diff --git a/daemons/controld/controld_remote_ra.c b/daemons/controld/controld_remote_ra.c index 74baebc80c..32cc5f7c1a 100644 --- a/daemons/controld/controld_remote_ra.c +++ b/daemons/controld/controld_remote_ra.c @@ -1,1476 +1,1476 @@ /* * Copyright 2013-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #define REMOTE_LRMD_RA "remote" /* The max start timeout before cmd retry */ #define MAX_START_TIMEOUT_MS 10000 #define cmd_set_flags(cmd, flags_to_set) do { \ (cmd)->status = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, \ "Remote command", (cmd)->rsc_id, (cmd)->status, \ (flags_to_set), #flags_to_set); \ } while (0) #define cmd_clear_flags(cmd, flags_to_clear) do { \ (cmd)->status = pcmk__clear_flags_as(__func__, __LINE__, LOG_TRACE, \ "Remote command", (cmd)->rsc_id, (cmd)->status, \ (flags_to_clear), #flags_to_clear); \ } while (0) enum remote_cmd_status { cmd_reported_success = (1 << 0), cmd_cancel = (1 << 1), }; typedef struct remote_ra_cmd_s { /*! the local node the cmd is issued from */ char *owner; /*! the remote node the cmd is executed on */ char *rsc_id; /*! the action to execute */ char *action; /*! some string the client wants us to give it back */ char *userdata; /*! start delay in ms */ int start_delay; /*! timer id used for start delay. */ int delay_id; /*! timeout in ms for cmd */ int timeout; int remaining_timeout; /*! recurring interval in ms */ guint interval_ms; /*! interval timer id */ int interval_id; int monitor_timeout_id; int takeover_timeout_id; /*! action parameters */ lrmd_key_value_t *params; pcmk__action_result_t result; int call_id; time_t start_time; uint32_t status; } remote_ra_cmd_t; #define lrm_remote_set_flags(lrm_state, flags_to_set) do { \ lrm_state_t *lrm = (lrm_state); \ remote_ra_data_t *ra = lrm->remote_ra_data; \ ra->status = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, "Remote", \ lrm->node_name, ra->status, \ (flags_to_set), #flags_to_set); \ } while (0) #define lrm_remote_clear_flags(lrm_state, flags_to_clear) do { \ lrm_state_t *lrm = (lrm_state); \ remote_ra_data_t *ra = lrm->remote_ra_data; \ ra->status = pcmk__clear_flags_as(__func__, __LINE__, LOG_TRACE, "Remote", \ lrm->node_name, ra->status, \ (flags_to_clear), #flags_to_clear); \ } while (0) enum remote_status { expect_takeover = (1 << 0), takeover_complete = (1 << 1), remote_active = (1 << 2), /* Maintenance mode is difficult to determine from the controller's context, * so we have it signalled back with the transition from the scheduler. */ remote_in_maint = (1 << 3), /* Similar for whether we are controlling a guest node or remote node. * Fortunately there is a meta-attribute in the transition already and * as the situation doesn't change over time we can use the * resource start for noting down the information for later use when * the attributes aren't at hand. */ controlling_guest = (1 << 4), }; typedef struct remote_ra_data_s { crm_trigger_t *work; remote_ra_cmd_t *cur_cmd; GList *cmds; GList *recurring_cmds; uint32_t status; } remote_ra_data_t; static int handle_remote_ra_start(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd, int timeout_ms); static void handle_remote_ra_stop(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd); static GList *fail_all_monitor_cmds(GList * list); static void free_cmd(gpointer user_data) { remote_ra_cmd_t *cmd = user_data; if (!cmd) { return; } if (cmd->delay_id) { g_source_remove(cmd->delay_id); } if (cmd->interval_id) { g_source_remove(cmd->interval_id); } if (cmd->monitor_timeout_id) { g_source_remove(cmd->monitor_timeout_id); } if (cmd->takeover_timeout_id) { g_source_remove(cmd->takeover_timeout_id); } free(cmd->owner); free(cmd->rsc_id); free(cmd->action); free(cmd->userdata); pcmk__reset_result(&(cmd->result)); lrmd_key_value_freeall(cmd->params); free(cmd); } static int generate_callid(void) { static int remote_ra_callid = 0; remote_ra_callid++; if (remote_ra_callid <= 0) { remote_ra_callid = 1; } return remote_ra_callid; } static gboolean recurring_helper(gpointer data) { remote_ra_cmd_t *cmd = data; lrm_state_t *connection_rsc = NULL; cmd->interval_id = 0; connection_rsc = lrm_state_find(cmd->rsc_id); if (connection_rsc && connection_rsc->remote_ra_data) { remote_ra_data_t *ra_data = connection_rsc->remote_ra_data; ra_data->recurring_cmds = g_list_remove(ra_data->recurring_cmds, cmd); ra_data->cmds = g_list_append(ra_data->cmds, cmd); mainloop_set_trigger(ra_data->work); } return FALSE; } static gboolean start_delay_helper(gpointer data) { remote_ra_cmd_t *cmd = data; lrm_state_t *connection_rsc = NULL; cmd->delay_id = 0; connection_rsc = lrm_state_find(cmd->rsc_id); if (connection_rsc && connection_rsc->remote_ra_data) { remote_ra_data_t *ra_data = connection_rsc->remote_ra_data; mainloop_set_trigger(ra_data->work); } return FALSE; } static bool should_purge_attributes(crm_node_t *node) { bool purge = true; crm_node_t *conn_node = NULL; lrm_state_t *connection_rsc = NULL; if (!node->conn_host) { return purge; } /* Get the node that was hosting the remote connection resource from the * peer cache. That's the one we really care about here. */ conn_node = crm_get_peer(0, node->conn_host); if (conn_node == NULL) { return purge; } /* Check the uptime of connection_rsc. If it hasn't been running long * enough, set purge=true. "Long enough" means it started running earlier * than the timestamp when we noticed it went away in the first place. */ connection_rsc = lrm_state_find(node->uname); if (connection_rsc != NULL) { lrmd_t *lrm = connection_rsc->conn; time_t uptime = lrmd__uptime(lrm); time_t now = time(NULL); /* Add 20s of fuzziness to give corosync a while to notice the remote * host is gone. On various error conditions (failure to get uptime, * peer_lost isn't set) we default to purging. */ if (uptime > 0 && conn_node->peer_lost > 0 && uptime + 20 >= now - conn_node->peer_lost) { purge = false; } } return purge; } static enum controld_section_e section_to_delete(bool purge) { if (pcmk_is_set(controld_globals.flags, controld_shutdown_lock_enabled)) { if (purge) { return controld_section_all_unlocked; } else { return controld_section_lrm_unlocked; } } else { if (purge) { return controld_section_all; } else { return controld_section_lrm; } } } static void purge_remote_node_attrs(int call_opt, crm_node_t *node) { bool purge = should_purge_attributes(node); enum controld_section_e section = section_to_delete(purge); /* Purge node from attrd's memory */ if (purge) { update_attrd_remote_node_removed(node->uname, NULL); } controld_delete_node_state(node->uname, section, call_opt); } /*! * \internal * \brief Handle cluster communication related to pacemaker_remote node joining * * \param[in] node_name Name of newly integrated pacemaker_remote node */ static void remote_node_up(const char *node_name) { int call_opt; xmlNode *update, *state; crm_node_t *node; lrm_state_t *connection_rsc = NULL; CRM_CHECK(node_name != NULL, return); crm_info("Announcing Pacemaker Remote node %s", node_name); call_opt = crmd_cib_smart_opt(); /* Delete node's probe_complete attribute. This serves two purposes: * * - @COMPAT DCs < 1.1.14 in a rolling upgrade might use it * - deleting it (or any attribute for that matter) here ensures the * attribute manager learns the node is remote */ update_attrd(node_name, CRM_OP_PROBED, NULL, NULL, TRUE); /* Ensure node is in the remote peer cache with member status */ node = crm_remote_peer_get(node_name); CRM_CHECK(node != NULL, return); purge_remote_node_attrs(call_opt, node); pcmk__update_peer_state(__func__, node, CRM_NODE_MEMBER, 0); /* Apply any start state that we were given from the environment on the * remote node. */ connection_rsc = lrm_state_find(node->uname); if (connection_rsc != NULL) { lrmd_t *lrm = connection_rsc->conn; const char *start_state = lrmd__node_start_state(lrm); if (start_state) { set_join_state(start_state, node->uname, node->uuid, true); } } /* pacemaker_remote nodes don't participate in the membership layer, * so cluster nodes don't automatically get notified when they come and go. * We send a cluster message to the DC, and update the CIB node state entry, * so the DC will get it sooner (via message) or later (via CIB refresh), * and any other interested parties can query the CIB. */ broadcast_remote_state_message(node_name, true); update = create_xml_node(NULL, XML_CIB_TAG_STATUS); state = create_node_state_update(node, node_update_cluster, update, __func__); /* Clear the XML_NODE_IS_FENCED flag in the node state. If the node ever * needs to be fenced, this flag will allow various actions to determine * whether the fencing has happened yet. */ crm_xml_add(state, XML_NODE_IS_FENCED, "0"); /* TODO: If the remote connection drops, and this (async) CIB update either * failed or has not yet completed, later actions could mistakenly think the * node has already been fenced (if the XML_NODE_IS_FENCED attribute was * previously set, because it won't have been cleared). This could prevent * actual fencing or allow recurring monitor failures to be cleared too * soon. Ideally, we wouldn't rely on the CIB for the fenced status. */ controld_update_cib(XML_CIB_TAG_STATUS, update, call_opt, NULL); free_xml(update); } enum down_opts { DOWN_KEEP_LRM, DOWN_ERASE_LRM }; /*! * \internal * \brief Handle cluster communication related to pacemaker_remote node leaving * * \param[in] node_name Name of lost node * \param[in] opts Whether to keep or erase LRM history */ static void remote_node_down(const char *node_name, const enum down_opts opts) { xmlNode *update; int call_opt = crmd_cib_smart_opt(); crm_node_t *node; /* Purge node from attrd's memory */ update_attrd_remote_node_removed(node_name, NULL); /* Normally, only node attributes should be erased, and the resource history * should be kept until the node comes back up. However, after a successful * fence, we want to clear the history as well, so we don't think resources * are still running on the node. */ if (opts == DOWN_ERASE_LRM) { controld_delete_node_state(node_name, controld_section_all, call_opt); } else { controld_delete_node_state(node_name, controld_section_attrs, call_opt); } /* Ensure node is in the remote peer cache with lost state */ node = crm_remote_peer_get(node_name); CRM_CHECK(node != NULL, return); pcmk__update_peer_state(__func__, node, CRM_NODE_LOST, 0); /* Notify DC */ broadcast_remote_state_message(node_name, false); /* Update CIB node state */ update = create_xml_node(NULL, XML_CIB_TAG_STATUS); create_node_state_update(node, node_update_cluster, update, __func__); controld_update_cib(XML_CIB_TAG_STATUS, update, call_opt, NULL); free_xml(update); } /*! * \internal * \brief Handle effects of a remote RA command on node state * * \param[in] cmd Completed remote RA command */ static void check_remote_node_state(const remote_ra_cmd_t *cmd) { /* Only successful actions can change node state */ if (!pcmk__result_ok(&(cmd->result))) { return; } if (pcmk__str_eq(cmd->action, PCMK_ACTION_START, pcmk__str_casei)) { remote_node_up(cmd->rsc_id); } else if (pcmk__str_eq(cmd->action, PCMK_ACTION_MIGRATE_FROM, pcmk__str_casei)) { /* After a successful migration, we don't need to do remote_node_up() * because the DC already knows the node is up, and we don't want to * clear LRM history etc. We do need to add the remote node to this * host's remote peer cache, because (unless it happens to be DC) * it hasn't been tracking the remote node, and other code relies on * the cache to distinguish remote nodes from unseen cluster nodes. */ crm_node_t *node = crm_remote_peer_get(cmd->rsc_id); CRM_CHECK(node != NULL, return); pcmk__update_peer_state(__func__, node, CRM_NODE_MEMBER, 0); } else if (pcmk__str_eq(cmd->action, PCMK_ACTION_STOP, pcmk__str_casei)) { lrm_state_t *lrm_state = lrm_state_find(cmd->rsc_id); remote_ra_data_t *ra_data = lrm_state? lrm_state->remote_ra_data : NULL; if (ra_data) { if (!pcmk_is_set(ra_data->status, takeover_complete)) { /* Stop means down if we didn't successfully migrate elsewhere */ remote_node_down(cmd->rsc_id, DOWN_KEEP_LRM); } else if (AM_I_DC == FALSE) { /* Only the connection host and DC track node state, * so if the connection migrated elsewhere and we aren't DC, * un-cache the node, so we don't have stale info */ crm_remote_peer_cache_remove(cmd->rsc_id); } } } /* We don't do anything for successful monitors, which is correct for * routine recurring monitors, and for monitors on nodes where the * connection isn't supposed to be (the cluster will stop the connection in * that case). However, if the initial probe finds the connection already * active on the node where we want it, we probably should do * remote_node_up(). Unfortunately, we can't distinguish that case here. * Given that connections have to be initiated by the cluster, the chance of * that should be close to zero. */ } static void report_remote_ra_result(remote_ra_cmd_t * cmd) { lrmd_event_data_t op = { 0, }; check_remote_node_state(cmd); op.type = lrmd_event_exec_complete; op.rsc_id = cmd->rsc_id; op.op_type = cmd->action; op.user_data = cmd->userdata; op.timeout = cmd->timeout; op.interval_ms = cmd->interval_ms; op.t_run = (unsigned int) cmd->start_time; op.t_rcchange = (unsigned int) cmd->start_time; lrmd__set_result(&op, cmd->result.exit_status, cmd->result.execution_status, cmd->result.exit_reason); if (pcmk_is_set(cmd->status, cmd_reported_success) && !pcmk__result_ok(&(cmd->result))) { op.t_rcchange = (unsigned int) time(NULL); /* This edge case will likely never ever occur, but if it does the * result is that a failure will not be processed correctly. This is only * remotely possible because we are able to detect a connection resource's tcp * connection has failed at any moment after start has completed. The actual * recurring operation is just a connectivity ping. * * basically, we are not guaranteed that the first successful monitor op and * a subsequent failed monitor op will not occur in the same timestamp. We have to * make it look like the operations occurred at separate times though. */ if (op.t_rcchange == op.t_run) { op.t_rcchange++; } } if (cmd->params) { lrmd_key_value_t *tmp; op.params = pcmk__strkey_table(free, free); for (tmp = cmd->params; tmp; tmp = tmp->next) { g_hash_table_insert(op.params, strdup(tmp->key), strdup(tmp->value)); } } op.call_id = cmd->call_id; op.remote_nodename = cmd->owner; lrm_op_callback(&op); if (op.params) { g_hash_table_destroy(op.params); } lrmd__reset_result(&op); } static void update_remaining_timeout(remote_ra_cmd_t * cmd) { cmd->remaining_timeout = ((cmd->timeout / 1000) - (time(NULL) - cmd->start_time)) * 1000; } static gboolean retry_start_cmd_cb(gpointer data) { lrm_state_t *lrm_state = data; remote_ra_data_t *ra_data = lrm_state->remote_ra_data; remote_ra_cmd_t *cmd = NULL; int rc = ETIME; if (!ra_data || !ra_data->cur_cmd) { return FALSE; } cmd = ra_data->cur_cmd; if (!pcmk__strcase_any_of(cmd->action, PCMK_ACTION_START, PCMK_ACTION_MIGRATE_FROM, NULL)) { return FALSE; } update_remaining_timeout(cmd); if (cmd->remaining_timeout > 0) { rc = handle_remote_ra_start(lrm_state, cmd, cmd->remaining_timeout); } else { pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_TIMEOUT, "Not enough time remains to retry remote connection"); } if (rc != pcmk_rc_ok) { report_remote_ra_result(cmd); if (ra_data->cmds) { mainloop_set_trigger(ra_data->work); } ra_data->cur_cmd = NULL; free_cmd(cmd); } else { /* wait for connection event */ } return FALSE; } static gboolean connection_takeover_timeout_cb(gpointer data) { lrm_state_t *lrm_state = NULL; remote_ra_cmd_t *cmd = data; crm_info("takeover event timed out for node %s", cmd->rsc_id); cmd->takeover_timeout_id = 0; lrm_state = lrm_state_find(cmd->rsc_id); handle_remote_ra_stop(lrm_state, cmd); free_cmd(cmd); return FALSE; } static gboolean monitor_timeout_cb(gpointer data) { lrm_state_t *lrm_state = NULL; remote_ra_cmd_t *cmd = data; lrm_state = lrm_state_find(cmd->rsc_id); crm_info("Timed out waiting for remote poke response from %s%s", cmd->rsc_id, (lrm_state? "" : " (no LRM state)")); cmd->monitor_timeout_id = 0; pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_TIMEOUT, "Remote executor did not respond"); if (lrm_state && lrm_state->remote_ra_data) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; if (ra_data->cur_cmd == cmd) { ra_data->cur_cmd = NULL; } if (ra_data->cmds) { mainloop_set_trigger(ra_data->work); } } report_remote_ra_result(cmd); free_cmd(cmd); if(lrm_state) { lrm_state_disconnect(lrm_state); } return FALSE; } static void synthesize_lrmd_success(lrm_state_t *lrm_state, const char *rsc_id, const char *op_type) { lrmd_event_data_t op = { 0, }; if (lrm_state == NULL) { /* if lrm_state not given assume local */ lrm_state = lrm_state_find(controld_globals.our_nodename); } CRM_ASSERT(lrm_state != NULL); op.type = lrmd_event_exec_complete; op.rsc_id = rsc_id; op.op_type = op_type; op.t_run = (unsigned int) time(NULL); op.t_rcchange = op.t_run; op.call_id = generate_callid(); lrmd__set_result(&op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); process_lrm_event(lrm_state, &op, NULL, NULL); } void remote_lrm_op_callback(lrmd_event_data_t * op) { gboolean cmd_handled = FALSE; lrm_state_t *lrm_state = NULL; remote_ra_data_t *ra_data = NULL; remote_ra_cmd_t *cmd = NULL; crm_debug("Processing '%s%s%s' event on remote connection to %s: %s " "(%d) status=%s (%d)", (op->op_type? op->op_type : ""), (op->op_type? " " : ""), lrmd_event_type2str(op->type), op->remote_nodename, services_ocf_exitcode_str(op->rc), op->rc, pcmk_exec_status_str(op->op_status), op->op_status); lrm_state = lrm_state_find(op->remote_nodename); if (!lrm_state || !lrm_state->remote_ra_data) { crm_debug("No state information found for remote connection event"); return; } ra_data = lrm_state->remote_ra_data; if (op->type == lrmd_event_new_client) { // Another client has connected to the remote daemon if (pcmk_is_set(ra_data->status, expect_takeover)) { // Great, we knew this was coming lrm_remote_clear_flags(lrm_state, expect_takeover); lrm_remote_set_flags(lrm_state, takeover_complete); } else { crm_err("Disconnecting from Pacemaker Remote node %s due to " "unexpected client takeover", op->remote_nodename); /* In this case, lrmd_tls_connection_destroy() will be called under the control of mainloop. */ /* Do not free lrm_state->conn yet. */ /* It'll be freed in the following stop action. */ lrm_state_disconnect_only(lrm_state); } return; } /* filter all EXEC events up */ if (op->type == lrmd_event_exec_complete) { if (pcmk_is_set(ra_data->status, takeover_complete)) { crm_debug("ignoring event, this connection is taken over by another node"); } else { lrm_op_callback(op); } return; } if ((op->type == lrmd_event_disconnect) && (ra_data->cur_cmd == NULL)) { if (!pcmk_is_set(ra_data->status, remote_active)) { crm_debug("Disconnection from Pacemaker Remote node %s complete", lrm_state->node_name); } else if (!remote_ra_is_in_maintenance(lrm_state)) { crm_err("Lost connection to Pacemaker Remote node %s", lrm_state->node_name); ra_data->recurring_cmds = fail_all_monitor_cmds(ra_data->recurring_cmds); ra_data->cmds = fail_all_monitor_cmds(ra_data->cmds); } else { crm_notice("Unmanaged Pacemaker Remote node %s disconnected", lrm_state->node_name); /* Do roughly what a 'stop' on the remote-resource would do */ handle_remote_ra_stop(lrm_state, NULL); remote_node_down(lrm_state->node_name, DOWN_KEEP_LRM); /* now fake the reply of a successful 'stop' */ synthesize_lrmd_success(NULL, lrm_state->node_name, PCMK_ACTION_STOP); } return; } if (!ra_data->cur_cmd) { crm_debug("no event to match"); return; } cmd = ra_data->cur_cmd; /* Start actions and migrate from actions complete after connection * comes back to us. */ if ((op->type == lrmd_event_connect) && pcmk__strcase_any_of(cmd->action, PCMK_ACTION_START, PCMK_ACTION_MIGRATE_FROM, NULL)) { if (op->connection_rc < 0) { update_remaining_timeout(cmd); if ((op->connection_rc == -ENOKEY) || (op->connection_rc == -EKEYREJECTED)) { // Hard error, don't retry pcmk__set_result(&(cmd->result), PCMK_OCF_INVALID_PARAM, PCMK_EXEC_ERROR, pcmk_strerror(op->connection_rc)); } else if (cmd->remaining_timeout > 3000) { crm_trace("rescheduling start, remaining timeout %d", cmd->remaining_timeout); g_timeout_add(1000, retry_start_cmd_cb, lrm_state); return; } else { crm_trace("can't reschedule start, remaining timeout too small %d", cmd->remaining_timeout); pcmk__format_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_TIMEOUT, "%s without enough time to retry", pcmk_strerror(op->connection_rc)); } } else { lrm_state_reset_tables(lrm_state, TRUE); pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); lrm_remote_set_flags(lrm_state, remote_active); } crm_debug("Remote connection event matched %s action", cmd->action); report_remote_ra_result(cmd); cmd_handled = TRUE; } else if ((op->type == lrmd_event_poke) && pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei)) { if (cmd->monitor_timeout_id) { g_source_remove(cmd->monitor_timeout_id); cmd->monitor_timeout_id = 0; } /* Only report success the first time, after that only worry about failures. * For this function, if we get the poke pack, it is always a success. Pokes * only fail if the send fails, or the response times out. */ if (!pcmk_is_set(cmd->status, cmd_reported_success)) { pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); report_remote_ra_result(cmd); cmd_set_flags(cmd, cmd_reported_success); } crm_debug("Remote poke event matched %s action", cmd->action); /* success, keep rescheduling if interval is present. */ if (cmd->interval_ms && !pcmk_is_set(cmd->status, cmd_cancel)) { ra_data->recurring_cmds = g_list_append(ra_data->recurring_cmds, cmd); cmd->interval_id = g_timeout_add(cmd->interval_ms, recurring_helper, cmd); cmd = NULL; /* prevent free */ } cmd_handled = TRUE; } else if ((op->type == lrmd_event_disconnect) && pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei)) { if (pcmk_is_set(ra_data->status, remote_active) && !pcmk_is_set(cmd->status, cmd_cancel)) { pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "Remote connection unexpectedly dropped " "during monitor"); report_remote_ra_result(cmd); crm_err("Remote connection to %s unexpectedly dropped during monitor", lrm_state->node_name); } cmd_handled = TRUE; } else if ((op->type == lrmd_event_new_client) && pcmk__str_eq(cmd->action, PCMK_ACTION_STOP, pcmk__str_casei)) { handle_remote_ra_stop(lrm_state, cmd); cmd_handled = TRUE; } else { crm_debug("Event did not match %s action", ra_data->cur_cmd->action); } if (cmd_handled) { ra_data->cur_cmd = NULL; if (ra_data->cmds) { mainloop_set_trigger(ra_data->work); } free_cmd(cmd); } } static void handle_remote_ra_stop(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd) { remote_ra_data_t *ra_data = NULL; CRM_ASSERT(lrm_state); ra_data = lrm_state->remote_ra_data; if (!pcmk_is_set(ra_data->status, takeover_complete)) { /* delete pending ops when ever the remote connection is intentionally stopped */ g_hash_table_remove_all(lrm_state->active_ops); } else { /* we no longer hold the history if this connection has been migrated, * however, we keep metadata cache for future use */ lrm_state_reset_tables(lrm_state, FALSE); } lrm_remote_clear_flags(lrm_state, remote_active); lrm_state_disconnect(lrm_state); if (ra_data->cmds) { g_list_free_full(ra_data->cmds, free_cmd); } if (ra_data->recurring_cmds) { g_list_free_full(ra_data->recurring_cmds, free_cmd); } ra_data->cmds = NULL; ra_data->recurring_cmds = NULL; ra_data->cur_cmd = NULL; if (cmd) { pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); report_remote_ra_result(cmd); } } // \return Standard Pacemaker return code static int handle_remote_ra_start(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd, int timeout_ms) { const char *server = NULL; lrmd_key_value_t *tmp = NULL; int port = 0; int timeout_used = timeout_ms > MAX_START_TIMEOUT_MS ? MAX_START_TIMEOUT_MS : timeout_ms; int rc = pcmk_rc_ok; for (tmp = cmd->params; tmp; tmp = tmp->next) { - if (pcmk__strcase_any_of(tmp->key, XML_RSC_ATTR_REMOTE_RA_ADDR, + if (pcmk__strcase_any_of(tmp->key, PCMK_REMOTE_RA_ADDR, XML_RSC_ATTR_REMOTE_RA_SERVER, NULL)) { server = tmp->value; } else if (pcmk__str_eq(tmp->key, XML_RSC_ATTR_REMOTE_RA_PORT, pcmk__str_casei)) { port = atoi(tmp->value); } else if (pcmk__str_eq(tmp->key, CRM_META "_" PCMK__META_CONTAINER, pcmk__str_none)) { lrm_remote_set_flags(lrm_state, controlling_guest); } } rc = controld_connect_remote_executor(lrm_state, server, port, timeout_used); if (rc != pcmk_rc_ok) { pcmk__format_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "Could not connect to Pacemaker Remote node %s: %s", lrm_state->node_name, pcmk_rc_str(rc)); } return rc; } static gboolean handle_remote_ra_exec(gpointer user_data) { int rc = 0; lrm_state_t *lrm_state = user_data; remote_ra_data_t *ra_data = lrm_state->remote_ra_data; remote_ra_cmd_t *cmd; GList *first = NULL; if (ra_data->cur_cmd) { /* still waiting on previous cmd */ return TRUE; } while (ra_data->cmds) { first = ra_data->cmds; cmd = first->data; if (cmd->delay_id) { /* still waiting for start delay timer to trip */ return TRUE; } ra_data->cmds = g_list_remove_link(ra_data->cmds, first); g_list_free_1(first); if (pcmk__str_any_of(cmd->action, PCMK_ACTION_START, PCMK_ACTION_MIGRATE_FROM, NULL)) { lrm_remote_clear_flags(lrm_state, expect_takeover | takeover_complete); if (handle_remote_ra_start(lrm_state, cmd, cmd->timeout) == pcmk_rc_ok) { /* take care of this later when we get async connection result */ crm_debug("Initiated async remote connection, %s action will complete after connect event", cmd->action); ra_data->cur_cmd = cmd; return TRUE; } report_remote_ra_result(cmd); } else if (!strcmp(cmd->action, PCMK_ACTION_MONITOR)) { if (lrm_state_is_connected(lrm_state) == TRUE) { rc = lrm_state_poke_connection(lrm_state); if (rc < 0) { pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, pcmk_strerror(rc)); } } else { rc = -1; pcmk__set_result(&(cmd->result), PCMK_OCF_NOT_RUNNING, PCMK_EXEC_DONE, "Remote connection inactive"); } if (rc == 0) { crm_debug("Poked Pacemaker Remote at node %s, waiting for async response", cmd->rsc_id); ra_data->cur_cmd = cmd; cmd->monitor_timeout_id = g_timeout_add(cmd->timeout, monitor_timeout_cb, cmd); return TRUE; } report_remote_ra_result(cmd); } else if (!strcmp(cmd->action, PCMK_ACTION_STOP)) { if (pcmk_is_set(ra_data->status, expect_takeover)) { /* briefly wait on stop for the takeover event to occur. If the * takeover event does not occur during the wait period, that's fine. * It just means that the remote-node's lrm_status section is going to get * cleared which will require all the resources running in the remote-node * to be explicitly re-detected via probe actions. If the takeover does occur * successfully, then we can leave the status section intact. */ cmd->takeover_timeout_id = g_timeout_add((cmd->timeout/2), connection_takeover_timeout_cb, cmd); ra_data->cur_cmd = cmd; return TRUE; } handle_remote_ra_stop(lrm_state, cmd); } else if (strcmp(cmd->action, PCMK_ACTION_MIGRATE_TO) == 0) { lrm_remote_clear_flags(lrm_state, takeover_complete); lrm_remote_set_flags(lrm_state, expect_takeover); pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); report_remote_ra_result(cmd); } else if (pcmk__str_any_of(cmd->action, PCMK_ACTION_RELOAD, PCMK_ACTION_RELOAD_AGENT, NULL)) { /* Currently the only reloadable parameter is reconnect_interval, * which is only used by the scheduler via the CIB, so reloads are a * no-op. * * @COMPAT DC <2.1.0: We only need to check for "reload" in case * we're in a rolling upgrade with a DC scheduling "reload" instead * of "reload-agent". An OCF 1.1 "reload" would be a no-op anyway, * so this would work for that purpose as well. */ pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); report_remote_ra_result(cmd); } free_cmd(cmd); } return TRUE; } static void remote_ra_data_init(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = NULL; if (lrm_state->remote_ra_data) { return; } ra_data = calloc(1, sizeof(remote_ra_data_t)); ra_data->work = mainloop_add_trigger(G_PRIORITY_HIGH, handle_remote_ra_exec, lrm_state); lrm_state->remote_ra_data = ra_data; } void remote_ra_cleanup(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; if (!ra_data) { return; } if (ra_data->cmds) { g_list_free_full(ra_data->cmds, free_cmd); } if (ra_data->recurring_cmds) { g_list_free_full(ra_data->recurring_cmds, free_cmd); } mainloop_destroy_trigger(ra_data->work); free(ra_data); lrm_state->remote_ra_data = NULL; } gboolean is_remote_lrmd_ra(const char *agent, const char *provider, const char *id) { if (agent && provider && !strcmp(agent, REMOTE_LRMD_RA) && !strcmp(provider, "pacemaker")) { return TRUE; } if ((id != NULL) && (lrm_state_find(id) != NULL) && !pcmk__str_eq(id, controld_globals.our_nodename, pcmk__str_casei)) { return TRUE; } return FALSE; } lrmd_rsc_info_t * remote_ra_get_rsc_info(lrm_state_t * lrm_state, const char *rsc_id) { lrmd_rsc_info_t *info = NULL; if ((lrm_state_find(rsc_id))) { info = calloc(1, sizeof(lrmd_rsc_info_t)); info->id = strdup(rsc_id); info->type = strdup(REMOTE_LRMD_RA); info->standard = strdup(PCMK_RESOURCE_CLASS_OCF); info->provider = strdup("pacemaker"); } return info; } static gboolean is_remote_ra_supported_action(const char *action) { return pcmk__str_any_of(action, PCMK_ACTION_START, PCMK_ACTION_STOP, PCMK_ACTION_MONITOR, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, PCMK_ACTION_RELOAD_AGENT, PCMK_ACTION_RELOAD, NULL); } static GList * fail_all_monitor_cmds(GList * list) { GList *rm_list = NULL; remote_ra_cmd_t *cmd = NULL; GList *gIter = NULL; for (gIter = list; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms > 0) && pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei)) { rm_list = g_list_append(rm_list, cmd); } } for (gIter = rm_list; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "Lost connection to remote executor"); crm_trace("Pre-emptively failing %s %s (interval=%u, %s)", cmd->action, cmd->rsc_id, cmd->interval_ms, cmd->userdata); report_remote_ra_result(cmd); list = g_list_remove(list, cmd); free_cmd(cmd); } /* frees only the list data, not the cmds */ g_list_free(rm_list); return list; } static GList * remove_cmd(GList * list, const char *action, guint interval_ms) { remote_ra_cmd_t *cmd = NULL; GList *gIter = NULL; for (gIter = list; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms == interval_ms) && pcmk__str_eq(cmd->action, action, pcmk__str_casei)) { break; } cmd = NULL; } if (cmd) { list = g_list_remove(list, cmd); free_cmd(cmd); } return list; } int remote_ra_cancel(lrm_state_t *lrm_state, const char *rsc_id, const char *action, guint interval_ms) { lrm_state_t *connection_rsc = NULL; remote_ra_data_t *ra_data = NULL; connection_rsc = lrm_state_find(rsc_id); if (!connection_rsc || !connection_rsc->remote_ra_data) { return -EINVAL; } ra_data = connection_rsc->remote_ra_data; ra_data->cmds = remove_cmd(ra_data->cmds, action, interval_ms); ra_data->recurring_cmds = remove_cmd(ra_data->recurring_cmds, action, interval_ms); if (ra_data->cur_cmd && (ra_data->cur_cmd->interval_ms == interval_ms) && (pcmk__str_eq(ra_data->cur_cmd->action, action, pcmk__str_casei))) { cmd_set_flags(ra_data->cur_cmd, cmd_cancel); } return 0; } static remote_ra_cmd_t * handle_dup_monitor(remote_ra_data_t *ra_data, guint interval_ms, const char *userdata) { GList *gIter = NULL; remote_ra_cmd_t *cmd = NULL; /* there are 3 places a potential duplicate monitor operation * could exist. * 1. recurring_cmds list. where the op is waiting for its next interval * 2. cmds list, where the op is queued to get executed immediately * 3. cur_cmd, which means the monitor op is in flight right now. */ if (interval_ms == 0) { return NULL; } if (ra_data->cur_cmd && !pcmk_is_set(ra_data->cur_cmd->status, cmd_cancel) && (ra_data->cur_cmd->interval_ms == interval_ms) && pcmk__str_eq(ra_data->cur_cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei)) { cmd = ra_data->cur_cmd; goto handle_dup; } for (gIter = ra_data->recurring_cmds; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms == interval_ms) && pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei)) { goto handle_dup; } } for (gIter = ra_data->cmds; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms == interval_ms) && pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei)) { goto handle_dup; } } return NULL; handle_dup: crm_trace("merging duplicate monitor cmd " PCMK__OP_FMT, cmd->rsc_id, PCMK_ACTION_MONITOR, interval_ms); /* update the userdata */ if (userdata) { free(cmd->userdata); cmd->userdata = strdup(userdata); } /* if we've already reported success, generate a new call id */ if (pcmk_is_set(cmd->status, cmd_reported_success)) { cmd->start_time = time(NULL); cmd->call_id = generate_callid(); cmd_clear_flags(cmd, cmd_reported_success); } /* if we have an interval_id set, that means we are in the process of * waiting for this cmd's next interval. instead of waiting, cancel * the timer and execute the action immediately */ if (cmd->interval_id) { g_source_remove(cmd->interval_id); cmd->interval_id = 0; recurring_helper(cmd); } return cmd; } /*! * \internal * \brief Execute an action using the (internal) ocf:pacemaker:remote agent * * \param[in] lrm_state Executor state object for remote connection * \param[in] rsc_id Connection resource ID * \param[in] action Action to execute * \param[in] userdata String to copy and pass to execution callback * \param[in] interval_ms Action interval (in milliseconds) * \param[in] timeout_ms Action timeout (in milliseconds) * \param[in] start_delay_ms Delay (in milliseconds) before executing action * \param[in,out] params Connection resource parameters * \param[out] call_id Where to store call ID on success * * \return Standard Pacemaker return code * \note This takes ownership of \p params, which should not be used or freed * after calling this function. */ int controld_execute_remote_agent(const lrm_state_t *lrm_state, const char *rsc_id, const char *action, const char *userdata, guint interval_ms, int timeout_ms, int start_delay_ms, lrmd_key_value_t *params, int *call_id) { lrm_state_t *connection_rsc = NULL; remote_ra_cmd_t *cmd = NULL; remote_ra_data_t *ra_data = NULL; *call_id = 0; CRM_CHECK((lrm_state != NULL) && (rsc_id != NULL) && (action != NULL) && (userdata != NULL) && (call_id != NULL), lrmd_key_value_freeall(params); return EINVAL); if (!is_remote_ra_supported_action(action)) { lrmd_key_value_freeall(params); return EOPNOTSUPP; } connection_rsc = lrm_state_find(rsc_id); if (connection_rsc == NULL) { lrmd_key_value_freeall(params); return ENOTCONN; } remote_ra_data_init(connection_rsc); ra_data = connection_rsc->remote_ra_data; cmd = handle_dup_monitor(ra_data, interval_ms, userdata); if (cmd) { *call_id = cmd->call_id; lrmd_key_value_freeall(params); return pcmk_rc_ok; } cmd = calloc(1, sizeof(remote_ra_cmd_t)); if (cmd == NULL) { lrmd_key_value_freeall(params); return ENOMEM; } cmd->owner = strdup(lrm_state->node_name); cmd->rsc_id = strdup(rsc_id); cmd->action = strdup(action); cmd->userdata = strdup(userdata); if ((cmd->owner == NULL) || (cmd->rsc_id == NULL) || (cmd->action == NULL) || (cmd->userdata == NULL)) { free_cmd(cmd); lrmd_key_value_freeall(params); return ENOMEM; } cmd->interval_ms = interval_ms; cmd->timeout = timeout_ms; cmd->start_delay = start_delay_ms; cmd->params = params; cmd->start_time = time(NULL); cmd->call_id = generate_callid(); if (cmd->start_delay) { cmd->delay_id = g_timeout_add(cmd->start_delay, start_delay_helper, cmd); } ra_data->cmds = g_list_append(ra_data->cmds, cmd); mainloop_set_trigger(ra_data->work); *call_id = cmd->call_id; return pcmk_rc_ok; } /*! * \internal * \brief Immediately fail all monitors of a remote node, if proxied here * * \param[in] node_name Name of pacemaker_remote node */ void remote_ra_fail(const char *node_name) { lrm_state_t *lrm_state = lrm_state_find(node_name); if (lrm_state && lrm_state_is_connected(lrm_state)) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; crm_info("Failing monitors on Pacemaker Remote node %s", node_name); ra_data->recurring_cmds = fail_all_monitor_cmds(ra_data->recurring_cmds); ra_data->cmds = fail_all_monitor_cmds(ra_data->cmds); } } /* A guest node fencing implied by host fencing looks like: * * * * * * * */ #define XPATH_PSEUDO_FENCE "/" XML_GRAPH_TAG_PSEUDO_EVENT \ "[@" XML_LRM_ATTR_TASK "='stonith']/" XML_GRAPH_TAG_DOWNED \ "/" XML_CIB_TAG_NODE /*! * \internal * \brief Check a pseudo-action for Pacemaker Remote node side effects * * \param[in,out] xml XML of pseudo-action to check */ void remote_ra_process_pseudo(xmlNode *xml) { xmlXPathObjectPtr search = xpath_search(xml, XPATH_PSEUDO_FENCE); if (numXpathResults(search) == 1) { xmlNode *result = getXpathResult(search, 0); /* Normally, we handle the necessary side effects of a guest node stop * action when reporting the remote agent's result. However, if the stop * is implied due to fencing, it will be a fencing pseudo-event, and * there won't be a result to report. Handle that case here. * * This will result in a duplicate call to remote_node_down() if the * guest stop was real instead of implied, but that shouldn't hurt. * * There is still one corner case that isn't handled: if a guest node * isn't running any resources when its host is fenced, it will appear * to be cleanly stopped, so there will be no pseudo-fence, and our * peer cache state will be incorrect unless and until the guest is * recovered. */ if (result) { const char *remote = ID(result); if (remote) { remote_node_down(remote, DOWN_ERASE_LRM); } } } freeXpathObject(search); } static void remote_ra_maintenance(lrm_state_t * lrm_state, gboolean maintenance) { xmlNode *update, *state; int call_opt; crm_node_t *node; call_opt = crmd_cib_smart_opt(); node = crm_remote_peer_get(lrm_state->node_name); CRM_CHECK(node != NULL, return); update = create_xml_node(NULL, XML_CIB_TAG_STATUS); state = create_node_state_update(node, node_update_none, update, __func__); crm_xml_add(state, XML_NODE_IS_MAINTENANCE, maintenance?"1":"0"); if (controld_update_cib(XML_CIB_TAG_STATUS, update, call_opt, NULL) == pcmk_rc_ok) { /* TODO: still not 100% sure that async update will succeed ... */ if (maintenance) { lrm_remote_set_flags(lrm_state, remote_in_maint); } else { lrm_remote_clear_flags(lrm_state, remote_in_maint); } } free_xml(update); } #define XPATH_PSEUDO_MAINTENANCE "//" XML_GRAPH_TAG_PSEUDO_EVENT \ "[@" XML_LRM_ATTR_TASK "='" PCMK_ACTION_MAINTENANCE_NODES "']/" \ XML_GRAPH_TAG_MAINTENANCE /*! * \internal * \brief Check a pseudo-action holding updates for maintenance state * * \param[in,out] xml XML of pseudo-action to check */ void remote_ra_process_maintenance_nodes(xmlNode *xml) { xmlXPathObjectPtr search = xpath_search(xml, XPATH_PSEUDO_MAINTENANCE); if (numXpathResults(search) == 1) { xmlNode *node; int cnt = 0, cnt_remote = 0; for (node = first_named_child(getXpathResult(search, 0), XML_CIB_TAG_NODE); node != NULL; node = crm_next_same_xml(node)) { lrm_state_t *lrm_state = lrm_state_find(ID(node)); cnt++; if (lrm_state && lrm_state->remote_ra_data && pcmk_is_set(((remote_ra_data_t *) lrm_state->remote_ra_data)->status, remote_active)) { int is_maint; cnt_remote++; pcmk__scan_min_int(crm_element_value(node, XML_NODE_IS_MAINTENANCE), &is_maint, 0); remote_ra_maintenance(lrm_state, is_maint); } } crm_trace("Action holds %d nodes (%d remotes found) adjusting " PCMK_OPT_MAINTENANCE_MODE, cnt, cnt_remote); } freeXpathObject(search); } gboolean remote_ra_is_in_maintenance(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; return pcmk_is_set(ra_data->status, remote_in_maint); } gboolean remote_ra_controlling_guest(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; return pcmk_is_set(ra_data->status, controlling_guest); } diff --git a/include/crm/msg_xml.h b/include/crm/msg_xml.h index b47980513d..3c80d87cf9 100644 --- a/include/crm/msg_xml.h +++ b/include/crm/msg_xml.h @@ -1,457 +1,464 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CRM_MSG_XML__H # define PCMK__CRM_MSG_XML__H # include #if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1) #include #endif #ifdef __cplusplus extern "C" { #endif /* This file defines constants for various XML syntax (mainly element and * attribute names). * * For consistency, new constants should start with "PCMK_", followed by "XE" * for XML element names, "XA" for XML attribute names, and "META" for meta * attribute names. Old names that don't follow this policy should eventually be * deprecated and replaced with names that do. */ /* * XML elements */ #define PCMK_XE_DATE_EXPRESSION "date_expression" #define PCMK_XE_OP_EXPRESSION "op_expression" /* This has been deprecated as a CIB element (an alias for with * "promotable" set to "true") since 2.0.0. */ #define PCMK_XE_PROMOTABLE_LEGACY "master" #define PCMK_XE_RSC_EXPRESSION "rsc_expression" /* * XML attributes */ #define PCMK_XA_ADMIN_EPOCH "admin_epoch" #define PCMK_XA_CIB_LAST_WRITTEN "cib-last-written" #define PCMK_XA_CRM_DEBUG_ORIGIN "crm-debug-origin" #define PCMK_XA_CRM_FEATURE_SET "crm_feature_set" #define PCMK_XA_CRM_TIMESTAMP "crm-timestamp" #define PCMK_XA_DESCRIPTION "description" #define PCMK_XA_EPOCH "epoch" #define PCMK_XA_FORMAT "format" #define PCMK_XA_HAVE_QUORUM "have-quorum" #define PCMK_XA_ID "id" #define PCMK_XA_NO_QUORUM_PANIC "no-quorum-panic" #define PCMK_XA_NUM_UPDATES "num_updates" #define PCMK_XA_VALIDATE_WITH "validate-with" #define PCMK_XA_VERSION "version" /* * Meta attributes */ #define PCMK_META_ALLOW_MIGRATE "allow-migrate" #define PCMK_META_CLONE_MAX "clone-max" #define PCMK_META_CLONE_MIN "clone-min" #define PCMK_META_CLONE_NODE_MAX "clone-node-max" #define PCMK_META_CONTAINER_ATTR_TARGET "container-attribute-target" #define PCMK_META_CRITICAL "critical" #define PCMK_META_ENABLED "enabled" #define PCMK_META_FAILURE_TIMEOUT "failure-timeout" #define PCMK_META_GLOBALLY_UNIQUE "globally-unique" #define PCMK_META_INTERLEAVE "interleave" #define PCMK_META_IS_MANAGED "is-managed" #define PCMK_META_MAINTENANCE "maintenance" #define PCMK_META_MIGRATION_THRESHOLD "migration-threshold" #define PCMK_META_MULTIPLE_ACTIVE "multiple-active" #define PCMK_META_NOTIFY "notify" #define PCMK_META_ORDERED "ordered" #define PCMK_META_PRIORITY "priority" #define PCMK_META_PROMOTABLE "promotable" #define PCMK_META_PROMOTED_MAX "promoted-max" #define PCMK_META_PROMOTED_NODE_MAX "promoted-node-max" #define PCMK_META_REMOTE_ADDR "remote-addr" #define PCMK_META_REMOTE_ALLOW_MIGRATE "remote-allow-migrate" #define PCMK_META_REMOTE_CONNECT_TIMEOUT "remote-connect-timeout" #define PCMK_META_REMOTE_NODE "remote-node" #define PCMK_META_REMOTE_PORT "remote-port" #define PCMK_META_REQUIRES "requires" #define PCMK_META_RESOURCE_STICKINESS "resource-stickiness" #define PCMK_META_TARGET_ROLE "target-role" +/* + * Remote resource instance attributes + */ + +#define PCMK_REMOTE_RA_ADDR "addr" + + /* * Older constants that don't follow current naming */ # ifndef F_ORIG # define F_ORIG "src" # endif # ifndef F_SEQ # define F_SEQ "seq" # endif # ifndef F_SUBTYPE # define F_SUBTYPE "subt" # endif # ifndef F_TYPE # define F_TYPE "t" # endif # ifndef F_CLIENTNAME # define F_CLIENTNAME "cn" # endif # ifndef F_XML_TAGNAME # define F_XML_TAGNAME "__name__" # endif # ifndef T_CRM # define T_CRM "crmd" # endif # ifndef T_ATTRD # define T_ATTRD "attrd" # endif # define CIB_OPTIONS_FIRST "cib-bootstrap-options" # define F_CRM_DATA "crm_xml" # define F_CRM_TASK "crm_task" # define F_CRM_HOST_TO "crm_host_to" # define F_CRM_MSG_TYPE F_SUBTYPE # define F_CRM_SYS_TO "crm_sys_to" # define F_CRM_SYS_FROM "crm_sys_from" # define F_CRM_HOST_FROM F_ORIG # define F_CRM_REFERENCE XML_ATTR_REFERENCE # define F_CRM_VERSION PCMK_XA_VERSION # define F_CRM_ORIGIN "origin" # define F_CRM_USER "crm_user" # define F_CRM_JOIN_ID "join_id" # define F_CRM_DC_LEAVING "dc-leaving" # define F_CRM_ELECTION_ID "election-id" # define F_CRM_ELECTION_AGE_S "election-age-sec" # define F_CRM_ELECTION_AGE_US "election-age-nano-sec" # define F_CRM_ELECTION_OWNER "election-owner" # define F_CRM_TGRAPH "crm-tgraph-file" # define F_CRM_TGRAPH_INPUT "crm-tgraph-in" # define F_CRM_THROTTLE_MODE "crm-limit-mode" # define F_CRM_THROTTLE_MAX "crm-limit-max" /*---- Common tags/attrs */ # define XML_DIFF_MARKER "__crm_diff_marker__" # define XML_TAG_CIB "cib" # define XML_TAG_FAILED "failed" # define XML_ATTR_TIMEOUT "timeout" # define XML_ATTR_NAME "name" # define XML_ATTR_IDREF "id-ref" # define XML_ATTR_ID_LONG "long-id" # define XML_ATTR_TYPE "type" # define XML_ATTR_OP "op" # define XML_ATTR_DC_UUID "dc-uuid" # define XML_ATTR_UPDATE_ORIG "update-origin" # define XML_ATTR_UPDATE_CLIENT "update-client" # define XML_ATTR_UPDATE_USER "update-user" # define XML_BOOLEAN_TRUE "true" # define XML_BOOLEAN_FALSE "false" # define XML_TAG_OPTIONS "options" /*---- top level tags/attrs */ # define XML_ATTR_REQUEST "request" # define XML_ATTR_RESPONSE "response" # define XML_ATTR_UNAME "uname" # define XML_ATTR_REFERENCE "reference" # define XML_CRM_TAG_PING "ping_response" # define XML_PING_ATTR_STATUS "result" # define XML_PING_ATTR_SYSFROM "crm_subsystem" # define XML_PING_ATTR_CRMDSTATE "crmd_state" # define XML_PING_ATTR_PACEMAKERDSTATE "pacemakerd_state" # define XML_PING_ATTR_PACEMAKERDSTATE_INIT "init" # define XML_PING_ATTR_PACEMAKERDSTATE_STARTINGDAEMONS "starting_daemons" # define XML_PING_ATTR_PACEMAKERDSTATE_WAITPING "wait_for_ping" # define XML_PING_ATTR_PACEMAKERDSTATE_RUNNING "running" # define XML_PING_ATTR_PACEMAKERDSTATE_SHUTTINGDOWN "shutting_down" # define XML_PING_ATTR_PACEMAKERDSTATE_SHUTDOWNCOMPLETE "shutdown_complete" # define XML_PING_ATTR_PACEMAKERDSTATE_REMOTE "remote" # define XML_FAIL_TAG_CIB "failed_update" # define XML_FAILCIB_ATTR_OBJTYPE "object_type" # define XML_FAILCIB_ATTR_OP "operation" # define XML_FAILCIB_ATTR_REASON "reason" /*---- CIB specific tags/attrs */ # define XML_CIB_TAG_SECTION_ALL "all" # define XML_CIB_TAG_CONFIGURATION "configuration" # define XML_CIB_TAG_STATUS "status" # define XML_CIB_TAG_RESOURCES "resources" # define XML_CIB_TAG_NODES "nodes" # define XML_CIB_TAG_CONSTRAINTS "constraints" # define XML_CIB_TAG_CRMCONFIG "crm_config" # define XML_CIB_TAG_OPCONFIG "op_defaults" # define XML_CIB_TAG_RSCCONFIG "rsc_defaults" # define XML_CIB_TAG_ACLS "acls" # define XML_CIB_TAG_ALERTS "alerts" # define XML_CIB_TAG_ALERT "alert" # define XML_CIB_TAG_ALERT_RECIPIENT "recipient" # define XML_CIB_TAG_ALERT_SELECT "select" # define XML_CIB_TAG_ALERT_ATTRIBUTES "select_attributes" # define XML_CIB_TAG_ALERT_FENCING "select_fencing" # define XML_CIB_TAG_ALERT_NODES "select_nodes" # define XML_CIB_TAG_ALERT_RESOURCES "select_resources" # define XML_CIB_TAG_ALERT_ATTR "attribute" # define XML_CIB_TAG_STATE "node_state" # define XML_CIB_TAG_NODE "node" # define XML_CIB_TAG_NVPAIR "nvpair" # define XML_CIB_TAG_PROPSET "cluster_property_set" # define XML_TAG_ATTR_SETS "instance_attributes" # define XML_TAG_META_SETS "meta_attributes" # define XML_TAG_ATTRS "attributes" # define XML_TAG_PARAMS "parameters" # define XML_TAG_PARAM "param" # define XML_TAG_UTILIZATION "utilization" # define XML_TAG_RESOURCE_REF "resource_ref" # define XML_CIB_TAG_RESOURCE "primitive" # define XML_CIB_TAG_GROUP "group" # define XML_CIB_TAG_INCARNATION "clone" # define XML_CIB_TAG_CONTAINER "bundle" # define XML_CIB_TAG_RSC_TEMPLATE "template" # define XML_RSC_ATTR_REMOTE_RA_ADDR "addr" # define XML_RSC_ATTR_REMOTE_RA_SERVER "server" # define XML_RSC_ATTR_REMOTE_RA_PORT "port" # define XML_REMOTE_ATTR_RECONNECT_INTERVAL "reconnect_interval" # define XML_OP_ATTR_ON_FAIL "on-fail" # define XML_OP_ATTR_START_DELAY "start-delay" # define XML_OP_ATTR_ORIGIN "interval-origin" # define XML_OP_ATTR_PENDING "record-pending" # define XML_OP_ATTR_DIGESTS_ALL "digests-all" # define XML_OP_ATTR_DIGESTS_SECURE "digests-secure" # define XML_CIB_TAG_LRM "lrm" # define XML_LRM_TAG_RESOURCES "lrm_resources" # define XML_LRM_TAG_RESOURCE "lrm_resource" # define XML_LRM_TAG_RSC_OP "lrm_rsc_op" # define XML_AGENT_ATTR_CLASS "class" # define XML_AGENT_ATTR_PROVIDER "provider" //! \deprecated Do not use (will be removed in a future release) # define XML_CIB_ATTR_REPLACE "replace" # define XML_CIB_ATTR_PRIORITY "priority" # define XML_NODE_IS_REMOTE "remote_node" # define XML_NODE_IS_FENCED "node_fenced" # define XML_NODE_IS_MAINTENANCE "node_in_maintenance" # define XML_CIB_ATTR_SHUTDOWN "shutdown" /* Aside from being an old name for the executor, LRM is a misnomer here because * the controller and scheduler use these to track actions, which are not always * executor operations. */ // XML attribute that takes interval specification (user-facing configuration) # define XML_LRM_ATTR_INTERVAL "interval" // XML attribute that takes interval in milliseconds (daemon APIs) // (identical value as above, but different constant allows clearer code intent) # define XML_LRM_ATTR_INTERVAL_MS XML_LRM_ATTR_INTERVAL # define XML_LRM_ATTR_TASK "operation" # define XML_LRM_ATTR_TASK_KEY "operation_key" # define XML_LRM_ATTR_TARGET "on_node" # define XML_LRM_ATTR_TARGET_UUID "on_node_uuid" /*! Actions to be executed on Pacemaker Remote nodes are routed through the * controller on the cluster node hosting the remote connection. That cluster * node is considered the router node for the action. */ # define XML_LRM_ATTR_ROUTER_NODE "router_node" # define XML_LRM_ATTR_RSCID "rsc-id" # define XML_LRM_ATTR_OPSTATUS "op-status" # define XML_LRM_ATTR_RC "rc-code" # define XML_LRM_ATTR_CALLID "call-id" # define XML_LRM_ATTR_OP_DIGEST "op-digest" # define XML_LRM_ATTR_OP_RESTART "op-force-restart" # define XML_LRM_ATTR_OP_SECURE "op-secure-params" # define XML_LRM_ATTR_RESTART_DIGEST "op-restart-digest" # define XML_LRM_ATTR_SECURE_DIGEST "op-secure-digest" # define XML_LRM_ATTR_EXIT_REASON "exit-reason" # define XML_RSC_OP_LAST_CHANGE "last-rc-change" # define XML_RSC_OP_T_EXEC "exec-time" # define XML_RSC_OP_T_QUEUE "queue-time" # define XML_LRM_ATTR_MIGRATE_SOURCE "migrate_source" # define XML_LRM_ATTR_MIGRATE_TARGET "migrate_target" # define XML_TAG_GRAPH "transition_graph" # define XML_GRAPH_TAG_RSC_OP "rsc_op" # define XML_GRAPH_TAG_PSEUDO_EVENT "pseudo_event" # define XML_GRAPH_TAG_CRM_EVENT "crm_event" # define XML_GRAPH_TAG_DOWNED "downed" # define XML_GRAPH_TAG_MAINTENANCE "maintenance" # define XML_TAG_RULE "rule" # define XML_RULE_ATTR_SCORE "score" # define XML_RULE_ATTR_SCORE_ATTRIBUTE "score-attribute" # define XML_RULE_ATTR_ROLE "role" # define XML_RULE_ATTR_BOOLEAN_OP "boolean-op" # define XML_TAG_EXPRESSION "expression" # define XML_EXPR_ATTR_ATTRIBUTE "attribute" # define XML_EXPR_ATTR_OPERATION "operation" # define XML_EXPR_ATTR_VALUE "value" # define XML_EXPR_ATTR_TYPE "type" # define XML_EXPR_ATTR_VALUE_SOURCE "value-source" # define XML_CONS_TAG_RSC_DEPEND "rsc_colocation" # define XML_CONS_TAG_RSC_ORDER "rsc_order" # define XML_CONS_TAG_RSC_LOCATION "rsc_location" # define XML_CONS_TAG_RSC_TICKET "rsc_ticket" # define XML_CONS_TAG_RSC_SET "resource_set" # define XML_CONS_ATTR_SYMMETRICAL "symmetrical" # define XML_LOCATION_ATTR_DISCOVERY "resource-discovery" # define XML_COLOC_ATTR_SOURCE "rsc" # define XML_COLOC_ATTR_SOURCE_ROLE "rsc-role" # define XML_COLOC_ATTR_TARGET "with-rsc" # define XML_COLOC_ATTR_TARGET_ROLE "with-rsc-role" # define XML_COLOC_ATTR_NODE_ATTR "node-attribute" # define XML_COLOC_ATTR_INFLUENCE "influence" //! \deprecated Deprecated since 2.1.5 # define XML_COLOC_ATTR_SOURCE_INSTANCE "rsc-instance" //! \deprecated Deprecated since 2.1.5 # define XML_COLOC_ATTR_TARGET_INSTANCE "with-rsc-instance" # define XML_LOC_ATTR_SOURCE "rsc" # define XML_LOC_ATTR_SOURCE_PATTERN "rsc-pattern" # define XML_ORDER_ATTR_FIRST "first" # define XML_ORDER_ATTR_THEN "then" # define XML_ORDER_ATTR_FIRST_ACTION "first-action" # define XML_ORDER_ATTR_THEN_ACTION "then-action" # define XML_ORDER_ATTR_KIND "kind" //! \deprecated Deprecated since 2.1.5 # define XML_ORDER_ATTR_FIRST_INSTANCE "first-instance" //! \deprecated Deprecated since 2.1.5 # define XML_ORDER_ATTR_THEN_INSTANCE "then-instance" # define XML_TICKET_ATTR_TICKET "ticket" # define XML_TICKET_ATTR_LOSS_POLICY "loss-policy" # define XML_NVPAIR_ATTR_NAME "name" # define XML_NVPAIR_ATTR_VALUE "value" # define XML_NODE_ATTR_RSC_DISCOVERY "resource-discovery-enabled" # define XML_ALERT_ATTR_PATH "path" # define XML_ALERT_ATTR_TIMEOUT "timeout" # define XML_ALERT_ATTR_TSTAMP_FORMAT "timestamp-format" # define XML_ALERT_ATTR_REC_VALUE "value" # define XML_CIB_TAG_GENERATION_TUPPLE "generation_tuple" # define XML_ATTR_TRANSITION_MAGIC "transition-magic" # define XML_ATTR_TRANSITION_KEY "transition-key" # define XML_ATTR_TE_NOWAIT "op_no_wait" # define XML_ATTR_TE_TARGET_RC "op_target_rc" # define XML_TAG_TRANSIENT_NODEATTRS "transient_attributes" //! \deprecated Do not use (will be removed in a future release) # define XML_TAG_DIFF_ADDED "diff-added" //! \deprecated Do not use (will be removed in a future release) # define XML_TAG_DIFF_REMOVED "diff-removed" # define XML_ACL_TAG_USER "acl_target" # define XML_ACL_TAG_USERv1 "acl_user" # define XML_ACL_TAG_GROUP "acl_group" # define XML_ACL_TAG_ROLE "acl_role" # define XML_ACL_TAG_PERMISSION "acl_permission" # define XML_ACL_TAG_ROLE_REF "role" # define XML_ACL_TAG_ROLE_REFv1 "role_ref" # define XML_ACL_ATTR_KIND "kind" # define XML_ACL_TAG_READ "read" # define XML_ACL_TAG_WRITE "write" # define XML_ACL_TAG_DENY "deny" # define XML_ACL_ATTR_REF "reference" # define XML_ACL_ATTR_REFv1 "ref" # define XML_ACL_ATTR_TAG "object-type" # define XML_ACL_ATTR_TAGv1 "tag" # define XML_ACL_ATTR_XPATH "xpath" # define XML_ACL_ATTR_ATTRIBUTE "attribute" # define XML_CIB_TAG_TICKETS "tickets" # define XML_CIB_TAG_TICKET_STATE "ticket_state" # define XML_CIB_TAG_TAGS "tags" # define XML_CIB_TAG_TAG "tag" # define XML_CIB_TAG_OBJ_REF "obj_ref" # define XML_TAG_FENCING_TOPOLOGY "fencing-topology" # define XML_TAG_FENCING_LEVEL "fencing-level" # define XML_ATTR_STONITH_INDEX "index" # define XML_ATTR_STONITH_TARGET "target" # define XML_ATTR_STONITH_TARGET_VALUE "target-value" # define XML_ATTR_STONITH_TARGET_PATTERN "target-pattern" # define XML_ATTR_STONITH_TARGET_ATTRIBUTE "target-attribute" # define XML_ATTR_STONITH_DEVICES "devices" # define XML_TAG_DIFF "diff" # define XML_DIFF_VERSION "version" # define XML_DIFF_VSOURCE "source" # define XML_DIFF_VTARGET "target" # define XML_DIFF_CHANGE "change" # define XML_DIFF_LIST "change-list" # define XML_DIFF_ATTR "change-attr" # define XML_DIFF_RESULT "change-result" # define XML_DIFF_OP "operation" # define XML_DIFF_PATH "path" # define XML_DIFF_POSITION "position" # define ID(x) crm_element_value(x, PCMK_XA_ID) #ifdef __cplusplus } #endif #endif diff --git a/lib/pacemaker/pcmk_sched_bundle.c b/lib/pacemaker/pcmk_sched_bundle.c index 0a9efe13f6..13f00653ac 100644 --- a/lib/pacemaker/pcmk_sched_bundle.c +++ b/lib/pacemaker/pcmk_sched_bundle.c @@ -1,1060 +1,1060 @@ /* - * Copyright 2004-2023 the Pacemaker project contributors + * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include "libpacemaker_private.h" struct assign_data { const pcmk_node_t *prefer; bool stop_if_fail; }; /*! * \internal * \brief Assign a single bundle replica's resources (other than container) * * \param[in,out] replica Replica to assign * \param[in] user_data Preferred node, if any * * \return true (to indicate that any further replicas should be processed) */ static bool assign_replica(pcmk__bundle_replica_t *replica, void *user_data) { pcmk_node_t *container_host = NULL; struct assign_data *assign_data = user_data; const pcmk_node_t *prefer = assign_data->prefer; bool stop_if_fail = assign_data->stop_if_fail; const pcmk_resource_t *bundle = pe__const_top_resource(replica->container, true); if (replica->ip != NULL) { pcmk__rsc_trace(bundle, "Assigning bundle %s IP %s", bundle->id, replica->ip->id); replica->ip->cmds->assign(replica->ip, prefer, stop_if_fail); } container_host = replica->container->allocated_to; if (replica->remote != NULL) { if (pe__is_guest_or_remote_node(container_host)) { /* REMOTE_CONTAINER_HACK: "Nested" connection resources must be on * the same host because Pacemaker Remote only supports a single * active connection. */ pcmk__new_colocation("#replica-remote-with-host-remote", NULL, INFINITY, replica->remote, container_host->details->remote_rsc, NULL, NULL, pcmk__coloc_influence); } pcmk__rsc_trace(bundle, "Assigning bundle %s connection %s", bundle->id, replica->remote->id); replica->remote->cmds->assign(replica->remote, prefer, stop_if_fail); } if (replica->child != NULL) { pcmk_node_t *node = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, replica->child->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { if (!pe__same_node(node, replica->node)) { node->weight = -INFINITY; } else if (!pcmk__threshold_reached(replica->child, node, NULL)) { node->weight = INFINITY; } } pe__set_resource_flags(replica->child->parent, pcmk_rsc_assigning); pcmk__rsc_trace(bundle, "Assigning bundle %s replica child %s", bundle->id, replica->child->id); replica->child->cmds->assign(replica->child, replica->node, stop_if_fail); pe__clear_resource_flags(replica->child->parent, pcmk_rsc_assigning); } return true; } /*! * \internal * \brief Assign a bundle resource to a node * * \param[in,out] rsc Resource to assign to a node * \param[in] prefer Node to prefer, if all else is equal * \param[in] stop_if_fail If \c true and a primitive descendant of \p rsc * can't be assigned to a node, set the * descendant's next role to stopped and update * existing actions * * \return Node that \p rsc is assigned to, if assigned entirely to one node * * \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can * completely undo the assignment. A successful assignment can be either * undone or left alone as final. A failed assignment has the same effect * as calling pcmk__unassign_resource(); there are no side effects on * roles or actions. */ pcmk_node_t * pcmk__bundle_assign(pcmk_resource_t *rsc, const pcmk_node_t *prefer, bool stop_if_fail) { GList *containers = NULL; pcmk_resource_t *bundled_resource = NULL; struct assign_data assign_data = { prefer, stop_if_fail }; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle)); pcmk__rsc_trace(rsc, "Assigning bundle %s", rsc->id); pe__set_resource_flags(rsc, pcmk_rsc_assigning); pe__show_node_scores(!pcmk_is_set(rsc->cluster->flags, pcmk_sched_output_scores), rsc, __func__, rsc->allowed_nodes, rsc->cluster); // Assign all containers first, so we know what nodes the bundle will be on containers = g_list_sort(pe__bundle_containers(rsc), pcmk__cmp_instance); pcmk__assign_instances(rsc, containers, pe__bundle_max(rsc), rsc->fns->max_per_node(rsc)); g_list_free(containers); // Then assign remaining replica resources pe__foreach_bundle_replica(rsc, assign_replica, (void *) &assign_data); // Finally, assign the bundled resources to each bundle node bundled_resource = pe__bundled_resource(rsc); if (bundled_resource != NULL) { pcmk_node_t *node = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, bundled_resource->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & node)) { if (pe__node_is_bundle_instance(rsc, node)) { node->weight = 0; } else { node->weight = -INFINITY; } } bundled_resource->cmds->assign(bundled_resource, prefer, stop_if_fail); } pe__clear_resource_flags(rsc, pcmk_rsc_assigning|pcmk_rsc_unassigned); return NULL; } /*! * \internal * \brief Create actions for a bundle replica's resources (other than child) * * \param[in,out] replica Replica to create actions for * \param[in] user_data Unused * * \return true (to indicate that any further replicas should be processed) */ static bool create_replica_actions(pcmk__bundle_replica_t *replica, void *user_data) { if (replica->ip != NULL) { replica->ip->cmds->create_actions(replica->ip); } if (replica->container != NULL) { replica->container->cmds->create_actions(replica->container); } if (replica->remote != NULL) { replica->remote->cmds->create_actions(replica->remote); } return true; } /*! * \internal * \brief Create all actions needed for a given bundle resource * * \param[in,out] rsc Bundle resource to create actions for */ void pcmk__bundle_create_actions(pcmk_resource_t *rsc) { pcmk_action_t *action = NULL; GList *containers = NULL; pcmk_resource_t *bundled_resource = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle)); pe__foreach_bundle_replica(rsc, create_replica_actions, NULL); containers = pe__bundle_containers(rsc); pcmk__create_instance_actions(rsc, containers); g_list_free(containers); bundled_resource = pe__bundled_resource(rsc); if (bundled_resource != NULL) { bundled_resource->cmds->create_actions(bundled_resource); if (pcmk_is_set(bundled_resource->flags, pcmk_rsc_promotable)) { pe__new_rsc_pseudo_action(rsc, PCMK_ACTION_PROMOTE, true, true); action = pe__new_rsc_pseudo_action(rsc, PCMK_ACTION_PROMOTED, true, true); action->priority = INFINITY; pe__new_rsc_pseudo_action(rsc, PCMK_ACTION_DEMOTE, true, true); action = pe__new_rsc_pseudo_action(rsc, PCMK_ACTION_DEMOTED, true, true); action->priority = INFINITY; } } } /*! * \internal * \brief Create internal constraints for a bundle replica's resources * * \param[in,out] replica Replica to create internal constraints for * \param[in,out] user_data Replica's parent bundle * * \return true (to indicate that any further replicas should be processed) */ static bool replica_internal_constraints(pcmk__bundle_replica_t *replica, void *user_data) { pcmk_resource_t *bundle = user_data; replica->container->cmds->internal_constraints(replica->container); // Start bundle -> start replica container pcmk__order_starts(bundle, replica->container, pcmk__ar_unrunnable_first_blocks |pcmk__ar_then_implies_first_graphed); // Stop bundle -> stop replica child and container if (replica->child != NULL) { pcmk__order_stops(bundle, replica->child, pcmk__ar_then_implies_first_graphed); } pcmk__order_stops(bundle, replica->container, pcmk__ar_then_implies_first_graphed); // Start replica container -> bundle is started pcmk__order_resource_actions(replica->container, PCMK_ACTION_START, bundle, PCMK_ACTION_RUNNING, pcmk__ar_first_implies_then_graphed); // Stop replica container -> bundle is stopped pcmk__order_resource_actions(replica->container, PCMK_ACTION_STOP, bundle, PCMK_ACTION_STOPPED, pcmk__ar_first_implies_then_graphed); if (replica->ip != NULL) { replica->ip->cmds->internal_constraints(replica->ip); // Replica IP address -> replica container (symmetric) pcmk__order_starts(replica->ip, replica->container, pcmk__ar_unrunnable_first_blocks |pcmk__ar_guest_allowed); pcmk__order_stops(replica->container, replica->ip, pcmk__ar_then_implies_first|pcmk__ar_guest_allowed); pcmk__new_colocation("#ip-with-container", NULL, INFINITY, replica->ip, replica->container, NULL, NULL, pcmk__coloc_influence); } if (replica->remote != NULL) { /* This handles ordering and colocating remote relative to container * (via "#resource-with-container"). Since IP is also ordered and * colocated relative to the container, we don't need to do anything * explicit here with IP. */ replica->remote->cmds->internal_constraints(replica->remote); } if (replica->child != NULL) { CRM_ASSERT(replica->remote != NULL); // "Start remote then child" is implicit in scheduler's remote logic } return true; } /*! * \internal * \brief Create implicit constraints needed for a bundle resource * * \param[in,out] rsc Bundle resource to create implicit constraints for */ void pcmk__bundle_internal_constraints(pcmk_resource_t *rsc) { pcmk_resource_t *bundled_resource = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle)); pe__foreach_bundle_replica(rsc, replica_internal_constraints, rsc); bundled_resource = pe__bundled_resource(rsc); if (bundled_resource == NULL) { return; } // Start bundle -> start bundled clone pcmk__order_resource_actions(rsc, PCMK_ACTION_START, bundled_resource, PCMK_ACTION_START, pcmk__ar_then_implies_first_graphed); // Bundled clone is started -> bundle is started pcmk__order_resource_actions(bundled_resource, PCMK_ACTION_RUNNING, rsc, PCMK_ACTION_RUNNING, pcmk__ar_first_implies_then_graphed); // Stop bundle -> stop bundled clone pcmk__order_resource_actions(rsc, PCMK_ACTION_STOP, bundled_resource, PCMK_ACTION_STOP, pcmk__ar_then_implies_first_graphed); // Bundled clone is stopped -> bundle is stopped pcmk__order_resource_actions(bundled_resource, PCMK_ACTION_STOPPED, rsc, PCMK_ACTION_STOPPED, pcmk__ar_first_implies_then_graphed); bundled_resource->cmds->internal_constraints(bundled_resource); if (!pcmk_is_set(bundled_resource->flags, pcmk_rsc_promotable)) { return; } pcmk__promotable_restart_ordering(rsc); // Demote bundle -> demote bundled clone pcmk__order_resource_actions(rsc, PCMK_ACTION_DEMOTE, bundled_resource, PCMK_ACTION_DEMOTE, pcmk__ar_then_implies_first_graphed); // Bundled clone is demoted -> bundle is demoted pcmk__order_resource_actions(bundled_resource, PCMK_ACTION_DEMOTED, rsc, PCMK_ACTION_DEMOTED, pcmk__ar_first_implies_then_graphed); // Promote bundle -> promote bundled clone pcmk__order_resource_actions(rsc, PCMK_ACTION_PROMOTE, bundled_resource, PCMK_ACTION_PROMOTE, pcmk__ar_then_implies_first_graphed); // Bundled clone is promoted -> bundle is promoted pcmk__order_resource_actions(bundled_resource, PCMK_ACTION_PROMOTED, rsc, PCMK_ACTION_PROMOTED, pcmk__ar_first_implies_then_graphed); } struct match_data { const pcmk_node_t *node; // Node to compare against replica pcmk_resource_t *container; // Replica container corresponding to node }; /*! * \internal * \brief Check whether a replica container is assigned to a given node * * \param[in] replica Replica to check * \param[in,out] user_data struct match_data with node to compare against * * \return true if the replica does not match (to indicate further replicas * should be processed), otherwise false */ static bool match_replica_container(const pcmk__bundle_replica_t *replica, void *user_data) { struct match_data *match_data = user_data; if (pcmk__instance_matches(replica->container, match_data->node, pcmk_role_unknown, false)) { match_data->container = replica->container; return false; // Match found, don't bother searching further replicas } return true; // No match, keep searching } /*! * \internal * \brief Get the host to which a bundle node is assigned * * \param[in] node Possible bundle node to check * * \return Node to which the container for \p node is assigned if \p node is a * bundle node, otherwise \p node itself */ static const pcmk_node_t * get_bundle_node_host(const pcmk_node_t *node) { if (pe__is_bundle_node(node)) { const pcmk_resource_t *container = node->details->remote_rsc->container; return container->fns->location(container, NULL, 0); } return node; } /*! * \internal * \brief Find a bundle container compatible with a dependent resource * * \param[in] dependent Dependent resource in colocation with bundle * \param[in] bundle Bundle that \p dependent is colocated with * * \return A container from \p bundle assigned to the same node as \p dependent * if assigned, otherwise assigned to any of dependent's allowed nodes, * otherwise NULL. */ static pcmk_resource_t * compatible_container(const pcmk_resource_t *dependent, const pcmk_resource_t *bundle) { GList *scratch = NULL; struct match_data match_data = { NULL, NULL }; // If dependent is assigned, only check there match_data.node = dependent->fns->location(dependent, NULL, 0); match_data.node = get_bundle_node_host(match_data.node); if (match_data.node != NULL) { pe__foreach_const_bundle_replica(bundle, match_replica_container, &match_data); return match_data.container; } // Otherwise, check for any of the dependent's allowed nodes scratch = g_hash_table_get_values(dependent->allowed_nodes); scratch = pcmk__sort_nodes(scratch, NULL); for (const GList *iter = scratch; iter != NULL; iter = iter->next) { match_data.node = iter->data; match_data.node = get_bundle_node_host(match_data.node); if (match_data.node == NULL) { continue; } pe__foreach_const_bundle_replica(bundle, match_replica_container, &match_data); if (match_data.container != NULL) { break; } } g_list_free(scratch); return match_data.container; } struct coloc_data { const pcmk__colocation_t *colocation; pcmk_resource_t *dependent; GList *container_hosts; }; /*! * \internal * \brief Apply a colocation score to replica node scores or resource priority * * \param[in] replica Replica of primary bundle resource in colocation * \param[in,out] user_data struct coloc_data for colocation being applied * * \return true (to indicate that any further replicas should be processed) */ static bool replica_apply_coloc_score(const pcmk__bundle_replica_t *replica, void *user_data) { struct coloc_data *coloc_data = user_data; pcmk_node_t *chosen = NULL; if (coloc_data->colocation->score < INFINITY) { replica->container->cmds->apply_coloc_score(coloc_data->dependent, replica->container, coloc_data->colocation, false); return true; } chosen = replica->container->fns->location(replica->container, NULL, 0); if ((chosen == NULL) || is_set_recursive(replica->container, pcmk_rsc_blocked, true)) { return true; } if ((coloc_data->colocation->primary_role >= pcmk_role_promoted) && ((replica->child == NULL) || (replica->child->next_role < pcmk_role_promoted))) { return true; } pcmk__rsc_trace(pe__const_top_resource(replica->container, true), "Allowing mandatory colocation %s using %s @%d", coloc_data->colocation->id, pe__node_name(chosen), chosen->weight); coloc_data->container_hosts = g_list_prepend(coloc_data->container_hosts, chosen); return true; } /*! * \internal * \brief Apply a colocation's score to node scores or resource priority * * Given a colocation constraint, apply its score to the dependent's * allowed node scores (if we are still placing resources) or priority (if * we are choosing promotable clone instance roles). * * \param[in,out] dependent Dependent resource in colocation * \param[in] primary Primary resource in colocation * \param[in] colocation Colocation constraint to apply * \param[in] for_dependent true if called on behalf of dependent */ void pcmk__bundle_apply_coloc_score(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent) { struct coloc_data coloc_data = { colocation, dependent, NULL }; /* This should never be called for the bundle itself as a dependent. * Instead, we add its colocation constraints to its containers and bundled * primitive and call the apply_coloc_score() method for them as dependents. */ CRM_ASSERT((primary != NULL) && (primary->variant == pcmk_rsc_variant_bundle) && (dependent != NULL) && (dependent->variant == pcmk_rsc_variant_primitive) && (colocation != NULL) && !for_dependent); if (pcmk_is_set(primary->flags, pcmk_rsc_unassigned)) { pcmk__rsc_trace(primary, "Skipping applying colocation %s " "because %s is still provisional", colocation->id, primary->id); return; } pcmk__rsc_trace(primary, "Applying colocation %s (%s with %s at %s)", colocation->id, dependent->id, primary->id, pcmk_readable_score(colocation->score)); /* If the constraint dependent is a clone or bundle, "dependent" here is one * of its instances. Look for a compatible instance of this bundle. */ if (colocation->dependent->variant > pcmk_rsc_variant_group) { const pcmk_resource_t *primary_container = NULL; primary_container = compatible_container(dependent, primary); if (primary_container != NULL) { // Success, we found one pcmk__rsc_debug(primary, "Pairing %s with %s", dependent->id, primary_container->id); dependent->cmds->apply_coloc_score(dependent, primary_container, colocation, true); } else if (colocation->score >= INFINITY) { // Failure, and it's fatal crm_notice("%s cannot run because there is no compatible " "instance of %s to colocate with", dependent->id, primary->id); pcmk__assign_resource(dependent, NULL, true, true); } else { // Failure, but we can ignore it pcmk__rsc_debug(primary, "%s cannot be colocated with any instance of %s", dependent->id, primary->id); } return; } pe__foreach_const_bundle_replica(primary, replica_apply_coloc_score, &coloc_data); if (colocation->score >= INFINITY) { pcmk__colocation_intersect_nodes(dependent, primary, colocation, coloc_data.container_hosts, false); } g_list_free(coloc_data.container_hosts); } // Bundle implementation of pcmk_assignment_methods_t:with_this_colocations() void pcmk__with_bundle_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list) { const pcmk_resource_t *bundled_rsc = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle) && (orig_rsc != NULL) && (list != NULL)); // The bundle itself and its containers always get its colocations if ((orig_rsc == rsc) || pcmk_is_set(orig_rsc->flags, pcmk_rsc_replica_container)) { pcmk__add_with_this_list(list, rsc->rsc_cons_lhs, orig_rsc); return; } /* The bundled resource gets the colocations if it's promotable and we've * begun choosing roles */ bundled_rsc = pe__bundled_resource(rsc); if ((bundled_rsc == NULL) || !pcmk_is_set(bundled_rsc->flags, pcmk_rsc_promotable) || (pe__const_top_resource(orig_rsc, false) != bundled_rsc)) { return; } if (orig_rsc == bundled_rsc) { if (pe__clone_flag_is_set(orig_rsc, pcmk__clone_promotion_constrained)) { /* orig_rsc is the clone and we're setting roles (or have already * done so) */ pcmk__add_with_this_list(list, rsc->rsc_cons_lhs, orig_rsc); } } else if (!pcmk_is_set(orig_rsc->flags, pcmk_rsc_unassigned)) { /* orig_rsc is an instance and is already assigned. If something * requests colocations for orig_rsc now, it's for setting roles. */ pcmk__add_with_this_list(list, rsc->rsc_cons_lhs, orig_rsc); } } // Bundle implementation of pcmk_assignment_methods_t:this_with_colocations() void pcmk__bundle_with_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list) { const pcmk_resource_t *bundled_rsc = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle) && (orig_rsc != NULL) && (list != NULL)); // The bundle itself and its containers always get its colocations if ((orig_rsc == rsc) || pcmk_is_set(orig_rsc->flags, pcmk_rsc_replica_container)) { pcmk__add_this_with_list(list, rsc->rsc_cons, orig_rsc); return; } /* The bundled resource gets the colocations if it's promotable and we've * begun choosing roles */ bundled_rsc = pe__bundled_resource(rsc); if ((bundled_rsc == NULL) || !pcmk_is_set(bundled_rsc->flags, pcmk_rsc_promotable) || (pe__const_top_resource(orig_rsc, false) != bundled_rsc)) { return; } if (orig_rsc == bundled_rsc) { if (pe__clone_flag_is_set(orig_rsc, pcmk__clone_promotion_constrained)) { /* orig_rsc is the clone and we're setting roles (or have already * done so) */ pcmk__add_this_with_list(list, rsc->rsc_cons, orig_rsc); } } else if (!pcmk_is_set(orig_rsc->flags, pcmk_rsc_unassigned)) { /* orig_rsc is an instance and is already assigned. If something * requests colocations for orig_rsc now, it's for setting roles. */ pcmk__add_this_with_list(list, rsc->rsc_cons, orig_rsc); } } /*! * \internal * \brief Return action flags for a given bundle resource action * * \param[in,out] action Bundle resource action to get flags for * \param[in] node If not NULL, limit effects to this node * * \return Flags appropriate to \p action on \p node */ uint32_t pcmk__bundle_action_flags(pcmk_action_t *action, const pcmk_node_t *node) { GList *containers = NULL; uint32_t flags = 0; pcmk_resource_t *bundled_resource = NULL; CRM_ASSERT((action != NULL) && (action->rsc != NULL) && (action->rsc->variant == pcmk_rsc_variant_bundle)); bundled_resource = pe__bundled_resource(action->rsc); if (bundled_resource != NULL) { // Clone actions are done on the bundled clone resource, not container switch (get_complex_task(bundled_resource, action->task)) { case pcmk_action_unspecified: case pcmk_action_notify: case pcmk_action_notified: case pcmk_action_promote: case pcmk_action_promoted: case pcmk_action_demote: case pcmk_action_demoted: return pcmk__collective_action_flags(action, bundled_resource->children, node); default: break; } } containers = pe__bundle_containers(action->rsc); flags = pcmk__collective_action_flags(action, containers, node); g_list_free(containers); return flags; } /*! * \internal * \brief Apply a location constraint to a bundle replica * * \param[in,out] replica Replica to apply constraint to * \param[in,out] user_data Location constraint to apply * * \return true (to indicate that any further replicas should be processed) */ static bool apply_location_to_replica(pcmk__bundle_replica_t *replica, void *user_data) { pcmk__location_t *location = user_data; if (replica->container != NULL) { replica->container->cmds->apply_location(replica->container, location); } if (replica->ip != NULL) { replica->ip->cmds->apply_location(replica->ip, location); } return true; } /*! * \internal * \brief Apply a location constraint to a bundle resource's allowed node scores * * \param[in,out] rsc Bundle resource to apply constraint to * \param[in,out] location Location constraint to apply */ void pcmk__bundle_apply_location(pcmk_resource_t *rsc, pcmk__location_t *location) { pcmk_resource_t *bundled_resource = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle) && (location != NULL)); pcmk__apply_location(rsc, location); pe__foreach_bundle_replica(rsc, apply_location_to_replica, location); bundled_resource = pe__bundled_resource(rsc); if ((bundled_resource != NULL) && ((location->role_filter == pcmk_role_unpromoted) || (location->role_filter == pcmk_role_promoted))) { bundled_resource->cmds->apply_location(bundled_resource, location); bundled_resource->rsc_location = g_list_prepend( bundled_resource->rsc_location, location); } } -#define XPATH_REMOTE "//nvpair[@name='" XML_RSC_ATTR_REMOTE_RA_ADDR "']" +#define XPATH_REMOTE "//nvpair[@name='" PCMK_REMOTE_RA_ADDR "']" /*! * \internal * \brief Add a bundle replica's actions to transition graph * * \param[in,out] replica Replica to add to graph * \param[in] user_data Bundle that replica belongs to (for logging only) * * \return true (to indicate that any further replicas should be processed) */ static bool add_replica_actions_to_graph(pcmk__bundle_replica_t *replica, void *user_data) { if ((replica->remote != NULL) && (replica->container != NULL) && pe__bundle_needs_remote_name(replica->remote)) { /* REMOTE_CONTAINER_HACK: Allow remote nodes to run containers that * run pacemaker-remoted inside, without needing a separate IP for * the container. This is done by configuring the inner remote's * connection host as the magic string "#uname", then * replacing it with the underlying host when needed. */ xmlNode *nvpair = get_xpath_object(XPATH_REMOTE, replica->remote->xml, LOG_ERR); const char *calculated_addr = NULL; // Replace the value in replica->remote->xml (if appropriate) calculated_addr = pe__add_bundle_remote_name(replica->remote, replica->remote->cluster, nvpair, "value"); if (calculated_addr != NULL) { /* Since this is for the bundle as a resource, and not any * particular action, replace the value in the default * parameters (not evaluated for node). create_graph_action() * will grab it from there to replace it in node-evaluated * parameters. */ GHashTable *params = pe_rsc_params(replica->remote, NULL, replica->remote->cluster); g_hash_table_replace(params, - strdup(XML_RSC_ATTR_REMOTE_RA_ADDR), + strdup(PCMK_REMOTE_RA_ADDR), strdup(calculated_addr)); } else { pcmk_resource_t *bundle = user_data; /* The only way to get here is if the remote connection is * neither currently running nor scheduled to run. That means we * won't be doing any operations that require addr (only start * requires it; we additionally use it to compare digests when * unpacking status, promote, and migrate_from history, but * that's already happened by this point). */ pcmk__rsc_info(bundle, "Unable to determine address for bundle %s " "remote connection", bundle->id); } } if (replica->ip != NULL) { replica->ip->cmds->add_actions_to_graph(replica->ip); } if (replica->container != NULL) { replica->container->cmds->add_actions_to_graph(replica->container); } if (replica->remote != NULL) { replica->remote->cmds->add_actions_to_graph(replica->remote); } return true; } /*! * \internal * \brief Add a bundle resource's actions to the transition graph * * \param[in,out] rsc Bundle resource whose actions should be added */ void pcmk__bundle_add_actions_to_graph(pcmk_resource_t *rsc) { pcmk_resource_t *bundled_resource = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle)); bundled_resource = pe__bundled_resource(rsc); if (bundled_resource != NULL) { bundled_resource->cmds->add_actions_to_graph(bundled_resource); } pe__foreach_bundle_replica(rsc, add_replica_actions_to_graph, rsc); } struct probe_data { pcmk_resource_t *bundle; // Bundle being probed pcmk_node_t *node; // Node to create probes on bool any_created; // Whether any probes have been created }; /*! * \internal * \brief Order a bundle replica's start after another replica's probe * * \param[in,out] replica Replica to order start for * \param[in,out] user_data Replica with probe to order after * * \return true (to indicate that any further replicas should be processed) */ static bool order_replica_start_after(pcmk__bundle_replica_t *replica, void *user_data) { pcmk__bundle_replica_t *probed_replica = user_data; if ((replica == probed_replica) || (replica->container == NULL)) { return true; } pcmk__new_ordering(probed_replica->container, pcmk__op_key(probed_replica->container->id, PCMK_ACTION_MONITOR, 0), NULL, replica->container, pcmk__op_key(replica->container->id, PCMK_ACTION_START, 0), NULL, pcmk__ar_ordered|pcmk__ar_if_on_same_node, replica->container->cluster); return true; } /*! * \internal * \brief Create probes for a bundle replica's resources * * \param[in,out] replica Replica to create probes for * \param[in,out] user_data struct probe_data * * \return true (to indicate that any further replicas should be processed) */ static bool create_replica_probes(pcmk__bundle_replica_t *replica, void *user_data) { struct probe_data *probe_data = user_data; if ((replica->ip != NULL) && replica->ip->cmds->create_probe(replica->ip, probe_data->node)) { probe_data->any_created = true; } if ((replica->child != NULL) && pe__same_node(probe_data->node, replica->node) && replica->child->cmds->create_probe(replica->child, probe_data->node)) { probe_data->any_created = true; } if ((replica->container != NULL) && replica->container->cmds->create_probe(replica->container, probe_data->node)) { probe_data->any_created = true; /* If we're limited to one replica per host (due to * the lack of an IP range probably), then we don't * want any of our peer containers starting until * we've established that no other copies are already * running. * * Partly this is to ensure that the maximum replicas per host is * observed, but also to ensure that the containers * don't fail to start because the necessary port * mappings (which won't include an IP for uniqueness) * are already taken */ if (probe_data->bundle->fns->max_per_node(probe_data->bundle) == 1) { pe__foreach_bundle_replica(probe_data->bundle, order_replica_start_after, replica); } } if ((replica->container != NULL) && (replica->remote != NULL) && replica->remote->cmds->create_probe(replica->remote, probe_data->node)) { /* Do not probe the remote resource until we know where the container is * running. This is required for REMOTE_CONTAINER_HACK to correctly * probe remote resources. */ char *probe_uuid = pcmk__op_key(replica->remote->id, PCMK_ACTION_MONITOR, 0); pcmk_action_t *probe = find_first_action(replica->remote->actions, probe_uuid, NULL, probe_data->node); free(probe_uuid); if (probe != NULL) { probe_data->any_created = true; pcmk__rsc_trace(probe_data->bundle, "Ordering %s probe on %s", replica->remote->id, pe__node_name(probe_data->node)); pcmk__new_ordering(replica->container, pcmk__op_key(replica->container->id, PCMK_ACTION_START, 0), NULL, replica->remote, NULL, probe, pcmk__ar_nested_remote_probe, probe_data->bundle->cluster); } } return true; } /*! * \internal * * \brief Schedule any probes needed for a bundle resource on a node * * \param[in,out] rsc Bundle resource to create probes for * \param[in,out] node Node to create probe on * * \return true if any probe was created, otherwise false */ bool pcmk__bundle_create_probe(pcmk_resource_t *rsc, pcmk_node_t *node) { struct probe_data probe_data = { rsc, node, false }; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle)); pe__foreach_bundle_replica(rsc, create_replica_probes, &probe_data); return probe_data.any_created; } /*! * \internal * \brief Output actions for one bundle replica * * \param[in,out] replica Replica to output actions for * \param[in] user_data Unused * * \return true (to indicate that any further replicas should be processed) */ static bool output_replica_actions(pcmk__bundle_replica_t *replica, void *user_data) { if (replica->ip != NULL) { replica->ip->cmds->output_actions(replica->ip); } if (replica->container != NULL) { replica->container->cmds->output_actions(replica->container); } if (replica->remote != NULL) { replica->remote->cmds->output_actions(replica->remote); } if (replica->child != NULL) { replica->child->cmds->output_actions(replica->child); } return true; } /*! * \internal * \brief Output a summary of scheduled actions for a bundle resource * * \param[in,out] rsc Bundle resource to output actions for */ void pcmk__output_bundle_actions(pcmk_resource_t *rsc) { CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle)); pe__foreach_bundle_replica(rsc, output_replica_actions, NULL); } // Bundle implementation of pcmk_assignment_methods_t:add_utilization() void pcmk__bundle_add_utilization(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization) { pcmk_resource_t *container = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle)); if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) { return; } /* All bundle replicas are identical, so using the utilization of the first * is sufficient for any. Only the implicit container resource can have * utilization values. */ container = pe__first_container(rsc); if (container != NULL) { container->cmds->add_utilization(container, orig_rsc, all_rscs, utilization); } } // Bundle implementation of pcmk_assignment_methods_t:shutdown_lock() void pcmk__bundle_shutdown_lock(pcmk_resource_t *rsc) { CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle)); // Bundles currently don't support shutdown locks } diff --git a/lib/pacemaker/pcmk_sched_remote.c b/lib/pacemaker/pcmk_sched_remote.c index d952dbcb24..c631807a45 100644 --- a/lib/pacemaker/pcmk_sched_remote.c +++ b/lib/pacemaker/pcmk_sched_remote.c @@ -1,735 +1,734 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include "libpacemaker_private.h" enum remote_connection_state { remote_state_unknown = 0, remote_state_alive = 1, remote_state_resting = 2, remote_state_failed = 3, remote_state_stopped = 4 }; static const char * state2text(enum remote_connection_state state) { switch (state) { case remote_state_unknown: return "unknown"; case remote_state_alive: return "alive"; case remote_state_resting: return "resting"; case remote_state_failed: return "failed"; case remote_state_stopped: return "stopped"; } return "impossible"; } /* We always use pcmk__ar_guest_allowed with these convenience functions to * exempt internally generated constraints from the prohibition of user * constraints involving remote connection resources. * * The start ordering additionally uses pcmk__ar_unrunnable_first_blocks so that * the specified action is not runnable if the start is not runnable. */ static inline void order_start_then_action(pcmk_resource_t *first_rsc, pcmk_action_t *then_action, uint32_t extra) { if ((first_rsc != NULL) && (then_action != NULL)) { pcmk__new_ordering(first_rsc, start_key(first_rsc), NULL, then_action->rsc, NULL, then_action, pcmk__ar_guest_allowed |pcmk__ar_unrunnable_first_blocks |extra, first_rsc->cluster); } } static inline void order_action_then_stop(pcmk_action_t *first_action, pcmk_resource_t *then_rsc, uint32_t extra) { if ((first_action != NULL) && (then_rsc != NULL)) { pcmk__new_ordering(first_action->rsc, NULL, first_action, then_rsc, stop_key(then_rsc), NULL, pcmk__ar_guest_allowed|extra, then_rsc->cluster); } } static enum remote_connection_state get_remote_node_state(const pcmk_node_t *node) { const pcmk_resource_t *remote_rsc = NULL; const pcmk_node_t *cluster_node = NULL; CRM_ASSERT(node != NULL); remote_rsc = node->details->remote_rsc; CRM_ASSERT(remote_rsc != NULL); cluster_node = pe__current_node(remote_rsc); /* If the cluster node the remote connection resource resides on * is unclean or went offline, we can't process any operations * on that remote node until after it starts elsewhere. */ if ((remote_rsc->next_role == pcmk_role_stopped) || (remote_rsc->allocated_to == NULL)) { // The connection resource is not going to run anywhere if ((cluster_node != NULL) && cluster_node->details->unclean) { /* The remote connection is failed because its resource is on a * failed node and can't be recovered elsewhere, so we must fence. */ return remote_state_failed; } if (!pcmk_is_set(remote_rsc->flags, pcmk_rsc_failed)) { /* Connection resource is cleanly stopped */ return remote_state_stopped; } /* Connection resource is failed */ if ((remote_rsc->next_role == pcmk_role_stopped) && remote_rsc->remote_reconnect_ms && node->details->remote_was_fenced && !pe__shutdown_requested(node)) { /* We won't know whether the connection is recoverable until the * reconnect interval expires and we reattempt connection. */ return remote_state_unknown; } /* The remote connection is in a failed state. If there are any * resources known to be active on it (stop) or in an unknown state * (probe), we must assume the worst and fence it. */ return remote_state_failed; } else if (cluster_node == NULL) { /* Connection is recoverable but not currently running anywhere, so see * if we can recover it first */ return remote_state_unknown; } else if (cluster_node->details->unclean || !(cluster_node->details->online)) { // Connection is running on a dead node, see if we can recover it first return remote_state_resting; } else if (pcmk__list_of_multiple(remote_rsc->running_on) && (remote_rsc->partial_migration_source != NULL) && (remote_rsc->partial_migration_target != NULL)) { /* We're in the middle of migrating a connection resource, so wait until * after the migration completes before performing any actions. */ return remote_state_resting; } return remote_state_alive; } /*! * \internal * \brief Order actions on remote node relative to actions for the connection * * \param[in,out] action An action scheduled on a Pacemaker Remote node */ static void apply_remote_ordering(pcmk_action_t *action) { pcmk_resource_t *remote_rsc = NULL; enum action_tasks task = text2task(action->task); enum remote_connection_state state = get_remote_node_state(action->node); uint32_t order_opts = pcmk__ar_none; if (action->rsc == NULL) { return; } CRM_ASSERT(pe__is_guest_or_remote_node(action->node)); remote_rsc = action->node->details->remote_rsc; CRM_ASSERT(remote_rsc != NULL); crm_trace("Order %s action %s relative to %s%s (state: %s)", action->task, action->uuid, pcmk_is_set(remote_rsc->flags, pcmk_rsc_failed)? "failed " : "", remote_rsc->id, state2text(state)); if (pcmk__strcase_any_of(action->task, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, NULL)) { /* Migration ops map to pcmk_action_unspecified, but we need to apply * the same ordering as for stop or demote (see get_router_node()). */ task = pcmk_action_stop; } switch (task) { case pcmk_action_start: case pcmk_action_promote: order_opts = pcmk__ar_none; if (state == remote_state_failed) { /* Force recovery, by making this action required */ pe__set_order_flags(order_opts, pcmk__ar_first_implies_then); } /* Ensure connection is up before running this action */ order_start_then_action(remote_rsc, action, order_opts); break; case pcmk_action_stop: if (state == remote_state_alive) { order_action_then_stop(action, remote_rsc, pcmk__ar_then_implies_first); } else if (state == remote_state_failed) { /* The resource is active on the node, but since we don't have a * valid connection, the only way to stop the resource is by * fencing the node. There is no need to order the stop relative * to the remote connection, since the stop will become implied * by the fencing. */ pe_fence_node(remote_rsc->cluster, action->node, "resources are active but " "connection is unrecoverable", FALSE); } else if (remote_rsc->next_role == pcmk_role_stopped) { /* State must be remote_state_unknown or remote_state_stopped. * Since the connection is not coming back up in this * transition, stop this resource first. */ order_action_then_stop(action, remote_rsc, pcmk__ar_then_implies_first); } else { /* The connection is going to be started somewhere else, so * stop this resource after that completes. */ order_start_then_action(remote_rsc, action, pcmk__ar_none); } break; case pcmk_action_demote: /* Only order this demote relative to the connection start if the * connection isn't being torn down. Otherwise, the demote would be * blocked because the connection start would not be allowed. */ if ((state == remote_state_resting) || (state == remote_state_unknown)) { order_start_then_action(remote_rsc, action, pcmk__ar_none); } /* Otherwise we can rely on the stop ordering */ break; default: /* Wait for the connection resource to be up */ if (pcmk__action_is_recurring(action)) { /* In case we ever get the recovery logic wrong, force * recurring monitors to be restarted, even if just * the connection was re-established */ order_start_then_action(remote_rsc, action, pcmk__ar_first_implies_then); } else { pcmk_node_t *cluster_node = pe__current_node(remote_rsc); if ((task == pcmk_action_monitor) && (state == remote_state_failed)) { /* We would only be here if we do not know the state of the * resource on the remote node. Since we have no way to find * out, it is necessary to fence the node. */ pe_fence_node(remote_rsc->cluster, action->node, "resources are in unknown state " "and connection is unrecoverable", FALSE); } if ((cluster_node != NULL) && (state == remote_state_stopped)) { /* The connection is currently up, but is going down * permanently. Make sure we check services are actually * stopped _before_ we let the connection get closed. */ order_action_then_stop(action, remote_rsc, pcmk__ar_unrunnable_first_blocks); } else { order_start_then_action(remote_rsc, action, pcmk__ar_none); } } break; } } static void apply_container_ordering(pcmk_action_t *action) { /* VMs are also classified as containers for these purposes... in * that they both involve a 'thing' running on a real or remote * cluster node. * * This allows us to be smarter about the type and extent of * recovery actions required in various scenarios */ pcmk_resource_t *remote_rsc = NULL; pcmk_resource_t *container = NULL; enum action_tasks task = text2task(action->task); CRM_ASSERT(action->rsc != NULL); CRM_ASSERT(action->node != NULL); CRM_ASSERT(pe__is_guest_or_remote_node(action->node)); remote_rsc = action->node->details->remote_rsc; CRM_ASSERT(remote_rsc != NULL); container = remote_rsc->container; CRM_ASSERT(container != NULL); if (pcmk_is_set(container->flags, pcmk_rsc_failed)) { pe_fence_node(action->rsc->cluster, action->node, "container failed", FALSE); } crm_trace("Order %s action %s relative to %s%s for %s%s", action->task, action->uuid, pcmk_is_set(remote_rsc->flags, pcmk_rsc_failed)? "failed " : "", remote_rsc->id, pcmk_is_set(container->flags, pcmk_rsc_failed)? "failed " : "", container->id); if (pcmk__strcase_any_of(action->task, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, NULL)) { /* Migration ops map to pcmk_action_unspecified, but we need to apply * the same ordering as for stop or demote (see get_router_node()). */ task = pcmk_action_stop; } switch (task) { case pcmk_action_start: case pcmk_action_promote: // Force resource recovery if the container is recovered order_start_then_action(container, action, pcmk__ar_first_implies_then); // Wait for the connection resource to be up, too order_start_then_action(remote_rsc, action, pcmk__ar_none); break; case pcmk_action_stop: case pcmk_action_demote: if (pcmk_is_set(container->flags, pcmk_rsc_failed)) { /* When the container representing a guest node fails, any stop * or demote actions for resources running on the guest node * are implied by the container stopping. This is similar to * how fencing operations work for cluster nodes and remote * nodes. */ } else { /* Ensure the operation happens before the connection is brought * down. * * If we really wanted to, we could order these after the * connection start, IFF the container's current role was * stopped (otherwise we re-introduce an ordering loop when the * connection is restarting). */ order_action_then_stop(action, remote_rsc, pcmk__ar_none); } break; default: /* Wait for the connection resource to be up */ if (pcmk__action_is_recurring(action)) { /* In case we ever get the recovery logic wrong, force * recurring monitors to be restarted, even if just * the connection was re-established */ if (task != pcmk_action_unspecified) { order_start_then_action(remote_rsc, action, pcmk__ar_first_implies_then); } } else { order_start_then_action(remote_rsc, action, pcmk__ar_none); } break; } } /*! * \internal * \brief Order all relevant actions relative to remote connection actions * * \param[in,out] scheduler Scheduler data */ void pcmk__order_remote_connection_actions(pcmk_scheduler_t *scheduler) { if (!pcmk_is_set(scheduler->flags, pcmk_sched_have_remote_nodes)) { return; } crm_trace("Creating remote connection orderings"); for (GList *iter = scheduler->actions; iter != NULL; iter = iter->next) { pcmk_action_t *action = iter->data; pcmk_resource_t *remote = NULL; // We are only interested in resource actions if (action->rsc == NULL) { continue; } /* Special case: If we are clearing the failcount of an actual * remote connection resource, then make sure this happens before * any start of the resource in this transition. */ if (action->rsc->is_remote_node && pcmk__str_eq(action->task, PCMK_ACTION_CLEAR_FAILCOUNT, pcmk__str_none)) { pcmk__new_ordering(action->rsc, NULL, action, action->rsc, pcmk__op_key(action->rsc->id, PCMK_ACTION_START, 0), NULL, pcmk__ar_ordered, scheduler); continue; } // We are only interested in actions assigned to a node if (action->node == NULL) { continue; } if (!pe__is_guest_or_remote_node(action->node)) { continue; } /* We are only interested in real actions. * * @TODO This is probably wrong; pseudo-actions might be converted to * real actions and vice versa later in update_actions() at the end of * pcmk__apply_orderings(). */ if (pcmk_is_set(action->flags, pcmk_action_pseudo)) { continue; } remote = action->node->details->remote_rsc; if (remote == NULL) { // Orphaned continue; } /* Another special case: if a resource is moving to a Pacemaker Remote * node, order the stop on the original node after any start of the * remote connection. This ensures that if the connection fails to * start, we leave the resource running on the original node. */ if (pcmk__str_eq(action->task, PCMK_ACTION_START, pcmk__str_none)) { for (GList *item = action->rsc->actions; item != NULL; item = item->next) { pcmk_action_t *rsc_action = item->data; if (!pe__same_node(rsc_action->node, action->node) && pcmk__str_eq(rsc_action->task, PCMK_ACTION_STOP, pcmk__str_none)) { pcmk__new_ordering(remote, start_key(remote), NULL, action->rsc, NULL, rsc_action, pcmk__ar_ordered, scheduler); } } } /* The action occurs across a remote connection, so create * ordering constraints that guarantee the action occurs while the node * is active (after start, before stop ... things like that). * * This is somewhat brittle in that we need to make sure the results of * this ordering are compatible with the result of get_router_node(). * It would probably be better to add XML_LRM_ATTR_ROUTER_NODE as part * of this logic rather than create_graph_action(). */ if (remote->container) { crm_trace("Container ordering for %s", action->uuid); apply_container_ordering(action); } else { crm_trace("Remote ordering for %s", action->uuid); apply_remote_ordering(action); } } } /*! * \internal * \brief Check whether a node is a failed remote node * * \param[in] node Node to check * * \return true if \p node is a failed remote node, false otherwise */ bool pcmk__is_failed_remote_node(const pcmk_node_t *node) { return pe__is_remote_node(node) && (node->details->remote_rsc != NULL) && (get_remote_node_state(node) == remote_state_failed); } /*! * \internal * \brief Check whether a given resource corresponds to a given node as guest * * \param[in] rsc Resource to check * \param[in] node Node to check * * \return true if \p node is a guest node and \p rsc is its containing * resource, otherwise false */ bool pcmk__rsc_corresponds_to_guest(const pcmk_resource_t *rsc, const pcmk_node_t *node) { return (rsc != NULL) && (rsc->fillers != NULL) && (node != NULL) && (node->details->remote_rsc != NULL) && (node->details->remote_rsc->container == rsc); } /*! * \internal * \brief Get proper connection host that a remote action must be routed through * * A remote connection resource might be starting, stopping, or migrating in the * same transition that an action needs to be executed on its Pacemaker Remote * node. Determine the proper node that the remote action should be routed * through. * * \param[in] action (Potentially remote) action to route * * \return Connection host that action should be routed through if remote, * otherwise NULL */ pcmk_node_t * pcmk__connection_host_for_action(const pcmk_action_t *action) { pcmk_node_t *began_on = NULL; pcmk_node_t *ended_on = NULL; bool partial_migration = false; const char *task = action->task; if (pcmk__str_eq(task, PCMK_ACTION_STONITH, pcmk__str_none) || !pe__is_guest_or_remote_node(action->node)) { return NULL; } CRM_ASSERT(action->node->details->remote_rsc != NULL); began_on = pe__current_node(action->node->details->remote_rsc); ended_on = action->node->details->remote_rsc->allocated_to; if (action->node->details->remote_rsc && (action->node->details->remote_rsc->container == NULL) && action->node->details->remote_rsc->partial_migration_target) { partial_migration = true; } if (began_on == NULL) { crm_trace("Routing %s for %s through remote connection's " "next node %s (starting)%s", action->task, (action->rsc? action->rsc->id : "no resource"), (ended_on? ended_on->details->uname : "none"), partial_migration? " (partial migration)" : ""); return ended_on; } if (ended_on == NULL) { crm_trace("Routing %s for %s through remote connection's " "current node %s (stopping)%s", action->task, (action->rsc? action->rsc->id : "no resource"), (began_on? began_on->details->uname : "none"), partial_migration? " (partial migration)" : ""); return began_on; } if (pe__same_node(began_on, ended_on)) { crm_trace("Routing %s for %s through remote connection's " "current node %s (not moving)%s", action->task, (action->rsc? action->rsc->id : "no resource"), (began_on? began_on->details->uname : "none"), partial_migration? " (partial migration)" : ""); return began_on; } /* If we get here, the remote connection is moving during this transition. * This means some actions for resources behind the connection will get * routed through the cluster node the connection resource is currently on, * and others are routed through the cluster node the connection will end up * on. */ if (pcmk__str_eq(task, PCMK_ACTION_NOTIFY, pcmk__str_none)) { task = g_hash_table_lookup(action->meta, "notify_operation"); } /* * Stop, demote, and migration actions must occur before the connection can * move (these actions are required before the remote resource can stop). In * this case, we know these actions have to be routed through the initial * cluster node the connection resource lived on before the move takes * place. * * The exception is a partial migration of a (non-guest) remote connection * resource; in that case, all actions (even these) will be ordered after * the connection's pseudo-start on the migration target, so the target is * the router node. */ if (pcmk__strcase_any_of(task, PCMK_ACTION_CANCEL, PCMK_ACTION_STOP, PCMK_ACTION_DEMOTE, PCMK_ACTION_MIGRATE_FROM, PCMK_ACTION_MIGRATE_TO, NULL) && !partial_migration) { crm_trace("Routing %s for %s through remote connection's " "current node %s (moving)%s", action->task, (action->rsc? action->rsc->id : "no resource"), (began_on? began_on->details->uname : "none"), partial_migration? " (partial migration)" : ""); return began_on; } /* Everything else (start, promote, monitor, probe, refresh, * clear failcount, delete, ...) must occur after the connection starts on * the node it is moving to. */ crm_trace("Routing %s for %s through remote connection's " "next node %s (moving)%s", action->task, (action->rsc? action->rsc->id : "no resource"), (ended_on? ended_on->details->uname : "none"), partial_migration? " (partial migration)" : ""); return ended_on; } /*! * \internal * \brief Replace remote connection's addr="#uname" with actual address * * REMOTE_CONTAINER_HACK: If a given resource is a remote connection resource * with its "addr" parameter set to "#uname", pull the actual value from the * parameters evaluated without a node (which was put there earlier in * pcmk__create_graph() when the bundle's expand() method was called). * * \param[in,out] rsc Resource to check * \param[in,out] params Resource parameters evaluated per node */ void pcmk__substitute_remote_addr(pcmk_resource_t *rsc, GHashTable *params) { - const char *remote_addr = g_hash_table_lookup(params, - XML_RSC_ATTR_REMOTE_RA_ADDR); + const char *remote_addr = g_hash_table_lookup(params, PCMK_REMOTE_RA_ADDR); if (pcmk__str_eq(remote_addr, "#uname", pcmk__str_none)) { GHashTable *base = pe_rsc_params(rsc, NULL, rsc->cluster); - remote_addr = g_hash_table_lookup(base, XML_RSC_ATTR_REMOTE_RA_ADDR); + remote_addr = g_hash_table_lookup(base, PCMK_REMOTE_RA_ADDR); if (remote_addr != NULL) { - g_hash_table_insert(params, strdup(XML_RSC_ATTR_REMOTE_RA_ADDR), + g_hash_table_insert(params, strdup(PCMK_REMOTE_RA_ADDR), strdup(remote_addr)); } } } /*! * \brief Add special guest node meta-attributes to XML * * If a given action will be executed on a guest node, add the following as XML * attributes (using meta-attribute naming): * * The resource's \c PCMK_META_CONTAINER_ATTR_TARGET meta-attribute (usually * set only for bundles), as \c PCMK_META_CONTAINER_ATTR_TARGET * * The guest's physical host (current host for "down" actions, next host for * "up" actions), as \c PCMK__META_PHYSICAL_HOST * * If the guest node has no physical host, then don't add either attribute. * * \param[in,out] args_xml XML to add attributes to * \param[in] action Action to check */ void pcmk__add_guest_meta_to_xml(xmlNode *args_xml, const pcmk_action_t *action) { const pcmk_node_t *guest = action->node; const pcmk_node_t *host = NULL; enum action_tasks task; if (!pe__is_guest_node(guest)) { return; } task = text2task(action->task); if ((task == pcmk_action_notify) || (task == pcmk_action_notified)) { task = text2task(g_hash_table_lookup(action->meta, "notify_operation")); } switch (task) { case pcmk_action_stop: case pcmk_action_stopped: case pcmk_action_demote: case pcmk_action_demoted: // "Down" actions take place on guest's current host host = pe__current_node(guest->details->remote_rsc->container); break; case pcmk_action_start: case pcmk_action_started: case pcmk_action_monitor: case pcmk_action_promote: case pcmk_action_promoted: // "Up" actions take place on guest's next host host = guest->details->remote_rsc->container->allocated_to; break; default: break; } if (host != NULL) { gpointer target = g_hash_table_lookup(action->rsc->meta, PCMK_META_CONTAINER_ATTR_TARGET); hash2metafield((gpointer) PCMK_META_CONTAINER_ATTR_TARGET, target, (gpointer) args_xml); hash2metafield((gpointer) PCMK__META_PHYSICAL_HOST, (gpointer) host->details->uname, (gpointer) args_xml); } } diff --git a/lib/pengine/bundle.c b/lib/pengine/bundle.c index 2842d89c9f..c0de71d30f 100644 --- a/lib/pengine/bundle.c +++ b/lib/pengine/bundle.c @@ -1,2228 +1,2228 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include enum pe__bundle_mount_flags { pe__bundle_mount_none = 0x00, // mount instance-specific subdirectory rather than source directly pe__bundle_mount_subdir = 0x01 }; typedef struct { char *source; char *target; char *options; uint32_t flags; // bitmask of pe__bundle_mount_flags } pe__bundle_mount_t; typedef struct { char *source; char *target; } pe__bundle_port_t; enum pe__container_agent { PE__CONTAINER_AGENT_UNKNOWN, PE__CONTAINER_AGENT_DOCKER, PE__CONTAINER_AGENT_RKT, PE__CONTAINER_AGENT_PODMAN, }; #define PE__CONTAINER_AGENT_UNKNOWN_S "unknown" #define PE__CONTAINER_AGENT_DOCKER_S "docker" #define PE__CONTAINER_AGENT_RKT_S "rkt" #define PE__CONTAINER_AGENT_PODMAN_S "podman" typedef struct pe__bundle_variant_data_s { int promoted_max; int nreplicas; int nreplicas_per_host; char *prefix; char *image; const char *ip_last; char *host_network; char *host_netmask; char *control_port; char *container_network; char *ip_range_start; gboolean add_host; gchar *container_host_options; char *container_command; char *launcher_options; const char *attribute_target; pcmk_resource_t *child; GList *replicas; // pcmk__bundle_replica_t * GList *ports; // pe__bundle_port_t * GList *mounts; // pe__bundle_mount_t * enum pe__container_agent agent_type; } pe__bundle_variant_data_t; #define get_bundle_variant_data(data, rsc) \ CRM_ASSERT(rsc != NULL); \ CRM_ASSERT(rsc->variant == pcmk_rsc_variant_bundle); \ CRM_ASSERT(rsc->variant_opaque != NULL); \ data = (pe__bundle_variant_data_t *) rsc->variant_opaque; /*! * \internal * \brief Get maximum number of bundle replicas allowed to run * * \param[in] rsc Bundle or bundled resource to check * * \return Maximum replicas for bundle corresponding to \p rsc */ int pe__bundle_max(const pcmk_resource_t *rsc) { const pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, pe__const_top_resource(rsc, true)); return bundle_data->nreplicas; } /*! * \internal * \brief Get the resource inside a bundle * * \param[in] bundle Bundle to check * * \return Resource inside \p bundle if any, otherwise NULL */ pcmk_resource_t * pe__bundled_resource(const pcmk_resource_t *rsc) { const pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, pe__const_top_resource(rsc, true)); return bundle_data->child; } /*! * \internal * \brief Get containerized resource corresponding to a given bundle container * * \param[in] instance Collective instance that might be a bundle container * * \return Bundled resource instance inside \p instance if it is a bundle * container instance, otherwise NULL */ const pcmk_resource_t * pe__get_rsc_in_container(const pcmk_resource_t *instance) { const pe__bundle_variant_data_t *data = NULL; const pcmk_resource_t *top = pe__const_top_resource(instance, true); if ((top == NULL) || (top->variant != pcmk_rsc_variant_bundle)) { return NULL; } get_bundle_variant_data(data, top); for (const GList *iter = data->replicas; iter != NULL; iter = iter->next) { const pcmk__bundle_replica_t *replica = iter->data; if (instance == replica->container) { return replica->child; } } return NULL; } /*! * \internal * \brief Check whether a given node is created by a bundle * * \param[in] bundle Bundle resource to check * \param[in] node Node to check * * \return true if \p node is an instance of \p bundle, otherwise false */ bool pe__node_is_bundle_instance(const pcmk_resource_t *bundle, const pcmk_node_t *node) { pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, bundle); for (GList *iter = bundle_data->replicas; iter != NULL; iter = iter->next) { pcmk__bundle_replica_t *replica = iter->data; if (pe__same_node(node, replica->node)) { return true; } } return false; } /*! * \internal * \brief Get the container of a bundle's first replica * * \param[in] bundle Bundle resource to get container for * * \return Container resource from first replica of \p bundle if any, * otherwise NULL */ pcmk_resource_t * pe__first_container(const pcmk_resource_t *bundle) { const pe__bundle_variant_data_t *bundle_data = NULL; const pcmk__bundle_replica_t *replica = NULL; get_bundle_variant_data(bundle_data, bundle); if (bundle_data->replicas == NULL) { return NULL; } replica = bundle_data->replicas->data; return replica->container; } /*! * \internal * \brief Iterate over bundle replicas * * \param[in,out] bundle Bundle to iterate over * \param[in] fn Function to call for each replica (its return value * indicates whether to continue iterating) * \param[in,out] user_data Pointer to pass to \p fn */ void pe__foreach_bundle_replica(pcmk_resource_t *bundle, bool (*fn)(pcmk__bundle_replica_t *, void *), void *user_data) { const pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, bundle); for (GList *iter = bundle_data->replicas; iter != NULL; iter = iter->next) { if (!fn((pcmk__bundle_replica_t *) iter->data, user_data)) { break; } } } /*! * \internal * \brief Iterate over const bundle replicas * * \param[in] bundle Bundle to iterate over * \param[in] fn Function to call for each replica (its return value * indicates whether to continue iterating) * \param[in,out] user_data Pointer to pass to \p fn */ void pe__foreach_const_bundle_replica(const pcmk_resource_t *bundle, bool (*fn)(const pcmk__bundle_replica_t *, void *), void *user_data) { const pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, bundle); for (const GList *iter = bundle_data->replicas; iter != NULL; iter = iter->next) { if (!fn((const pcmk__bundle_replica_t *) iter->data, user_data)) { break; } } } static char * next_ip(const char *last_ip) { unsigned int oct1 = 0; unsigned int oct2 = 0; unsigned int oct3 = 0; unsigned int oct4 = 0; int rc = sscanf(last_ip, "%u.%u.%u.%u", &oct1, &oct2, &oct3, &oct4); if (rc != 4) { /*@ TODO check for IPv6 */ return NULL; } else if (oct3 > 253) { return NULL; } else if (oct4 > 253) { ++oct3; oct4 = 1; } else { ++oct4; } return crm_strdup_printf("%u.%u.%u.%u", oct1, oct2, oct3, oct4); } static void allocate_ip(pe__bundle_variant_data_t *data, pcmk__bundle_replica_t *replica, GString *buffer) { if(data->ip_range_start == NULL) { return; } else if(data->ip_last) { replica->ipaddr = next_ip(data->ip_last); } else { replica->ipaddr = strdup(data->ip_range_start); } data->ip_last = replica->ipaddr; switch (data->agent_type) { case PE__CONTAINER_AGENT_DOCKER: case PE__CONTAINER_AGENT_PODMAN: if (data->add_host) { g_string_append_printf(buffer, " --add-host=%s-%d:%s", data->prefix, replica->offset, replica->ipaddr); } else { g_string_append_printf(buffer, " --hosts-entry=%s=%s-%d", replica->ipaddr, data->prefix, replica->offset); } break; case PE__CONTAINER_AGENT_RKT: g_string_append_printf(buffer, " --hosts-entry=%s=%s-%d", replica->ipaddr, data->prefix, replica->offset); break; default: // PE__CONTAINER_AGENT_UNKNOWN break; } } static xmlNode * create_resource(const char *name, const char *provider, const char *kind) { xmlNode *rsc = create_xml_node(NULL, XML_CIB_TAG_RESOURCE); crm_xml_add(rsc, PCMK_XA_ID, name); crm_xml_add(rsc, XML_AGENT_ATTR_CLASS, PCMK_RESOURCE_CLASS_OCF); crm_xml_add(rsc, XML_AGENT_ATTR_PROVIDER, provider); crm_xml_add(rsc, XML_ATTR_TYPE, kind); return rsc; } /*! * \internal * \brief Check whether cluster can manage resource inside container * * \param[in,out] data Container variant data * * \return TRUE if networking configuration is acceptable, FALSE otherwise * * \note The resource is manageable if an IP range or control port has been * specified. If a control port is used without an IP range, replicas per * host must be 1. */ static bool valid_network(pe__bundle_variant_data_t *data) { if(data->ip_range_start) { return TRUE; } if(data->control_port) { if(data->nreplicas_per_host > 1) { pcmk__config_err("Specifying the 'control-port' for %s requires " "'replicas-per-host=1'", data->prefix); data->nreplicas_per_host = 1; // @TODO to be sure: // pe__clear_resource_flags(rsc, pcmk_rsc_unique); } return TRUE; } return FALSE; } static int create_ip_resource(pcmk_resource_t *parent, pe__bundle_variant_data_t *data, pcmk__bundle_replica_t *replica) { if(data->ip_range_start) { char *id = NULL; xmlNode *xml_ip = NULL; xmlNode *xml_obj = NULL; id = crm_strdup_printf("%s-ip-%s", data->prefix, replica->ipaddr); crm_xml_sanitize_id(id); xml_ip = create_resource(id, "heartbeat", "IPaddr2"); free(id); xml_obj = create_xml_node(xml_ip, XML_TAG_ATTR_SETS); crm_xml_set_id(xml_obj, "%s-attributes-%d", data->prefix, replica->offset); crm_create_nvpair_xml(xml_obj, NULL, "ip", replica->ipaddr); if(data->host_network) { crm_create_nvpair_xml(xml_obj, NULL, "nic", data->host_network); } if(data->host_netmask) { crm_create_nvpair_xml(xml_obj, NULL, "cidr_netmask", data->host_netmask); } else { crm_create_nvpair_xml(xml_obj, NULL, "cidr_netmask", "32"); } xml_obj = create_xml_node(xml_ip, "operations"); crm_create_op_xml(xml_obj, ID(xml_ip), PCMK_ACTION_MONITOR, "60s", NULL); // TODO: Other ops? Timeouts and intervals from underlying resource? if (pe__unpack_resource(xml_ip, &replica->ip, parent, parent->cluster) != pcmk_rc_ok) { return pcmk_rc_unpack_error; } parent->children = g_list_append(parent->children, replica->ip); } return pcmk_rc_ok; } static const char* container_agent_str(enum pe__container_agent t) { switch (t) { case PE__CONTAINER_AGENT_DOCKER: return PE__CONTAINER_AGENT_DOCKER_S; case PE__CONTAINER_AGENT_RKT: return PE__CONTAINER_AGENT_RKT_S; case PE__CONTAINER_AGENT_PODMAN: return PE__CONTAINER_AGENT_PODMAN_S; default: // PE__CONTAINER_AGENT_UNKNOWN break; } return PE__CONTAINER_AGENT_UNKNOWN_S; } static int create_container_resource(pcmk_resource_t *parent, const pe__bundle_variant_data_t *data, pcmk__bundle_replica_t *replica) { char *id = NULL; xmlNode *xml_container = NULL; xmlNode *xml_obj = NULL; // Agent-specific const char *hostname_opt = NULL; const char *env_opt = NULL; const char *agent_str = NULL; int volid = 0; // rkt-only GString *buffer = NULL; GString *dbuffer = NULL; // Where syntax differences are drop-in replacements, set them now switch (data->agent_type) { case PE__CONTAINER_AGENT_DOCKER: case PE__CONTAINER_AGENT_PODMAN: hostname_opt = "-h "; env_opt = "-e "; break; case PE__CONTAINER_AGENT_RKT: hostname_opt = "--hostname="; env_opt = "--environment="; break; default: // PE__CONTAINER_AGENT_UNKNOWN return pcmk_rc_unpack_error; } agent_str = container_agent_str(data->agent_type); buffer = g_string_sized_new(4096); id = crm_strdup_printf("%s-%s-%d", data->prefix, agent_str, replica->offset); crm_xml_sanitize_id(id); xml_container = create_resource(id, "heartbeat", agent_str); free(id); xml_obj = create_xml_node(xml_container, XML_TAG_ATTR_SETS); crm_xml_set_id(xml_obj, "%s-attributes-%d", data->prefix, replica->offset); crm_create_nvpair_xml(xml_obj, NULL, "image", data->image); crm_create_nvpair_xml(xml_obj, NULL, "allow_pull", XML_BOOLEAN_TRUE); crm_create_nvpair_xml(xml_obj, NULL, "force_kill", XML_BOOLEAN_FALSE); crm_create_nvpair_xml(xml_obj, NULL, "reuse", XML_BOOLEAN_FALSE); if (data->agent_type == PE__CONTAINER_AGENT_DOCKER) { g_string_append(buffer, " --restart=no"); } /* Set a container hostname only if we have an IP to map it to. The user can * set -h or --uts=host themselves if they want a nicer name for logs, but * this makes applications happy who need their hostname to match the IP * they bind to. */ if (data->ip_range_start != NULL) { g_string_append_printf(buffer, " %s%s-%d", hostname_opt, data->prefix, replica->offset); } pcmk__g_strcat(buffer, " ", env_opt, "PCMK_stderr=1", NULL); if (data->container_network != NULL) { pcmk__g_strcat(buffer, " --net=", data->container_network, NULL); } if (data->control_port != NULL) { pcmk__g_strcat(buffer, " ", env_opt, "PCMK_" PCMK__ENV_REMOTE_PORT "=", data->control_port, NULL); } else { g_string_append_printf(buffer, " %sPCMK_" PCMK__ENV_REMOTE_PORT "=%d", env_opt, DEFAULT_REMOTE_PORT); } for (GList *iter = data->mounts; iter != NULL; iter = iter->next) { pe__bundle_mount_t *mount = (pe__bundle_mount_t *) iter->data; char *source = NULL; if (pcmk_is_set(mount->flags, pe__bundle_mount_subdir)) { source = crm_strdup_printf("%s/%s-%d", mount->source, data->prefix, replica->offset); pcmk__add_separated_word(&dbuffer, 1024, source, ","); } switch (data->agent_type) { case PE__CONTAINER_AGENT_DOCKER: case PE__CONTAINER_AGENT_PODMAN: pcmk__g_strcat(buffer, " -v ", pcmk__s(source, mount->source), ":", mount->target, NULL); if (mount->options != NULL) { pcmk__g_strcat(buffer, ":", mount->options, NULL); } break; case PE__CONTAINER_AGENT_RKT: g_string_append_printf(buffer, " --volume vol%d,kind=host," "source=%s%s%s " "--mount volume=vol%d,target=%s", volid, pcmk__s(source, mount->source), (mount->options != NULL)? "," : "", pcmk__s(mount->options, ""), volid, mount->target); volid++; break; default: break; } free(source); } for (GList *iter = data->ports; iter != NULL; iter = iter->next) { pe__bundle_port_t *port = (pe__bundle_port_t *) iter->data; switch (data->agent_type) { case PE__CONTAINER_AGENT_DOCKER: case PE__CONTAINER_AGENT_PODMAN: if (replica->ipaddr != NULL) { pcmk__g_strcat(buffer, " -p ", replica->ipaddr, ":", port->source, ":", port->target, NULL); } else if (!pcmk__str_eq(data->container_network, "host", pcmk__str_none)) { // No need to do port mapping if net == host pcmk__g_strcat(buffer, " -p ", port->source, ":", port->target, NULL); } break; case PE__CONTAINER_AGENT_RKT: if (replica->ipaddr != NULL) { pcmk__g_strcat(buffer, " --port=", port->target, ":", replica->ipaddr, ":", port->source, NULL); } else { pcmk__g_strcat(buffer, " --port=", port->target, ":", port->source, NULL); } break; default: break; } } /* @COMPAT: We should use pcmk__add_word() here, but we can't yet, because * it would cause restarts during rolling upgrades. * * In a previous version of the container resource creation logic, if * data->launcher_options is not NULL, we append * (" %s", data->launcher_options) even if data->launcher_options is an * empty string. Likewise for data->container_host_options. Using * * pcmk__add_word(buffer, 0, data->launcher_options) * * removes that extra trailing space, causing a resource definition change. */ if (data->launcher_options != NULL) { pcmk__g_strcat(buffer, " ", data->launcher_options, NULL); } if (data->container_host_options != NULL) { pcmk__g_strcat(buffer, " ", data->container_host_options, NULL); } crm_create_nvpair_xml(xml_obj, NULL, "run_opts", (const char *) buffer->str); g_string_free(buffer, TRUE); crm_create_nvpair_xml(xml_obj, NULL, "mount_points", (dbuffer != NULL)? (const char *) dbuffer->str : ""); if (dbuffer != NULL) { g_string_free(dbuffer, TRUE); } if (replica->child != NULL) { if (data->container_command != NULL) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", data->container_command); } else { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", SBIN_DIR "/pacemaker-remoted"); } /* TODO: Allow users to specify their own? * * We just want to know if the container is alive; we'll monitor the * child independently. */ crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", "/bin/true"); #if 0 /* @TODO Consider supporting the use case where we can start and stop * resources, but not proxy local commands (such as setting node * attributes), by running the local executor in stand-alone mode. * However, this would probably be better done via ACLs as with other * Pacemaker Remote nodes. */ } else if ((child != NULL) && data->untrusted) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", CRM_DAEMON_DIR "/pacemaker-execd"); crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", CRM_DAEMON_DIR "/pacemaker/cts-exec-helper -c poke"); #endif } else { if (data->container_command != NULL) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", data->container_command); } /* TODO: Allow users to specify their own? * * We don't know what's in the container, so we just want to know if it * is alive. */ crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", "/bin/true"); } xml_obj = create_xml_node(xml_container, "operations"); crm_create_op_xml(xml_obj, ID(xml_container), PCMK_ACTION_MONITOR, "60s", NULL); // TODO: Other ops? Timeouts and intervals from underlying resource? if (pe__unpack_resource(xml_container, &replica->container, parent, parent->cluster) != pcmk_rc_ok) { return pcmk_rc_unpack_error; } pe__set_resource_flags(replica->container, pcmk_rsc_replica_container); parent->children = g_list_append(parent->children, replica->container); return pcmk_rc_ok; } /*! * \brief Ban a node from a resource's (and its children's) allowed nodes list * * \param[in,out] rsc Resource to modify * \param[in] uname Name of node to ban */ static void disallow_node(pcmk_resource_t *rsc, const char *uname) { gpointer match = g_hash_table_lookup(rsc->allowed_nodes, uname); if (match) { ((pcmk_node_t *) match)->weight = -INFINITY; ((pcmk_node_t *) match)->rsc_discover_mode = pcmk_probe_never; } if (rsc->children) { g_list_foreach(rsc->children, (GFunc) disallow_node, (gpointer) uname); } } static int create_remote_resource(pcmk_resource_t *parent, pe__bundle_variant_data_t *data, pcmk__bundle_replica_t *replica) { if (replica->child && valid_network(data)) { GHashTableIter gIter; pcmk_node_t *node = NULL; xmlNode *xml_remote = NULL; char *id = crm_strdup_printf("%s-%d", data->prefix, replica->offset); char *port_s = NULL; const char *uname = NULL; const char *connect_name = NULL; if (pe_find_resource(parent->cluster->resources, id) != NULL) { free(id); // The biggest hammer we have id = crm_strdup_printf("pcmk-internal-%s-remote-%d", replica->child->id, replica->offset); //@TODO return error instead of asserting? CRM_ASSERT(pe_find_resource(parent->cluster->resources, id) == NULL); } /* REMOTE_CONTAINER_HACK: Using "#uname" as the server name when the * connection does not have its own IP is a magic string that we use to * support nested remotes (i.e. a bundle running on a remote node). */ connect_name = (replica->ipaddr? replica->ipaddr : "#uname"); if (data->control_port == NULL) { port_s = pcmk__itoa(DEFAULT_REMOTE_PORT); } /* This sets replica->container as replica->remote's container, which is * similar to what happens with guest nodes. This is how the scheduler * knows that the bundle node is fenced by recovering the container, and * that remote should be ordered relative to the container. */ xml_remote = pe_create_remote_xml(NULL, id, replica->container->id, NULL, NULL, NULL, connect_name, (data->control_port? data->control_port : port_s)); free(port_s); /* Abandon our created ID, and pull the copy from the XML, because we * need something that will get freed during scheduler data cleanup to * use as the node ID and uname. */ free(id); id = NULL; uname = ID(xml_remote); /* Ensure a node has been created for the guest (it may have already * been, if it has a permanent node attribute), and ensure its weight is * -INFINITY so no other resources can run on it. */ node = pe_find_node(parent->cluster->nodes, uname); if (node == NULL) { node = pe_create_node(uname, uname, "remote", "-INFINITY", parent->cluster); } else { node->weight = -INFINITY; } node->rsc_discover_mode = pcmk_probe_never; /* unpack_remote_nodes() ensures that each remote node and guest node * has a pcmk_node_t entry. Ideally, it would do the same for bundle * nodes. Unfortunately, a bundle has to be mostly unpacked before it's * obvious what nodes will be needed, so we do it just above. * * Worse, that means that the node may have been utilized while * unpacking other resources, without our weight correction. The most * likely place for this to happen is when pe__unpack_resource() calls * resource_location() to set a default score in symmetric clusters. * This adds a node *copy* to each resource's allowed nodes, and these * copies will have the wrong weight. * * As a hacky workaround, fix those copies here. * * @TODO Possible alternative: ensure bundles are unpacked before other * resources, so the weight is correct before any copies are made. */ g_list_foreach(parent->cluster->resources, (GFunc) disallow_node, (gpointer) uname); replica->node = pe__copy_node(node); replica->node->weight = 500; replica->node->rsc_discover_mode = pcmk_probe_exclusive; /* Ensure the node shows up as allowed and with the correct discovery set */ if (replica->child->allowed_nodes != NULL) { g_hash_table_destroy(replica->child->allowed_nodes); } replica->child->allowed_nodes = pcmk__strkey_table(NULL, free); g_hash_table_insert(replica->child->allowed_nodes, (gpointer) replica->node->details->id, pe__copy_node(replica->node)); { pcmk_node_t *copy = pe__copy_node(replica->node); copy->weight = -INFINITY; g_hash_table_insert(replica->child->parent->allowed_nodes, (gpointer) replica->node->details->id, copy); } if (pe__unpack_resource(xml_remote, &replica->remote, parent, parent->cluster) != pcmk_rc_ok) { return pcmk_rc_unpack_error; } g_hash_table_iter_init(&gIter, replica->remote->allowed_nodes); while (g_hash_table_iter_next(&gIter, NULL, (void **)&node)) { if (pe__is_guest_or_remote_node(node)) { /* Remote resources can only run on 'normal' cluster node */ node->weight = -INFINITY; } } replica->node->details->remote_rsc = replica->remote; // Ensure pe__is_guest_node() functions correctly immediately replica->remote->container = replica->container; /* A bundle's #kind is closer to "container" (guest node) than the * "remote" set by pe_create_node(). */ g_hash_table_insert(replica->node->details->attrs, strdup(CRM_ATTR_KIND), strdup("container")); /* One effect of this is that setup_container() will add * replica->remote to replica->container's fillers, which will make * pe__resource_contains_guest_node() true for replica->container. * * replica->child does NOT get added to replica->container's fillers. * The only noticeable effect if it did would be for its fail count to * be taken into account when checking replica->container's migration * threshold. */ parent->children = g_list_append(parent->children, replica->remote); } return pcmk_rc_ok; } static int create_replica_resources(pcmk_resource_t *parent, pe__bundle_variant_data_t *data, pcmk__bundle_replica_t *replica) { int rc = pcmk_rc_ok; rc = create_container_resource(parent, data, replica); if (rc != pcmk_rc_ok) { return rc; } rc = create_ip_resource(parent, data, replica); if (rc != pcmk_rc_ok) { return rc; } rc = create_remote_resource(parent, data, replica); if (rc != pcmk_rc_ok) { return rc; } if ((replica->child != NULL) && (replica->ipaddr != NULL)) { add_hash_param(replica->child->meta, "external-ip", replica->ipaddr); } if (replica->remote != NULL) { /* * Allow the remote connection resource to be allocated to a * different node than the one on which the container is active. * * This makes it possible to have Pacemaker Remote nodes running * containers with pacemaker-remoted inside in order to start * services inside those containers. */ pe__set_resource_flags(replica->remote, pcmk_rsc_remote_nesting_allowed); } return rc; } static void mount_add(pe__bundle_variant_data_t *bundle_data, const char *source, const char *target, const char *options, uint32_t flags) { pe__bundle_mount_t *mount = calloc(1, sizeof(pe__bundle_mount_t)); CRM_ASSERT(mount != NULL); mount->source = strdup(source); mount->target = strdup(target); pcmk__str_update(&mount->options, options); mount->flags = flags; bundle_data->mounts = g_list_append(bundle_data->mounts, mount); } static void mount_free(pe__bundle_mount_t *mount) { free(mount->source); free(mount->target); free(mount->options); free(mount); } static void port_free(pe__bundle_port_t *port) { free(port->source); free(port->target); free(port); } static pcmk__bundle_replica_t * replica_for_remote(pcmk_resource_t *remote) { pcmk_resource_t *top = remote; pe__bundle_variant_data_t *bundle_data = NULL; if (top == NULL) { return NULL; } while (top->parent != NULL) { top = top->parent; } get_bundle_variant_data(bundle_data, top); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; if (replica->remote == remote) { return replica; } } CRM_LOG_ASSERT(FALSE); return NULL; } bool pe__bundle_needs_remote_name(pcmk_resource_t *rsc) { const char *value; GHashTable *params = NULL; if (rsc == NULL) { return false; } // Use NULL node since pcmk__bundle_expand() uses that to set value params = pe_rsc_params(rsc, NULL, rsc->cluster); - value = g_hash_table_lookup(params, XML_RSC_ATTR_REMOTE_RA_ADDR); + value = g_hash_table_lookup(params, PCMK_REMOTE_RA_ADDR); return pcmk__str_eq(value, "#uname", pcmk__str_casei) && xml_contains_remote_node(rsc->xml); } const char * pe__add_bundle_remote_name(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler, xmlNode *xml, const char *field) { // REMOTE_CONTAINER_HACK: Allow remote nodes that start containers with pacemaker remote inside pcmk_node_t *node = NULL; pcmk__bundle_replica_t *replica = NULL; if (!pe__bundle_needs_remote_name(rsc)) { return NULL; } replica = replica_for_remote(rsc); if (replica == NULL) { return NULL; } node = replica->container->allocated_to; if (node == NULL) { /* If it won't be running anywhere after the * transition, go with where it's running now. */ node = pe__current_node(replica->container); } if(node == NULL) { crm_trace("Cannot determine address for bundle connection %s", rsc->id); return NULL; } crm_trace("Setting address for bundle connection %s to bundle host %s", rsc->id, pe__node_name(node)); if(xml != NULL && field != NULL) { crm_xml_add(xml, field, node->details->uname); } return node->details->uname; } #define pe__set_bundle_mount_flags(mount_xml, flags, flags_to_set) do { \ flags = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, \ "Bundle mount", ID(mount_xml), flags, \ (flags_to_set), #flags_to_set); \ } while (0) gboolean pe__unpack_bundle(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler) { const char *value = NULL; xmlNode *xml_obj = NULL; xmlNode *xml_resource = NULL; pe__bundle_variant_data_t *bundle_data = NULL; bool need_log_mount = TRUE; CRM_ASSERT(rsc != NULL); pcmk__rsc_trace(rsc, "Processing resource %s...", rsc->id); bundle_data = calloc(1, sizeof(pe__bundle_variant_data_t)); rsc->variant_opaque = bundle_data; bundle_data->prefix = strdup(rsc->id); xml_obj = first_named_child(rsc->xml, PE__CONTAINER_AGENT_DOCKER_S); if (xml_obj != NULL) { bundle_data->agent_type = PE__CONTAINER_AGENT_DOCKER; } else { xml_obj = first_named_child(rsc->xml, PE__CONTAINER_AGENT_RKT_S); if (xml_obj != NULL) { bundle_data->agent_type = PE__CONTAINER_AGENT_RKT; } else { xml_obj = first_named_child(rsc->xml, PE__CONTAINER_AGENT_PODMAN_S); if (xml_obj != NULL) { bundle_data->agent_type = PE__CONTAINER_AGENT_PODMAN; } else { return FALSE; } } } // Use 0 for default, minimum, and invalid promoted-max value = crm_element_value(xml_obj, PCMK_META_PROMOTED_MAX); if (value == NULL) { // @COMPAT deprecated since 2.0.0 value = crm_element_value(xml_obj, "masters"); } pcmk__scan_min_int(value, &bundle_data->promoted_max, 0); // Default replicas to promoted-max if it was specified and 1 otherwise value = crm_element_value(xml_obj, "replicas"); if ((value == NULL) && (bundle_data->promoted_max > 0)) { bundle_data->nreplicas = bundle_data->promoted_max; } else { pcmk__scan_min_int(value, &bundle_data->nreplicas, 1); } /* * Communication between containers on the same host via the * floating IPs only works if the container is started with: * --userland-proxy=false --ip-masq=false */ value = crm_element_value(xml_obj, "replicas-per-host"); pcmk__scan_min_int(value, &bundle_data->nreplicas_per_host, 1); if (bundle_data->nreplicas_per_host == 1) { pe__clear_resource_flags(rsc, pcmk_rsc_unique); } bundle_data->container_command = crm_element_value_copy(xml_obj, "run-command"); bundle_data->launcher_options = crm_element_value_copy(xml_obj, "options"); bundle_data->image = crm_element_value_copy(xml_obj, "image"); bundle_data->container_network = crm_element_value_copy(xml_obj, "network"); xml_obj = first_named_child(rsc->xml, "network"); if(xml_obj) { bundle_data->ip_range_start = crm_element_value_copy(xml_obj, "ip-range-start"); bundle_data->host_netmask = crm_element_value_copy(xml_obj, "host-netmask"); bundle_data->host_network = crm_element_value_copy(xml_obj, "host-interface"); bundle_data->control_port = crm_element_value_copy(xml_obj, "control-port"); value = crm_element_value(xml_obj, "add-host"); if (crm_str_to_boolean(value, &bundle_data->add_host) != 1) { bundle_data->add_host = TRUE; } for (xmlNode *xml_child = pcmk__xe_first_child(xml_obj); xml_child != NULL; xml_child = pcmk__xe_next(xml_child)) { pe__bundle_port_t *port = calloc(1, sizeof(pe__bundle_port_t)); port->source = crm_element_value_copy(xml_child, "port"); if(port->source == NULL) { port->source = crm_element_value_copy(xml_child, "range"); } else { port->target = crm_element_value_copy(xml_child, "internal-port"); } if(port->source != NULL && strlen(port->source) > 0) { if(port->target == NULL) { port->target = strdup(port->source); } bundle_data->ports = g_list_append(bundle_data->ports, port); } else { pcmk__config_err("Invalid port directive %s", ID(xml_child)); port_free(port); } } } xml_obj = first_named_child(rsc->xml, "storage"); for (xmlNode *xml_child = pcmk__xe_first_child(xml_obj); xml_child != NULL; xml_child = pcmk__xe_next(xml_child)) { const char *source = crm_element_value(xml_child, "source-dir"); const char *target = crm_element_value(xml_child, "target-dir"); const char *options = crm_element_value(xml_child, "options"); int flags = pe__bundle_mount_none; if (source == NULL) { source = crm_element_value(xml_child, "source-dir-root"); pe__set_bundle_mount_flags(xml_child, flags, pe__bundle_mount_subdir); } if (source && target) { mount_add(bundle_data, source, target, options, flags); if (strcmp(target, "/var/log") == 0) { need_log_mount = FALSE; } } else { pcmk__config_err("Invalid mount directive %s", ID(xml_child)); } } xml_obj = first_named_child(rsc->xml, "primitive"); if (xml_obj && valid_network(bundle_data)) { char *value = NULL; xmlNode *xml_set = NULL; xml_resource = create_xml_node(NULL, XML_CIB_TAG_INCARNATION); /* @COMPAT We no longer use the tag, but we need to keep it as * part of the resource name, so that bundles don't restart in a rolling * upgrade. (It also avoids needing to change regression tests.) */ crm_xml_set_id(xml_resource, "%s-%s", bundle_data->prefix, (bundle_data->promoted_max? "master" : (const char *)xml_resource->name)); xml_set = create_xml_node(xml_resource, XML_TAG_META_SETS); crm_xml_set_id(xml_set, "%s-%s-meta", bundle_data->prefix, xml_resource->name); crm_create_nvpair_xml(xml_set, NULL, PCMK_META_ORDERED, XML_BOOLEAN_TRUE); value = pcmk__itoa(bundle_data->nreplicas); crm_create_nvpair_xml(xml_set, NULL, PCMK_META_CLONE_MAX, value); free(value); value = pcmk__itoa(bundle_data->nreplicas_per_host); crm_create_nvpair_xml(xml_set, NULL, PCMK_META_CLONE_NODE_MAX, value); free(value); crm_create_nvpair_xml(xml_set, NULL, PCMK_META_GLOBALLY_UNIQUE, pcmk__btoa(bundle_data->nreplicas_per_host > 1)); if (bundle_data->promoted_max) { crm_create_nvpair_xml(xml_set, NULL, PCMK_META_PROMOTABLE, XML_BOOLEAN_TRUE); value = pcmk__itoa(bundle_data->promoted_max); crm_create_nvpair_xml(xml_set, NULL, PCMK_META_PROMOTED_MAX, value); free(value); } //crm_xml_add(xml_obj, PCMK_XA_ID, bundle_data->prefix); add_node_copy(xml_resource, xml_obj); } else if(xml_obj) { pcmk__config_err("Cannot control %s inside %s without either " "ip-range-start or control-port", rsc->id, ID(xml_obj)); return FALSE; } if(xml_resource) { int lpc = 0; GList *childIter = NULL; pe__bundle_port_t *port = NULL; GString *buffer = NULL; if (pe__unpack_resource(xml_resource, &(bundle_data->child), rsc, scheduler) != pcmk_rc_ok) { return FALSE; } /* Currently, we always map the default authentication key location * into the same location inside the container. * * Ideally, we would respect the host's PCMK_authkey_location, but: * - it may be different on different nodes; * - the actual connection will do extra checking to make sure the key * file exists and is readable, that we can't do here on the DC * - tools such as crm_resource and crm_simulate may not have the same * environment variables as the cluster, causing operation digests to * differ * * Always using the default location inside the container is fine, * because we control the pacemaker_remote environment, and it avoids * having to pass another environment variable to the container. * * @TODO A better solution may be to have only pacemaker_remote use the * environment variable, and have the cluster nodes use a new * cluster option for key location. This would introduce the limitation * of the location being the same on all cluster nodes, but that's * reasonable. */ mount_add(bundle_data, DEFAULT_REMOTE_KEY_LOCATION, DEFAULT_REMOTE_KEY_LOCATION, NULL, pe__bundle_mount_none); if (need_log_mount) { mount_add(bundle_data, CRM_BUNDLE_DIR, "/var/log", NULL, pe__bundle_mount_subdir); } port = calloc(1, sizeof(pe__bundle_port_t)); if(bundle_data->control_port) { port->source = strdup(bundle_data->control_port); } else { /* If we wanted to respect PCMK_remote_port, we could use * crm_default_remote_port() here and elsewhere in this file instead * of DEFAULT_REMOTE_PORT. * * However, it gains nothing, since we control both the container * environment and the connection resource parameters, and the user * can use a different port if desired by setting control-port. */ port->source = pcmk__itoa(DEFAULT_REMOTE_PORT); } port->target = strdup(port->source); bundle_data->ports = g_list_append(bundle_data->ports, port); buffer = g_string_sized_new(1024); for (childIter = bundle_data->child->children; childIter != NULL; childIter = childIter->next) { pcmk__bundle_replica_t *replica = NULL; replica = calloc(1, sizeof(pcmk__bundle_replica_t)); replica->child = childIter->data; replica->child->exclusive_discover = TRUE; replica->offset = lpc++; // Ensure the child's notify gets set based on the underlying primitive's value if (pcmk_is_set(replica->child->flags, pcmk_rsc_notify)) { pe__set_resource_flags(bundle_data->child, pcmk_rsc_notify); } allocate_ip(bundle_data, replica, buffer); bundle_data->replicas = g_list_append(bundle_data->replicas, replica); bundle_data->attribute_target = g_hash_table_lookup(replica->child->meta, PCMK_META_CONTAINER_ATTR_TARGET); } bundle_data->container_host_options = g_string_free(buffer, FALSE); if (bundle_data->attribute_target) { g_hash_table_replace(rsc->meta, strdup(PCMK_META_CONTAINER_ATTR_TARGET), strdup(bundle_data->attribute_target)); g_hash_table_replace(bundle_data->child->meta, strdup(PCMK_META_CONTAINER_ATTR_TARGET), strdup(bundle_data->attribute_target)); } } else { // Just a naked container, no pacemaker-remote GString *buffer = g_string_sized_new(1024); for (int lpc = 0; lpc < bundle_data->nreplicas; lpc++) { pcmk__bundle_replica_t *replica = NULL; replica = calloc(1, sizeof(pcmk__bundle_replica_t)); replica->offset = lpc; allocate_ip(bundle_data, replica, buffer); bundle_data->replicas = g_list_append(bundle_data->replicas, replica); } bundle_data->container_host_options = g_string_free(buffer, FALSE); } for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; if (create_replica_resources(rsc, bundle_data, replica) != pcmk_rc_ok) { pcmk__config_err("Failed unpacking resource %s", rsc->id); rsc->fns->free(rsc); return FALSE; } /* Utilization needs special handling for bundles. It makes no sense for * the inner primitive to have utilization, because it is tied * one-to-one to the guest node created by the container resource -- and * there's no way to set capacities for that guest node anyway. * * What the user really wants is to configure utilization for the * container. However, the schema only allows utilization for * primitives, and the container resource is implicit anyway, so the * user can *only* configure utilization for the inner primitive. If * they do, move the primitive's utilization values to the container. * * @TODO This means that bundles without an inner primitive can't have * utilization. An alternative might be to allow utilization values in * the top-level bundle XML in the schema, and copy those to each * container. */ if (replica->child != NULL) { GHashTable *empty = replica->container->utilization; replica->container->utilization = replica->child->utilization; replica->child->utilization = empty; } } if (bundle_data->child) { rsc->children = g_list_append(rsc->children, bundle_data->child); } return TRUE; } static int replica_resource_active(pcmk_resource_t *rsc, gboolean all) { if (rsc) { gboolean child_active = rsc->fns->active(rsc, all); if (child_active && !all) { return TRUE; } else if (!child_active && all) { return FALSE; } } return -1; } gboolean pe__bundle_active(pcmk_resource_t *rsc, gboolean all) { pe__bundle_variant_data_t *bundle_data = NULL; GList *iter = NULL; get_bundle_variant_data(bundle_data, rsc); for (iter = bundle_data->replicas; iter != NULL; iter = iter->next) { pcmk__bundle_replica_t *replica = iter->data; int rsc_active; rsc_active = replica_resource_active(replica->ip, all); if (rsc_active >= 0) { return (gboolean) rsc_active; } rsc_active = replica_resource_active(replica->child, all); if (rsc_active >= 0) { return (gboolean) rsc_active; } rsc_active = replica_resource_active(replica->container, all); if (rsc_active >= 0) { return (gboolean) rsc_active; } rsc_active = replica_resource_active(replica->remote, all); if (rsc_active >= 0) { return (gboolean) rsc_active; } } /* If "all" is TRUE, we've already checked that no resources were inactive, * so return TRUE; if "all" is FALSE, we didn't find any active resources, * so return FALSE. */ return all; } /*! * \internal * \brief Find the bundle replica corresponding to a given node * * \param[in] bundle Top-level bundle resource * \param[in] node Node to search for * * \return Bundle replica if found, NULL otherwise */ pcmk_resource_t * pe__find_bundle_replica(const pcmk_resource_t *bundle, const pcmk_node_t *node) { pe__bundle_variant_data_t *bundle_data = NULL; CRM_ASSERT(bundle && node); get_bundle_variant_data(bundle_data, bundle); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica && replica->node); if (replica->node->details == node->details) { return replica->child; } } return NULL; } /*! * \internal * \deprecated This function will be removed in a future release */ static void print_rsc_in_list(pcmk_resource_t *rsc, const char *pre_text, long options, void *print_data) { if (rsc != NULL) { if (options & pe_print_html) { status_print("
  • "); } rsc->fns->print(rsc, pre_text, options, print_data); if (options & pe_print_html) { status_print("
    • \n"); } } } /*! * \internal * \deprecated This function will be removed in a future release */ static void bundle_print_xml(pcmk_resource_t *rsc, const char *pre_text, long options, void *print_data) { pe__bundle_variant_data_t *bundle_data = NULL; char *child_text = NULL; CRM_CHECK(rsc != NULL, return); if (pre_text == NULL) { pre_text = ""; } child_text = crm_strdup_printf("%s ", pre_text); get_bundle_variant_data(bundle_data, rsc); status_print("%sid); status_print("type=\"%s\" ", container_agent_str(bundle_data->agent_type)); status_print("image=\"%s\" ", bundle_data->image); status_print("unique=\"%s\" ", pe__rsc_bool_str(rsc, pcmk_rsc_unique)); status_print("managed=\"%s\" ", pe__rsc_bool_str(rsc, pcmk_rsc_managed)); status_print("failed=\"%s\" ", pe__rsc_bool_str(rsc, pcmk_rsc_failed)); status_print(">\n"); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica); status_print("%s \n", pre_text, replica->offset); print_rsc_in_list(replica->ip, child_text, options, print_data); print_rsc_in_list(replica->child, child_text, options, print_data); print_rsc_in_list(replica->container, child_text, options, print_data); print_rsc_in_list(replica->remote, child_text, options, print_data); status_print("%s \n", pre_text); } status_print("%s\n", pre_text); free(child_text); } PCMK__OUTPUT_ARGS("bundle", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__bundle_xml(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); pe__bundle_variant_data_t *bundle_data = NULL; int rc = pcmk_rc_no_output; gboolean printed_header = FALSE; gboolean print_everything = TRUE; const char *desc = NULL; CRM_ASSERT(rsc != NULL); get_bundle_variant_data(bundle_data, rsc); if (rsc->fns->is_filtered(rsc, only_rsc, TRUE)) { return rc; } print_everything = pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; char *id = NULL; gboolean print_ip, print_child, print_ctnr, print_remote; CRM_ASSERT(replica); if (pcmk__rsc_filtered_by_node(replica->container, only_node)) { continue; } print_ip = replica->ip != NULL && !replica->ip->fns->is_filtered(replica->ip, only_rsc, print_everything); print_child = replica->child != NULL && !replica->child->fns->is_filtered(replica->child, only_rsc, print_everything); print_ctnr = !replica->container->fns->is_filtered(replica->container, only_rsc, print_everything); print_remote = replica->remote != NULL && !replica->remote->fns->is_filtered(replica->remote, only_rsc, print_everything); if (!print_everything && !print_ip && !print_child && !print_ctnr && !print_remote) { continue; } if (!printed_header) { printed_header = TRUE; desc = pe__resource_description(rsc, show_opts); rc = pe__name_and_nvpairs_xml(out, true, "bundle", 8, PCMK_XA_ID, rsc->id, "type", container_agent_str(bundle_data->agent_type), "image", bundle_data->image, "unique", pe__rsc_bool_str(rsc, pcmk_rsc_unique), "maintenance", pe__rsc_bool_str(rsc, pcmk_rsc_maintenance), "managed", pe__rsc_bool_str(rsc, pcmk_rsc_managed), "failed", pe__rsc_bool_str(rsc, pcmk_rsc_failed), PCMK_XA_DESCRIPTION, desc); CRM_ASSERT(rc == pcmk_rc_ok); } id = pcmk__itoa(replica->offset); rc = pe__name_and_nvpairs_xml(out, true, "replica", 1, PCMK_XA_ID, id); free(id); CRM_ASSERT(rc == pcmk_rc_ok); if (print_ip) { out->message(out, crm_map_element_name(replica->ip->xml), show_opts, replica->ip, only_node, only_rsc); } if (print_child) { out->message(out, crm_map_element_name(replica->child->xml), show_opts, replica->child, only_node, only_rsc); } if (print_ctnr) { out->message(out, crm_map_element_name(replica->container->xml), show_opts, replica->container, only_node, only_rsc); } if (print_remote) { out->message(out, crm_map_element_name(replica->remote->xml), show_opts, replica->remote, only_node, only_rsc); } pcmk__output_xml_pop_parent(out); // replica } if (printed_header) { pcmk__output_xml_pop_parent(out); // bundle } return rc; } static void pe__bundle_replica_output_html(pcmk__output_t *out, pcmk__bundle_replica_t *replica, pcmk_node_t *node, uint32_t show_opts) { pcmk_resource_t *rsc = replica->child; int offset = 0; char buffer[LINE_MAX]; if(rsc == NULL) { rsc = replica->container; } if (replica->remote) { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->remote)); } else { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->container)); } if (replica->ipaddr) { offset += snprintf(buffer + offset, LINE_MAX - offset, " (%s)", replica->ipaddr); } pe__common_output_html(out, rsc, buffer, node, show_opts); } /*! * \internal * \brief Get a string describing a resource's unmanaged state or lack thereof * * \param[in] rsc Resource to describe * * \return A string indicating that a resource is in maintenance mode or * otherwise unmanaged, or an empty string otherwise */ static const char * get_unmanaged_str(const pcmk_resource_t *rsc) { if (pcmk_is_set(rsc->flags, pcmk_rsc_maintenance)) { return " (maintenance)"; } if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { return " (unmanaged)"; } return ""; } PCMK__OUTPUT_ARGS("bundle", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__bundle_html(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); const char *desc = NULL; pe__bundle_variant_data_t *bundle_data = NULL; int rc = pcmk_rc_no_output; gboolean print_everything = TRUE; CRM_ASSERT(rsc != NULL); get_bundle_variant_data(bundle_data, rsc); desc = pe__resource_description(rsc, show_opts); if (rsc->fns->is_filtered(rsc, only_rsc, TRUE)) { return rc; } print_everything = pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; gboolean print_ip, print_child, print_ctnr, print_remote; CRM_ASSERT(replica); if (pcmk__rsc_filtered_by_node(replica->container, only_node)) { continue; } print_ip = replica->ip != NULL && !replica->ip->fns->is_filtered(replica->ip, only_rsc, print_everything); print_child = replica->child != NULL && !replica->child->fns->is_filtered(replica->child, only_rsc, print_everything); print_ctnr = !replica->container->fns->is_filtered(replica->container, only_rsc, print_everything); print_remote = replica->remote != NULL && !replica->remote->fns->is_filtered(replica->remote, only_rsc, print_everything); if (pcmk_is_set(show_opts, pcmk_show_implicit_rscs) || (print_everything == FALSE && (print_ip || print_child || print_ctnr || print_remote))) { /* The text output messages used below require pe_print_implicit to * be set to do anything. */ uint32_t new_show_opts = show_opts | pcmk_show_implicit_rscs; PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Container bundle%s: %s [%s]%s%s%s%s%s", (bundle_data->nreplicas > 1)? " set" : "", rsc->id, bundle_data->image, pcmk_is_set(rsc->flags, pcmk_rsc_unique)? " (unique)" : "", desc ? " (" : "", desc ? desc : "", desc ? ")" : "", get_unmanaged_str(rsc)); if (pcmk__list_of_multiple(bundle_data->replicas)) { out->begin_list(out, NULL, NULL, "Replica[%d]", replica->offset); } if (print_ip) { out->message(out, crm_map_element_name(replica->ip->xml), new_show_opts, replica->ip, only_node, only_rsc); } if (print_child) { out->message(out, crm_map_element_name(replica->child->xml), new_show_opts, replica->child, only_node, only_rsc); } if (print_ctnr) { out->message(out, crm_map_element_name(replica->container->xml), new_show_opts, replica->container, only_node, only_rsc); } if (print_remote) { out->message(out, crm_map_element_name(replica->remote->xml), new_show_opts, replica->remote, only_node, only_rsc); } if (pcmk__list_of_multiple(bundle_data->replicas)) { out->end_list(out); } } else if (print_everything == FALSE && !(print_ip || print_child || print_ctnr || print_remote)) { continue; } else { PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Container bundle%s: %s [%s]%s%s%s%s%s", (bundle_data->nreplicas > 1)? " set" : "", rsc->id, bundle_data->image, pcmk_is_set(rsc->flags, pcmk_rsc_unique)? " (unique)" : "", desc ? " (" : "", desc ? desc : "", desc ? ")" : "", get_unmanaged_str(rsc)); pe__bundle_replica_output_html(out, replica, pe__current_node(replica->container), show_opts); } } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } static void pe__bundle_replica_output_text(pcmk__output_t *out, pcmk__bundle_replica_t *replica, pcmk_node_t *node, uint32_t show_opts) { const pcmk_resource_t *rsc = replica->child; int offset = 0; char buffer[LINE_MAX]; if(rsc == NULL) { rsc = replica->container; } if (replica->remote) { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->remote)); } else { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->container)); } if (replica->ipaddr) { offset += snprintf(buffer + offset, LINE_MAX - offset, " (%s)", replica->ipaddr); } pe__common_output_text(out, rsc, buffer, node, show_opts); } PCMK__OUTPUT_ARGS("bundle", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__bundle_text(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); const char *desc = NULL; pe__bundle_variant_data_t *bundle_data = NULL; int rc = pcmk_rc_no_output; gboolean print_everything = TRUE; desc = pe__resource_description(rsc, show_opts); get_bundle_variant_data(bundle_data, rsc); CRM_ASSERT(rsc != NULL); if (rsc->fns->is_filtered(rsc, only_rsc, TRUE)) { return rc; } print_everything = pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; gboolean print_ip, print_child, print_ctnr, print_remote; CRM_ASSERT(replica); if (pcmk__rsc_filtered_by_node(replica->container, only_node)) { continue; } print_ip = replica->ip != NULL && !replica->ip->fns->is_filtered(replica->ip, only_rsc, print_everything); print_child = replica->child != NULL && !replica->child->fns->is_filtered(replica->child, only_rsc, print_everything); print_ctnr = !replica->container->fns->is_filtered(replica->container, only_rsc, print_everything); print_remote = replica->remote != NULL && !replica->remote->fns->is_filtered(replica->remote, only_rsc, print_everything); if (pcmk_is_set(show_opts, pcmk_show_implicit_rscs) || (print_everything == FALSE && (print_ip || print_child || print_ctnr || print_remote))) { /* The text output messages used below require pe_print_implicit to * be set to do anything. */ uint32_t new_show_opts = show_opts | pcmk_show_implicit_rscs; PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Container bundle%s: %s [%s]%s%s%s%s%s", (bundle_data->nreplicas > 1)? " set" : "", rsc->id, bundle_data->image, pcmk_is_set(rsc->flags, pcmk_rsc_unique)? " (unique)" : "", desc ? " (" : "", desc ? desc : "", desc ? ")" : "", get_unmanaged_str(rsc)); if (pcmk__list_of_multiple(bundle_data->replicas)) { out->list_item(out, NULL, "Replica[%d]", replica->offset); } out->begin_list(out, NULL, NULL, NULL); if (print_ip) { out->message(out, crm_map_element_name(replica->ip->xml), new_show_opts, replica->ip, only_node, only_rsc); } if (print_child) { out->message(out, crm_map_element_name(replica->child->xml), new_show_opts, replica->child, only_node, only_rsc); } if (print_ctnr) { out->message(out, crm_map_element_name(replica->container->xml), new_show_opts, replica->container, only_node, only_rsc); } if (print_remote) { out->message(out, crm_map_element_name(replica->remote->xml), new_show_opts, replica->remote, only_node, only_rsc); } out->end_list(out); } else if (print_everything == FALSE && !(print_ip || print_child || print_ctnr || print_remote)) { continue; } else { PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Container bundle%s: %s [%s]%s%s%s%s%s", (bundle_data->nreplicas > 1)? " set" : "", rsc->id, bundle_data->image, pcmk_is_set(rsc->flags, pcmk_rsc_unique)? " (unique)" : "", desc ? " (" : "", desc ? desc : "", desc ? ")" : "", get_unmanaged_str(rsc)); pe__bundle_replica_output_text(out, replica, pe__current_node(replica->container), show_opts); } } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } /*! * \internal * \deprecated This function will be removed in a future release */ static void print_bundle_replica(pcmk__bundle_replica_t *replica, const char *pre_text, long options, void *print_data) { pcmk_node_t *node = NULL; pcmk_resource_t *rsc = replica->child; int offset = 0; char buffer[LINE_MAX]; if(rsc == NULL) { rsc = replica->container; } if (replica->remote) { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->remote)); } else { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->container)); } if (replica->ipaddr) { offset += snprintf(buffer + offset, LINE_MAX - offset, " (%s)", replica->ipaddr); } node = pe__current_node(replica->container); common_print(rsc, pre_text, buffer, node, options, print_data); } /*! * \internal * \deprecated This function will be removed in a future release */ void pe__print_bundle(pcmk_resource_t *rsc, const char *pre_text, long options, void *print_data) { pe__bundle_variant_data_t *bundle_data = NULL; char *child_text = NULL; CRM_CHECK(rsc != NULL, return); if (options & pe_print_xml) { bundle_print_xml(rsc, pre_text, options, print_data); return; } get_bundle_variant_data(bundle_data, rsc); if (pre_text == NULL) { pre_text = " "; } status_print("%sContainer bundle%s: %s [%s]%s%s\n", pre_text, ((bundle_data->nreplicas > 1)? " set" : ""), rsc->id, bundle_data->image, pcmk_is_set(rsc->flags, pcmk_rsc_unique)? " (unique)" : "", pcmk_is_set(rsc->flags, pcmk_rsc_managed)? "" : " (unmanaged)"); if (options & pe_print_html) { status_print("
    \n
      \n"); } for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica); if (options & pe_print_html) { status_print("
    • "); } if (pcmk_is_set(options, pe_print_implicit)) { child_text = crm_strdup_printf(" %s", pre_text); if (pcmk__list_of_multiple(bundle_data->replicas)) { status_print(" %sReplica[%d]\n", pre_text, replica->offset); } if (options & pe_print_html) { status_print("
      \n
        \n"); } print_rsc_in_list(replica->ip, child_text, options, print_data); print_rsc_in_list(replica->container, child_text, options, print_data); print_rsc_in_list(replica->remote, child_text, options, print_data); print_rsc_in_list(replica->child, child_text, options, print_data); if (options & pe_print_html) { status_print("
      \n"); } } else { child_text = crm_strdup_printf("%s ", pre_text); print_bundle_replica(replica, child_text, options, print_data); } free(child_text); if (options & pe_print_html) { status_print("
      • \n"); } } if (options & pe_print_html) { status_print("
    \n"); } } static void free_bundle_replica(pcmk__bundle_replica_t *replica) { if (replica == NULL) { return; } if (replica->node) { free(replica->node); replica->node = NULL; } if (replica->ip) { free_xml(replica->ip->xml); replica->ip->xml = NULL; replica->ip->fns->free(replica->ip); replica->ip = NULL; } if (replica->container) { free_xml(replica->container->xml); replica->container->xml = NULL; replica->container->fns->free(replica->container); replica->container = NULL; } if (replica->remote) { free_xml(replica->remote->xml); replica->remote->xml = NULL; replica->remote->fns->free(replica->remote); replica->remote = NULL; } free(replica->ipaddr); free(replica); } void pe__free_bundle(pcmk_resource_t *rsc) { pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(rsc != NULL, return); get_bundle_variant_data(bundle_data, rsc); pcmk__rsc_trace(rsc, "Freeing %s", rsc->id); free(bundle_data->prefix); free(bundle_data->image); free(bundle_data->control_port); free(bundle_data->host_network); free(bundle_data->host_netmask); free(bundle_data->ip_range_start); free(bundle_data->container_network); free(bundle_data->launcher_options); free(bundle_data->container_command); g_free(bundle_data->container_host_options); g_list_free_full(bundle_data->replicas, (GDestroyNotify) free_bundle_replica); g_list_free_full(bundle_data->mounts, (GDestroyNotify)mount_free); g_list_free_full(bundle_data->ports, (GDestroyNotify)port_free); g_list_free(rsc->children); if(bundle_data->child) { free_xml(bundle_data->child->xml); bundle_data->child->xml = NULL; bundle_data->child->fns->free(bundle_data->child); } common_free(rsc); } enum rsc_role_e pe__bundle_resource_state(const pcmk_resource_t *rsc, gboolean current) { enum rsc_role_e container_role = pcmk_role_unknown; return container_role; } /*! * \brief Get the number of configured replicas in a bundle * * \param[in] rsc Bundle resource * * \return Number of configured replicas, or 0 on error */ int pe_bundle_replicas(const pcmk_resource_t *rsc) { if ((rsc == NULL) || (rsc->variant != pcmk_rsc_variant_bundle)) { return 0; } else { pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, rsc); return bundle_data->nreplicas; } } void pe__count_bundle(pcmk_resource_t *rsc) { pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, rsc); for (GList *item = bundle_data->replicas; item != NULL; item = item->next) { pcmk__bundle_replica_t *replica = item->data; if (replica->ip) { replica->ip->fns->count(replica->ip); } if (replica->child) { replica->child->fns->count(replica->child); } if (replica->container) { replica->container->fns->count(replica->container); } if (replica->remote) { replica->remote->fns->count(replica->remote); } } } gboolean pe__bundle_is_filtered(const pcmk_resource_t *rsc, GList *only_rsc, gboolean check_parent) { gboolean passes = FALSE; pe__bundle_variant_data_t *bundle_data = NULL; if (pcmk__str_in_list(rsc_printable_id(rsc), only_rsc, pcmk__str_star_matches)) { passes = TRUE; } else { get_bundle_variant_data(bundle_data, rsc); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; if (replica->ip != NULL && !replica->ip->fns->is_filtered(replica->ip, only_rsc, FALSE)) { passes = TRUE; break; } else if (replica->child != NULL && !replica->child->fns->is_filtered(replica->child, only_rsc, FALSE)) { passes = TRUE; break; } else if (!replica->container->fns->is_filtered(replica->container, only_rsc, FALSE)) { passes = TRUE; break; } else if (replica->remote != NULL && !replica->remote->fns->is_filtered(replica->remote, only_rsc, FALSE)) { passes = TRUE; break; } } } return !passes; } /*! * \internal * \brief Get a list of a bundle's containers * * \param[in] bundle Bundle resource * * \return Newly created list of \p bundle's containers * \note It is the caller's responsibility to free the result with * g_list_free(). */ GList * pe__bundle_containers(const pcmk_resource_t *bundle) { GList *containers = NULL; const pe__bundle_variant_data_t *data = NULL; get_bundle_variant_data(data, bundle); for (GList *iter = data->replicas; iter != NULL; iter = iter->next) { pcmk__bundle_replica_t *replica = iter->data; containers = g_list_append(containers, replica->container); } return containers; } // Bundle implementation of pcmk_rsc_methods_t:active_node() pcmk_node_t * pe__bundle_active_node(const pcmk_resource_t *rsc, unsigned int *count_all, unsigned int *count_clean) { pcmk_node_t *active = NULL; pcmk_node_t *node = NULL; pcmk_resource_t *container = NULL; GList *containers = NULL; GList *iter = NULL; GHashTable *nodes = NULL; const pe__bundle_variant_data_t *data = NULL; if (count_all != NULL) { *count_all = 0; } if (count_clean != NULL) { *count_clean = 0; } if (rsc == NULL) { return NULL; } /* For the purposes of this method, we only care about where the bundle's * containers are active, so build a list of active containers. */ get_bundle_variant_data(data, rsc); for (iter = data->replicas; iter != NULL; iter = iter->next) { pcmk__bundle_replica_t *replica = iter->data; if (replica->container->running_on != NULL) { containers = g_list_append(containers, replica->container); } } if (containers == NULL) { return NULL; } /* If the bundle has only a single active container, just use that * container's method. If live migration is ever supported for bundle * containers, this will allow us to prefer the migration source when there * is only one container and it is migrating. For now, this just lets us * avoid creating the nodes table. */ if (pcmk__list_of_1(containers)) { container = containers->data; node = container->fns->active_node(container, count_all, count_clean); g_list_free(containers); return node; } // Add all containers' active nodes to a hash table (for uniqueness) nodes = g_hash_table_new(NULL, NULL); for (iter = containers; iter != NULL; iter = iter->next) { container = iter->data; for (GList *node_iter = container->running_on; node_iter != NULL; node_iter = node_iter->next) { node = node_iter->data; // If insert returns true, we haven't counted this node yet if (g_hash_table_insert(nodes, (gpointer) node->details, (gpointer) node) && !pe__count_active_node(rsc, node, &active, count_all, count_clean)) { goto done; } } } done: g_list_free(containers); g_hash_table_destroy(nodes); return active; } /*! * \internal * \brief Get maximum bundle resource instances per node * * \param[in] rsc Bundle resource to check * * \return Maximum number of \p rsc instances that can be active on one node */ unsigned int pe__bundle_max_per_node(const pcmk_resource_t *rsc) { pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, rsc); CRM_ASSERT(bundle_data->nreplicas_per_host >= 0); return (unsigned int) bundle_data->nreplicas_per_host; } diff --git a/lib/pengine/pe_digest.c b/lib/pengine/pe_digest.c index 7f1f365b2a..934b55bcca 100644 --- a/lib/pengine/pe_digest.c +++ b/lib/pengine/pe_digest.c @@ -1,609 +1,609 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include "pe_status_private.h" extern bool pcmk__is_daemon; /*! * \internal * \brief Free an operation digest cache entry * * \param[in,out] ptr Pointer to cache entry to free * * \note The argument is a gpointer so this can be used as a hash table * free function. */ void pe__free_digests(gpointer ptr) { pcmk__op_digest_t *data = ptr; if (data != NULL) { free_xml(data->params_all); free_xml(data->params_secure); free_xml(data->params_restart); free(data->digest_all_calc); free(data->digest_restart_calc); free(data->digest_secure_calc); free(data); } } // Return true if XML attribute name is not substring of a given string static bool attr_not_in_string(xmlAttrPtr a, void *user_data) { bool filter = false; char *name = crm_strdup_printf(" %s ", (const char *) a->name); if (strstr((const char *) user_data, name) == NULL) { crm_trace("Filtering %s (not found in '%s')", (const char *) a->name, (const char *) user_data); filter = true; } free(name); return filter; } // Return true if XML attribute name is substring of a given string static bool attr_in_string(xmlAttrPtr a, void *user_data) { bool filter = false; char *name = crm_strdup_printf(" %s ", (const char *) a->name); if (strstr((const char *) user_data, name) != NULL) { crm_trace("Filtering %s (found in '%s')", (const char *) a->name, (const char *) user_data); filter = true; } free(name); return filter; } /*! * \internal * \brief Add digest of all parameters to a digest cache entry * * \param[out] data Digest cache entry to modify * \param[in,out] rsc Resource that action was for * \param[in] node Node action was performed on * \param[in] params Resource parameters evaluated for node * \param[in] task Name of action performed * \param[in,out] interval_ms Action's interval (will be reset if in overrides) * \param[in] xml_op Unused * \param[in] op_version CRM feature set to use for digest calculation * \param[in] overrides Key/value table to override resource parameters * \param[in,out] scheduler Scheduler data */ static void calculate_main_digest(pcmk__op_digest_t *data, pcmk_resource_t *rsc, const pcmk_node_t *node, GHashTable *params, const char *task, guint *interval_ms, const xmlNode *xml_op, const char *op_version, GHashTable *overrides, pcmk_scheduler_t *scheduler) { xmlNode *action_config = NULL; data->params_all = create_xml_node(NULL, XML_TAG_PARAMS); /* REMOTE_CONTAINER_HACK: Allow Pacemaker Remote nodes to run containers * that themselves are Pacemaker Remote nodes */ (void) pe__add_bundle_remote_name(rsc, scheduler, data->params_all, - XML_RSC_ATTR_REMOTE_RA_ADDR); + PCMK_REMOTE_RA_ADDR); if (overrides != NULL) { // If interval was overridden, reset it const char *interval_s = g_hash_table_lookup(overrides, CRM_META "_" XML_LRM_ATTR_INTERVAL); if (interval_s != NULL) { long long value_ll; if ((pcmk__scan_ll(interval_s, &value_ll, 0LL) == pcmk_rc_ok) && (value_ll >= 0) && (value_ll <= G_MAXUINT)) { *interval_ms = (guint) value_ll; } } // Add overrides to list of all parameters g_hash_table_foreach(overrides, hash2field, data->params_all); } // Add provided instance parameters g_hash_table_foreach(params, hash2field, data->params_all); // Find action configuration XML in CIB action_config = pcmk__find_action_config(rsc, task, *interval_ms, true); /* Add action-specific resource instance attributes to the digest list. * * If this is a one-time action with action-specific instance attributes, * enforce a restart instead of reload-agent in case the main digest doesn't * match, even if the restart digest does. This ensures any changes of the * action-specific parameters get applied for this specific action, and * digests calculated for the resulting history will be correct. Default the * result to RSC_DIGEST_RESTART for the case where the main digest doesn't * match. */ params = pcmk__unpack_action_rsc_params(action_config, node->details->attrs, scheduler); if ((*interval_ms == 0) && (g_hash_table_size(params) > 0)) { data->rc = pcmk__digest_restart; } g_hash_table_foreach(params, hash2field, data->params_all); g_hash_table_destroy(params); // Add action meta-attributes params = pcmk__unpack_action_meta(rsc, node, task, *interval_ms, action_config); g_hash_table_foreach(params, hash2metafield, data->params_all); g_hash_table_destroy(params); pcmk__filter_op_for_digest(data->params_all); data->digest_all_calc = calculate_operation_digest(data->params_all, op_version); } // Return true if XML attribute name is a Pacemaker-defined fencing parameter static bool is_fence_param(xmlAttrPtr attr, void *user_data) { return pcmk_stonith_param((const char *) attr->name); } /*! * \internal * \brief Add secure digest to a digest cache entry * * \param[out] data Digest cache entry to modify * \param[in] rsc Resource that action was for * \param[in] params Resource parameters evaluated for node * \param[in] xml_op XML of operation in CIB status (if available) * \param[in] op_version CRM feature set to use for digest calculation * \param[in] overrides Key/value hash table to override resource parameters */ static void calculate_secure_digest(pcmk__op_digest_t *data, const pcmk_resource_t *rsc, GHashTable *params, const xmlNode *xml_op, const char *op_version, GHashTable *overrides) { const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS); const char *secure_list = NULL; bool old_version = (compare_version(op_version, "3.16.0") < 0); if (xml_op == NULL) { secure_list = " passwd password user "; } else { secure_list = crm_element_value(xml_op, XML_LRM_ATTR_OP_SECURE); } if (old_version) { data->params_secure = create_xml_node(NULL, XML_TAG_PARAMS); if (overrides != NULL) { g_hash_table_foreach(overrides, hash2field, data->params_secure); } g_hash_table_foreach(params, hash2field, data->params_secure); } else { // Start with a copy of all parameters data->params_secure = copy_xml(data->params_all); } if (secure_list != NULL) { pcmk__xe_remove_matching_attrs(data->params_secure, attr_in_string, (void *) secure_list); } if (old_version && pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_fence_params)) { /* For stonith resources, Pacemaker adds special parameters, * but these are not listed in fence agent meta-data, so with older * versions of DC, the controller will not hash them. That means we have * to filter them out before calculating our hash for comparison. */ pcmk__xe_remove_matching_attrs(data->params_secure, is_fence_param, NULL); } pcmk__filter_op_for_digest(data->params_secure); /* CRM_meta_timeout *should* be part of a digest for recurring operations. * However, with older versions of DC, the controller does not add timeout * to secure digests, because it only includes parameters declared by the * resource agent. * Remove any timeout that made it this far, to match. */ if (old_version) { xml_remove_prop(data->params_secure, CRM_META "_" XML_ATTR_TIMEOUT); } data->digest_secure_calc = calculate_operation_digest(data->params_secure, op_version); } /*! * \internal * \brief Add restart digest to a digest cache entry * * \param[out] data Digest cache entry to modify * \param[in] xml_op XML of operation in CIB status (if available) * \param[in] op_version CRM feature set to use for digest calculation * * \note This function doesn't need to handle overrides because it starts with * data->params_all, which already has overrides applied. */ static void calculate_restart_digest(pcmk__op_digest_t *data, const xmlNode *xml_op, const char *op_version) { const char *value = NULL; // We must have XML of resource operation history if (xml_op == NULL) { return; } // And the history must have a restart digest to compare against if (crm_element_value(xml_op, XML_LRM_ATTR_RESTART_DIGEST) == NULL) { return; } // Start with a copy of all parameters data->params_restart = copy_xml(data->params_all); // Then filter out reloadable parameters, if any value = crm_element_value(xml_op, XML_LRM_ATTR_OP_RESTART); if (value != NULL) { pcmk__xe_remove_matching_attrs(data->params_restart, attr_not_in_string, (void *) value); } value = crm_element_value(xml_op, PCMK_XA_CRM_FEATURE_SET); data->digest_restart_calc = calculate_operation_digest(data->params_restart, value); } /*! * \internal * \brief Create a new digest cache entry with calculated digests * * \param[in,out] rsc Resource that action was for * \param[in] task Name of action performed * \param[in,out] interval_ms Action's interval (will be reset if in overrides) * \param[in] node Node action was performed on * \param[in] xml_op XML of operation in CIB status (if available) * \param[in] overrides Key/value table to override resource parameters * \param[in] calc_secure Whether to calculate secure digest * \param[in,out] scheduler Scheduler data * * \return Pointer to new digest cache entry (or NULL on memory error) * \note It is the caller's responsibility to free the result using * pe__free_digests(). */ pcmk__op_digest_t * pe__calculate_digests(pcmk_resource_t *rsc, const char *task, guint *interval_ms, const pcmk_node_t *node, const xmlNode *xml_op, GHashTable *overrides, bool calc_secure, pcmk_scheduler_t *scheduler) { pcmk__op_digest_t *data = calloc(1, sizeof(pcmk__op_digest_t)); const char *op_version = NULL; GHashTable *params = NULL; if (data == NULL) { pcmk__sched_err("Could not allocate memory for operation digest"); return NULL; } data->rc = pcmk__digest_match; if (xml_op != NULL) { op_version = crm_element_value(xml_op, PCMK_XA_CRM_FEATURE_SET); } if (op_version == NULL && scheduler != NULL && scheduler->input != NULL) { op_version = crm_element_value(scheduler->input, PCMK_XA_CRM_FEATURE_SET); } if (op_version == NULL) { op_version = CRM_FEATURE_SET; } params = pe_rsc_params(rsc, node, scheduler); calculate_main_digest(data, rsc, node, params, task, interval_ms, xml_op, op_version, overrides, scheduler); if (calc_secure) { calculate_secure_digest(data, rsc, params, xml_op, op_version, overrides); } calculate_restart_digest(data, xml_op, op_version); return data; } /*! * \internal * \brief Calculate action digests and store in node's digest cache * * \param[in,out] rsc Resource that action was for * \param[in] task Name of action performed * \param[in] interval_ms Action's interval * \param[in,out] node Node action was performed on * \param[in] xml_op XML of operation in CIB status (if available) * \param[in] calc_secure Whether to calculate secure digest * \param[in,out] scheduler Scheduler data * * \return Pointer to node's digest cache entry */ static pcmk__op_digest_t * rsc_action_digest(pcmk_resource_t *rsc, const char *task, guint interval_ms, pcmk_node_t *node, const xmlNode *xml_op, bool calc_secure, pcmk_scheduler_t *scheduler) { pcmk__op_digest_t *data = NULL; char *key = pcmk__op_key(rsc->id, task, interval_ms); data = g_hash_table_lookup(node->details->digest_cache, key); if (data == NULL) { data = pe__calculate_digests(rsc, task, &interval_ms, node, xml_op, NULL, calc_secure, scheduler); CRM_ASSERT(data != NULL); g_hash_table_insert(node->details->digest_cache, strdup(key), data); } free(key); return data; } /*! * \internal * \brief Calculate operation digests and compare against an XML history entry * * \param[in,out] rsc Resource to check * \param[in] xml_op Resource history XML * \param[in,out] node Node to use for digest calculation * \param[in,out] scheduler Scheduler data * * \return Pointer to node's digest cache entry, with comparison result set */ pcmk__op_digest_t * rsc_action_digest_cmp(pcmk_resource_t *rsc, const xmlNode *xml_op, pcmk_node_t *node, pcmk_scheduler_t *scheduler) { pcmk__op_digest_t *data = NULL; guint interval_ms = 0; const char *op_version; const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK); const char *digest_all; const char *digest_restart; CRM_ASSERT(node != NULL); op_version = crm_element_value(xml_op, PCMK_XA_CRM_FEATURE_SET); digest_all = crm_element_value(xml_op, XML_LRM_ATTR_OP_DIGEST); digest_restart = crm_element_value(xml_op, XML_LRM_ATTR_RESTART_DIGEST); crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); data = rsc_action_digest(rsc, task, interval_ms, node, xml_op, pcmk_is_set(scheduler->flags, pcmk_sched_sanitized), scheduler); if (digest_restart && data->digest_restart_calc && strcmp(data->digest_restart_calc, digest_restart) != 0) { pcmk__rsc_info(rsc, "Parameters to %ums-interval %s action for %s on %s " "changed: hash was %s vs. now %s (restart:%s) %s", interval_ms, task, rsc->id, pe__node_name(node), pcmk__s(digest_restart, "missing"), data->digest_restart_calc, op_version, crm_element_value(xml_op, XML_ATTR_TRANSITION_MAGIC)); data->rc = pcmk__digest_restart; } else if (digest_all == NULL) { /* it is unknown what the previous op digest was */ data->rc = pcmk__digest_unknown; } else if (strcmp(digest_all, data->digest_all_calc) != 0) { /* Given a non-recurring operation with extra parameters configured, * in case that the main digest doesn't match, even if the restart * digest matches, enforce a restart rather than a reload-agent anyway. * So that it ensures any changes of the extra parameters get applied * for this specific operation, and the digests calculated for the * resulting lrm_rsc_op will be correct. * Preserve the implied rc pcmk__digest_restart for the case that the * main digest doesn't match. */ if ((interval_ms == 0) && (data->rc == pcmk__digest_restart)) { pcmk__rsc_info(rsc, "Parameters containing extra ones to %ums-interval" " %s action for %s on %s " "changed: hash was %s vs. now %s (restart:%s) %s", interval_ms, task, rsc->id, pe__node_name(node), pcmk__s(digest_all, "missing"), data->digest_all_calc, op_version, crm_element_value(xml_op, XML_ATTR_TRANSITION_MAGIC)); } else { pcmk__rsc_info(rsc, "Parameters to %ums-interval %s action for %s on %s " "changed: hash was %s vs. now %s (%s:%s) %s", interval_ms, task, rsc->id, pe__node_name(node), pcmk__s(digest_all, "missing"), data->digest_all_calc, (interval_ms > 0)? "reschedule" : "reload", op_version, crm_element_value(xml_op, XML_ATTR_TRANSITION_MAGIC)); data->rc = pcmk__digest_mismatch; } } else { data->rc = pcmk__digest_match; } return data; } /*! * \internal * \brief Create an unfencing summary for use in special node attribute * * Create a string combining a fence device's resource ID, agent type, and * parameter digest (whether for all parameters or just non-private parameters). * This can be stored in a special node attribute, allowing us to detect changes * in either the agent type or parameters, to know whether unfencing must be * redone or can be safely skipped when the device's history is cleaned. * * \param[in] rsc_id Fence device resource ID * \param[in] agent_type Fence device agent * \param[in] param_digest Fence device parameter digest * * \return Newly allocated string with unfencing digest * \note The caller is responsible for freeing the result. */ static inline char * create_unfencing_summary(const char *rsc_id, const char *agent_type, const char *param_digest) { return crm_strdup_printf("%s:%s:%s", rsc_id, agent_type, param_digest); } /*! * \internal * \brief Check whether a node can skip unfencing * * Check whether a fence device's current definition matches a node's * stored summary of when it was last unfenced by the device. * * \param[in] rsc_id Fence device's resource ID * \param[in] agent Fence device's agent type * \param[in] digest_calc Fence device's current parameter digest * \param[in] node_summary Value of node's special unfencing node attribute * (a comma-separated list of unfencing summaries for * all devices that have unfenced this node) * * \return TRUE if digest matches, FALSE otherwise */ static bool unfencing_digest_matches(const char *rsc_id, const char *agent, const char *digest_calc, const char *node_summary) { bool matches = FALSE; if (rsc_id && agent && digest_calc && node_summary) { char *search_secure = create_unfencing_summary(rsc_id, agent, digest_calc); /* The digest was calculated including the device ID and agent, * so there is no risk of collision using strstr(). */ matches = (strstr(node_summary, search_secure) != NULL); crm_trace("Calculated unfencing digest '%s' %sfound in '%s'", search_secure, matches? "" : "not ", node_summary); free(search_secure); } return matches; } /* Magic string to use as action name for digest cache entries used for * unfencing checks. This is not a real action name (i.e. "on"), so * pcmk__check_action_config() won't confuse these entries with real actions. */ #define STONITH_DIGEST_TASK "stonith-on" /*! * \internal * \brief Calculate fence device digests and digest comparison result * * \param[in,out] rsc Fence device resource * \param[in] agent Fence device's agent type * \param[in,out] node Node with digest cache to use * \param[in,out] scheduler Scheduler data * * \return Node's digest cache entry */ pcmk__op_digest_t * pe__compare_fencing_digest(pcmk_resource_t *rsc, const char *agent, pcmk_node_t *node, pcmk_scheduler_t *scheduler) { const char *node_summary = NULL; // Calculate device's current parameter digests pcmk__op_digest_t *data = rsc_action_digest(rsc, STONITH_DIGEST_TASK, 0U, node, NULL, TRUE, scheduler); // Check whether node has special unfencing summary node attribute node_summary = pe_node_attribute_raw(node, CRM_ATTR_DIGESTS_ALL); if (node_summary == NULL) { data->rc = pcmk__digest_unknown; return data; } // Check whether full parameter digest matches if (unfencing_digest_matches(rsc->id, agent, data->digest_all_calc, node_summary)) { data->rc = pcmk__digest_match; return data; } // Check whether secure parameter digest matches node_summary = pe_node_attribute_raw(node, CRM_ATTR_DIGESTS_SECURE); if (unfencing_digest_matches(rsc->id, agent, data->digest_secure_calc, node_summary)) { data->rc = pcmk__digest_match; if (!pcmk__is_daemon && scheduler->priv != NULL) { pcmk__output_t *out = scheduler->priv; out->info(out, "Only 'private' parameters to %s " "for unfencing %s changed", rsc->id, pe__node_name(node)); } return data; } // Parameters don't match data->rc = pcmk__digest_mismatch; if (pcmk_is_set(scheduler->flags, pcmk_sched_sanitized) && (data->digest_secure_calc != NULL)) { if (scheduler->priv != NULL) { pcmk__output_t *out = scheduler->priv; char *digest = create_unfencing_summary(rsc->id, agent, data->digest_secure_calc); out->info(out, "Parameters to %s for unfencing " "%s changed, try '%s'", rsc->id, pe__node_name(node), digest); free(digest); } else if (!pcmk__is_daemon) { char *digest = create_unfencing_summary(rsc->id, agent, data->digest_secure_calc); printf("Parameters to %s for unfencing %s changed, try '%s'\n", rsc->id, pe__node_name(node), digest); free(digest); } } return data; } diff --git a/lib/pengine/remote.c b/lib/pengine/remote.c index 42163f61f9..22808925ca 100644 --- a/lib/pengine/remote.c +++ b/lib/pengine/remote.c @@ -1,275 +1,274 @@ /* * Copyright 2013-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include bool pe__resource_is_remote_conn(const pcmk_resource_t *rsc) { return (rsc != NULL) && rsc->is_remote_node && pe__is_remote_node(pe_find_node(rsc->cluster->nodes, rsc->id)); } bool pe__is_remote_node(const pcmk_node_t *node) { return (node != NULL) && (node->details->type == pcmk_node_variant_remote) && ((node->details->remote_rsc == NULL) || (node->details->remote_rsc->container == NULL)); } bool pe__is_guest_node(const pcmk_node_t *node) { return (node != NULL) && (node->details->type == pcmk_node_variant_remote) && (node->details->remote_rsc != NULL) && (node->details->remote_rsc->container != NULL); } bool pe__is_guest_or_remote_node(const pcmk_node_t *node) { return (node != NULL) && (node->details->type == pcmk_node_variant_remote); } bool pe__is_bundle_node(const pcmk_node_t *node) { return pe__is_guest_node(node) && pe_rsc_is_bundled(node->details->remote_rsc); } /*! * \internal * \brief Check whether a resource creates a guest node * * If a given resource contains a filler resource that is a remote connection, * return that filler resource (or NULL if none is found). * * \param[in] scheduler Scheduler data * \param[in] rsc Resource to check * * \return Filler resource with remote connection, or NULL if none found */ pcmk_resource_t * pe__resource_contains_guest_node(const pcmk_scheduler_t *scheduler, const pcmk_resource_t *rsc) { if ((rsc != NULL) && (scheduler != NULL) && pcmk_is_set(scheduler->flags, pcmk_sched_have_remote_nodes)) { for (GList *gIter = rsc->fillers; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *filler = gIter->data; if (filler->is_remote_node) { return filler; } } } return NULL; } bool xml_contains_remote_node(xmlNode *xml) { const char *value = NULL; if (xml == NULL) { return false; } value = crm_element_value(xml, XML_ATTR_TYPE); if (!pcmk__str_eq(value, "remote", pcmk__str_casei)) { return false; } value = crm_element_value(xml, XML_AGENT_ATTR_CLASS); if (!pcmk__str_eq(value, PCMK_RESOURCE_CLASS_OCF, pcmk__str_casei)) { return false; } value = crm_element_value(xml, XML_AGENT_ATTR_PROVIDER); if (!pcmk__str_eq(value, "pacemaker", pcmk__str_casei)) { return false; } return true; } /*! * \internal * \brief Execute a supplied function for each guest node running on a host * * \param[in] scheduler Scheduler data * \param[in] host Host node to check * \param[in] helper Function to call for each guest node * \param[in,out] user_data Pointer to pass to helper function */ void pe_foreach_guest_node(const pcmk_scheduler_t *scheduler, const pcmk_node_t *host, void (*helper)(const pcmk_node_t*, void*), void *user_data) { GList *iter; CRM_CHECK(scheduler && host && host->details && helper, return); if (!pcmk_is_set(scheduler->flags, pcmk_sched_have_remote_nodes)) { return; } for (iter = host->details->running_rsc; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; if (rsc->is_remote_node && (rsc->container != NULL)) { pcmk_node_t *guest_node = pe_find_node(scheduler->nodes, rsc->id); if (guest_node) { (*helper)(guest_node, user_data); } } } } /*! * \internal * \brief Create CIB XML for an implicit remote connection * * \param[in,out] parent If not NULL, use as parent XML element * \param[in] uname Name of Pacemaker Remote node * \param[in] container If not NULL, use this as connection container * \param[in] migrateable If not NULL, use as allow-migrate value * \param[in] is_managed If not NULL, use as is-managed value * \param[in] start_timeout If not NULL, use as remote connect timeout * \param[in] server If not NULL, use as remote server value * \param[in] port If not NULL, use as remote port value * * \return Newly created XML */ xmlNode * pe_create_remote_xml(xmlNode *parent, const char *uname, const char *container_id, const char *migrateable, const char *is_managed, const char *start_timeout, const char *server, const char *port) { xmlNode *remote; xmlNode *xml_sub; remote = create_xml_node(parent, XML_CIB_TAG_RESOURCE); // Add identity crm_xml_add(remote, PCMK_XA_ID, uname); crm_xml_add(remote, XML_AGENT_ATTR_CLASS, PCMK_RESOURCE_CLASS_OCF); crm_xml_add(remote, XML_AGENT_ATTR_PROVIDER, "pacemaker"); crm_xml_add(remote, XML_ATTR_TYPE, "remote"); // Add meta-attributes xml_sub = create_xml_node(remote, XML_TAG_META_SETS); crm_xml_set_id(xml_sub, "%s-%s", uname, XML_TAG_META_SETS); crm_create_nvpair_xml(xml_sub, NULL, PCMK__META_INTERNAL_RSC, XML_BOOLEAN_TRUE); if (container_id) { crm_create_nvpair_xml(xml_sub, NULL, PCMK__META_CONTAINER, container_id); } if (migrateable) { crm_create_nvpair_xml(xml_sub, NULL, PCMK_META_ALLOW_MIGRATE, migrateable); } if (is_managed) { crm_create_nvpair_xml(xml_sub, NULL, PCMK_META_IS_MANAGED, is_managed); } // Add instance attributes if (port || server) { xml_sub = create_xml_node(remote, XML_TAG_ATTR_SETS); crm_xml_set_id(xml_sub, "%s-%s", uname, XML_TAG_ATTR_SETS); if (server) { - crm_create_nvpair_xml(xml_sub, NULL, XML_RSC_ATTR_REMOTE_RA_ADDR, - server); + crm_create_nvpair_xml(xml_sub, NULL, PCMK_REMOTE_RA_ADDR, server); } if (port) { crm_create_nvpair_xml(xml_sub, NULL, XML_RSC_ATTR_REMOTE_RA_PORT, port); } } // Add operations xml_sub = create_xml_node(remote, "operations"); crm_create_op_xml(xml_sub, uname, PCMK_ACTION_MONITOR, "30s", "30s"); if (start_timeout) { crm_create_op_xml(xml_sub, uname, PCMK_ACTION_START, "0", start_timeout); } return remote; } // History entry to be checked for fail count clearing struct check_op { const xmlNode *rsc_op; // History entry XML pcmk_resource_t *rsc; // Known resource corresponding to history entry pcmk_node_t *node; // Known node corresponding to history entry enum pcmk__check_parameters check_type; // What needs checking }; void pe__add_param_check(const xmlNode *rsc_op, pcmk_resource_t *rsc, pcmk_node_t *node, enum pcmk__check_parameters flag, pcmk_scheduler_t *scheduler) { struct check_op *check_op = NULL; CRM_CHECK(scheduler && rsc_op && rsc && node, return); check_op = calloc(1, sizeof(struct check_op)); CRM_ASSERT(check_op != NULL); crm_trace("Deferring checks of %s until after allocation", ID(rsc_op)); check_op->rsc_op = rsc_op; check_op->rsc = rsc; check_op->node = node; check_op->check_type = flag; scheduler->param_check = g_list_prepend(scheduler->param_check, check_op); } /*! * \internal * \brief Call a function for each action to be checked for addr substitution * * \param[in,out] scheduler Scheduler data * \param[in] cb Function to be called */ void pe__foreach_param_check(pcmk_scheduler_t *scheduler, void (*cb)(pcmk_resource_t*, pcmk_node_t*, const xmlNode*, enum pcmk__check_parameters)) { CRM_CHECK(scheduler && cb, return); for (GList *item = scheduler->param_check; item != NULL; item = item->next) { struct check_op *check_op = item->data; cb(check_op->rsc, check_op->node, check_op->rsc_op, check_op->check_type); } } void pe__free_param_checks(pcmk_scheduler_t *scheduler) { if (scheduler && scheduler->param_check) { g_list_free_full(scheduler->param_check, free); scheduler->param_check = NULL; } }