diff --git a/daemons/attrd/attrd_cib.c b/daemons/attrd/attrd_cib.c index b8f509ab7d..6129f54c75 100644 --- a/daemons/attrd/attrd_cib.c +++ b/daemons/attrd/attrd_cib.c @@ -1,692 +1,692 @@ /* * 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 // PRIu32 #include #include #include #include // cib__* #include #include #include #include #include // pcmk__get_node() #include "pacemaker-attrd.h" static int last_cib_op_done = 0; static void write_attribute(attribute_t *a, bool ignore_delay); static void attrd_cib_destroy_cb(gpointer user_data) { cib_t *cib = user_data; cib->cmds->signoff(cib); if (attrd_shutting_down(false)) { crm_info("Disconnected from the CIB manager"); } else { // @TODO This should trigger a reconnect, not a shutdown crm_crit("Lost connection to the CIB manager, shutting down"); attrd_exit_status = CRM_EX_DISCONNECT; attrd_shutdown(0); } } static void attrd_cib_updated_cb(const char *event, xmlNode *msg) { const xmlNode *patchset = NULL; const char *client_name = NULL; bool status_changed = false; if (cib__get_notify_patchset(msg, &patchset) != pcmk_rc_ok) { return; } if (pcmk__cib_element_in_patchset(patchset, PCMK_XE_ALERTS)) { if (attrd_shutting_down(true)) { crm_debug("Ignoring alerts change in CIB during shutdown"); } else { mainloop_set_trigger(attrd_config_read); } } status_changed = pcmk__cib_element_in_patchset(patchset, PCMK_XE_STATUS); client_name = crm_element_value(msg, PCMK__XA_CIB_CLIENTNAME); if (!cib__client_triggers_refresh(client_name)) { /* This change came from a source that ensured the CIB is consistent * with our attributes table, so we don't need to write anything out. */ return; } if (!attrd_election_won()) { // Don't write attributes if we're not the writer return; } if (status_changed || pcmk__cib_element_in_patchset(patchset, PCMK_XE_NODES)) { if (attrd_shutting_down(true)) { crm_debug("Ignoring node change in CIB during shutdown"); return; } /* An unsafe client modified the PCMK_XE_NODES or PCMK_XE_STATUS * section. Write transient attributes to ensure they're up-to-date in * the CIB. */ if (client_name == NULL) { client_name = crm_element_value(msg, PCMK__XA_CIB_CLIENTID); } crm_notice("Updating all attributes after %s event triggered by %s", event, pcmk__s(client_name, "unidentified client")); attrd_write_attributes(attrd_write_all); } } int attrd_cib_connect(int max_retry) { static int attempts = 0; int rc = -ENOTCONN; the_cib = cib_new(); if (the_cib == NULL) { return -ENOTCONN; } do { if (attempts > 0) { sleep(attempts); } attempts++; crm_debug("Connection attempt %d to the CIB manager", attempts); rc = the_cib->cmds->signon(the_cib, crm_system_name, cib_command); } while ((rc != pcmk_ok) && (attempts < max_retry)); if (rc != pcmk_ok) { crm_err("Connection to the CIB manager failed: %s " QB_XS " rc=%d", pcmk_strerror(rc), rc); goto cleanup; } crm_debug("Connected to the CIB manager after %d attempts", attempts); rc = the_cib->cmds->set_connection_dnotify(the_cib, attrd_cib_destroy_cb); if (rc != pcmk_ok) { crm_err("Could not set disconnection callback"); goto cleanup; } rc = the_cib->cmds->add_notify_callback(the_cib, PCMK__VALUE_CIB_DIFF_NOTIFY, attrd_cib_updated_cb); if (rc != pcmk_ok) { crm_err("Could not set CIB notification callback"); goto cleanup; } return pcmk_ok; cleanup: cib__clean_up_connection(&the_cib); return -ENOTCONN; } void attrd_cib_disconnect(void) { CRM_CHECK(the_cib != NULL, return); the_cib->cmds->del_notify_callback(the_cib, PCMK__VALUE_CIB_DIFF_NOTIFY, attrd_cib_updated_cb); cib__clean_up_connection(&the_cib); mainloop_destroy_trigger(attrd_config_read); } static void attrd_erase_cb(xmlNode *msg, int call_id, int rc, xmlNode *output, void *user_data) { const char *node = pcmk__s((const char *) user_data, "a node"); if (rc == pcmk_ok) { crm_info("Cleared transient node attributes for %s from CIB", node); } else { crm_err("Unable to clear transient node attributes for %s from CIB: %s", node, pcmk_strerror(rc)); } } #define XPATH_TRANSIENT "//" PCMK__XE_NODE_STATE \ "[@" PCMK_XA_UNAME "='%s']" \ "/" PCMK__XE_TRANSIENT_ATTRIBUTES /*! * \internal * \brief Wipe all transient node attributes for a node from the CIB * * \param[in] node Node to clear attributes for */ void attrd_cib_erase_transient_attrs(const char *node) { int call_id = 0; char *xpath = NULL; CRM_CHECK(node != NULL, return); xpath = crm_strdup_printf(XPATH_TRANSIENT, node); crm_debug("Clearing transient node attributes for %s from CIB using %s", node, xpath); call_id = the_cib->cmds->remove(the_cib, xpath, NULL, cib_xpath); free(xpath); the_cib->cmds->register_callback_full(the_cib, call_id, 120, FALSE, pcmk__str_copy(node), "attrd_erase_cb", attrd_erase_cb, free); } /*! * \internal * \brief Prepare the CIB after cluster is connected */ void attrd_cib_init(void) { /* We have no attribute values in memory, so wipe the CIB to match. This is * normally done by the DC's controller when this node leaves the cluster, but * this handles the case where the node restarted so quickly that the * cluster layer didn't notice. * * \todo If the attribute manager respawns after crashing (see * PCMK_ENV_RESPAWNED), ideally we'd skip this and sync our attributes * from the writer. However, currently we reject any values for us * that the writer has, in attrd_peer_update(). */ attrd_cib_erase_transient_attrs(attrd_cluster->priv->node_name); // Set a trigger for reading the CIB (for the alerts section) attrd_config_read = mainloop_add_trigger(G_PRIORITY_HIGH, attrd_read_options, NULL); // Always read the CIB at start-up mainloop_set_trigger(attrd_config_read); } static gboolean attribute_timer_cb(gpointer data) { attribute_t *a = data; crm_trace("Dampen interval expired for %s", a->id); attrd_write_or_elect_attribute(a); return FALSE; } static void attrd_cib_callback(xmlNode *msg, int call_id, int rc, xmlNode *output, void *user_data) { int level = LOG_ERR; GHashTableIter iter; const char *peer = NULL; attribute_value_t *v = NULL; char *name = user_data; attribute_t *a = g_hash_table_lookup(attributes, name); if(a == NULL) { crm_info("Attribute %s no longer exists", name); return; } a->update = 0; if (rc == pcmk_ok && call_id < 0) { rc = call_id; } switch (rc) { case pcmk_ok: level = LOG_INFO; last_cib_op_done = call_id; if (a->timer && !a->timeout_ms) { // Remove temporary dampening for failed writes mainloop_timer_del(a->timer); a->timer = NULL; } break; case -pcmk_err_diff_failed: /* When an attr changes while the CIB is syncing */ case -ETIME: /* When an attr changes while there is a DC election */ case -ENXIO: /* When an attr changes while the CIB is syncing a * newer config from a node that just came up */ level = LOG_WARNING; break; } do_crm_log(level, "CIB update %d result for %s: %s " QB_XS " rc=%d", call_id, a->id, pcmk_strerror(rc), rc); g_hash_table_iter_init(&iter, a->values); while (g_hash_table_iter_next(&iter, (gpointer *) & peer, (gpointer *) & v)) { if (rc == pcmk_ok) { crm_info("* Wrote %s[%s]=%s", a->id, peer, pcmk__s(v->requested, "(unset)")); pcmk__str_update(&(v->requested), NULL); } else { do_crm_log(level, "* Could not write %s[%s]=%s", a->id, peer, pcmk__s(v->requested, "(unset)")); /* Reattempt write below if we are still the writer */ attrd_set_attr_flags(a, attrd_attr_changed); } } if (pcmk_is_set(a->flags, attrd_attr_changed) && attrd_election_won()) { if (rc == pcmk_ok) { /* We deferred a write of a new update because this update was in * progress. Write out the new value without additional delay. */ crm_debug("Pending update for %s can be written now", a->id); write_attribute(a, false); /* We're re-attempting a write because the original failed; delay * the next attempt so we don't potentially flood the CIB manager * and logs with a zillion attempts per second. * * @TODO We could elect a new writer instead. However, we'd have to * somehow downgrade our vote, and we'd still need something like this * if all peers similarly fail to write this attribute (which may * indicate a corrupted attribute entry rather than a CIB issue). */ } else if (a->timer) { // Attribute has a dampening value, so use that as delay if (!mainloop_timer_running(a->timer)) { crm_trace("Delayed re-attempted write for %s by %s", name, pcmk__readable_interval(a->timeout_ms)); mainloop_timer_start(a->timer); } } else { /* Set a temporary dampening of 2 seconds (timer will continue * to exist until the attribute's dampening gets set or the * write succeeds). */ a->timer = attrd_add_timer(a->id, 2000, a); mainloop_timer_start(a->timer); } } } /*! * \internal * \brief Add a set-attribute update request to the current CIB transaction * * \param[in] attr Attribute to update * \param[in] attr_id ID of attribute to update * \param[in] node_id ID of node for which to update attribute value * \param[in] set_id ID of attribute set * \param[in] value New value for attribute * * \return Standard Pacemaker return code */ static int add_set_attr_update(const attribute_t *attr, const char *attr_id, const char *node_id, const char *set_id, const char *value) { xmlNode *update = pcmk__xe_create(NULL, PCMK__XE_NODE_STATE); xmlNode *child = update; int rc = ENOMEM; crm_xml_add(child, PCMK_XA_ID, node_id); child = pcmk__xe_create(child, PCMK__XE_TRANSIENT_ATTRIBUTES); crm_xml_add(child, PCMK_XA_ID, node_id); child = pcmk__xe_create(child, attr->set_type); crm_xml_add(child, PCMK_XA_ID, set_id); child = pcmk__xe_create(child, PCMK_XE_NVPAIR); crm_xml_add(child, PCMK_XA_ID, attr_id); crm_xml_add(child, PCMK_XA_NAME, attr->id); crm_xml_add(child, PCMK_XA_VALUE, value); rc = the_cib->cmds->modify(the_cib, PCMK_XE_STATUS, update, cib_can_create|cib_transaction); rc = pcmk_legacy2rc(rc); pcmk__xml_free(update); return rc; } /*! * \internal * \brief Add an unset-attribute update request to the current CIB transaction * * \param[in] attr Attribute to update * \param[in] attr_id ID of attribute to update * \param[in] node_id ID of node for which to update attribute value * \param[in] set_id ID of attribute set * * \return Standard Pacemaker return code */ static int add_unset_attr_update(const attribute_t *attr, const char *attr_id, const char *node_id, const char *set_id) { char *xpath = crm_strdup_printf("/" PCMK_XE_CIB "/" PCMK_XE_STATUS "/" PCMK__XE_NODE_STATE "[@" PCMK_XA_ID "='%s']" "/" PCMK__XE_TRANSIENT_ATTRIBUTES "[@" PCMK_XA_ID "='%s']" "/%s[@" PCMK_XA_ID "='%s']" "/" PCMK_XE_NVPAIR "[@" PCMK_XA_ID "='%s' " "and @" PCMK_XA_NAME "='%s']", node_id, node_id, attr->set_type, set_id, attr_id, attr->id); int rc = the_cib->cmds->remove(the_cib, xpath, NULL, cib_xpath|cib_transaction); free(xpath); return pcmk_legacy2rc(rc); } /*! * \internal * \brief Add an attribute update request to the current CIB transaction * * \param[in] attr Attribute to update * \param[in] value New value for attribute * \param[in] node_id ID of node for which to update attribute value * * \return Standard Pacemaker return code */ static int add_attr_update(const attribute_t *attr, const char *value, const char *node_id) { char *set_id = attrd_set_id(attr, node_id); char *nvpair_id = attrd_nvpair_id(attr, node_id); int rc = pcmk_rc_ok; if (value == NULL) { rc = add_unset_attr_update(attr, nvpair_id, node_id, set_id); } else { rc = add_set_attr_update(attr, nvpair_id, node_id, set_id, value); } free(set_id); free(nvpair_id); return rc; } static void send_alert_attributes_value(attribute_t *a, GHashTable *t) { int rc = 0; attribute_value_t *at = NULL; GHashTableIter vIter; g_hash_table_iter_init(&vIter, t); while (g_hash_table_iter_next(&vIter, NULL, (gpointer *) & at)) { rc = attrd_send_attribute_alert(at->nodename, at->nodeid, a->id, at->current); crm_trace("Sent alerts for %s[%s]=%s: nodeid=%d rc=%d", a->id, at->nodename, at->current, at->nodeid, rc); } } static void set_alert_attribute_value(GHashTable *t, attribute_value_t *v) { attribute_value_t *a_v = pcmk__assert_alloc(1, sizeof(attribute_value_t)); a_v->nodeid = v->nodeid; a_v->nodename = pcmk__str_copy(v->nodename); a_v->current = pcmk__str_copy(v->current); g_hash_table_replace(t, a_v->nodename, a_v); } mainloop_timer_t * attrd_add_timer(const char *id, int timeout_ms, attribute_t *attr) { return mainloop_timer_add(id, timeout_ms, FALSE, attribute_timer_cb, attr); } /*! * \internal * \brief Write an attribute's values to the CIB if appropriate * * \param[in,out] a Attribute to write * \param[in] ignore_delay If true, write attribute now regardless of any * configured delay */ static void write_attribute(attribute_t *a, bool ignore_delay) { int private_updates = 0, cib_updates = 0; attribute_value_t *v = NULL; GHashTableIter iter; GHashTable *alert_attribute_value = NULL; int rc = pcmk_ok; if (a == NULL) { return; } /* If this attribute will be written to the CIB ... */ if (!stand_alone && !pcmk_is_set(a->flags, attrd_attr_is_private)) { /* Defer the write if now's not a good time */ if (a->update && (a->update < last_cib_op_done)) { crm_info("Write out of '%s' continuing: update %d considered lost", a->id, a->update); a->update = 0; // Don't log this message again } else if (a->update) { crm_info("Write out of '%s' delayed: update %d in progress", a->id, a->update); goto done; } else if (mainloop_timer_running(a->timer)) { if (ignore_delay) { mainloop_timer_stop(a->timer); crm_debug("Overriding '%s' write delay", a->id); } else { crm_info("Delaying write of '%s'", a->id); goto done; } } // Initiate a transaction for all the peer value updates CRM_CHECK(the_cib != NULL, goto done); the_cib->cmds->set_user(the_cib, a->user); rc = the_cib->cmds->init_transaction(the_cib); if (rc != pcmk_ok) { crm_err("Failed to write %s (set %s): Could not initiate " "CIB transaction", a->id, pcmk__s(a->set_id, "unspecified")); goto done; } } /* Attribute will be written shortly, so clear changed flag and force * write flag, and initialize UUID missing flag to false. */ attrd_clear_attr_flags(a, attrd_attr_changed|attrd_attr_uuid_missing|attrd_attr_force_write); /* Make the table for the attribute trap */ alert_attribute_value = pcmk__strikey_table(NULL, attrd_free_attribute_value); /* Iterate over each peer value of this attribute */ g_hash_table_iter_init(&iter, a->values); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &v)) { const char *node_xml_id = NULL; // Try to get the XML ID used for the node in the CIB if (pcmk_is_set(v->flags, attrd_value_remote)) { // A Pacemaker Remote node's XML ID is the same as its name node_xml_id = v->nodename; } else { /* Get cluster node XML IDs from the peer caches. * This will create a cluster node cache entry if none exists. */ pcmk__node_status_t *peer = pcmk__get_node(v->nodeid, v->nodename, NULL, pcmk__node_search_any); - node_xml_id = peer->xml_id; + node_xml_id = pcmk__cluster_node_uuid(peer); // Remember peer's node ID if we're just now learning it if ((peer->cluster_layer_id != 0) && (v->nodeid == 0)) { crm_trace("Learned ID %" PRIu32 " for node %s", peer->cluster_layer_id, v->nodename); v->nodeid = peer->cluster_layer_id; } } /* If this is a private attribute, no update needs to be sent */ if (stand_alone || pcmk_is_set(a->flags, attrd_attr_is_private)) { private_updates++; continue; } // Defer write if this is a cluster node that's never been seen if (node_xml_id == NULL) { attrd_set_attr_flags(a, attrd_attr_uuid_missing); crm_notice("Cannot write %s[%s]='%s' to CIB because node's XML ID " "is unknown (will retry if learned)", a->id, v->nodename, v->current); continue; } // Update this value as part of the CIB transaction we're building rc = add_attr_update(a, v->current, node_xml_id); if (rc != pcmk_rc_ok) { crm_err("Couldn't add %s[%s]='%s' to CIB transaction: %s " QB_XS " node XML ID %s", a->id, v->nodename, v->current, pcmk_rc_str(rc), node_xml_id); continue; } crm_debug("Added %s[%s]=%s to CIB transaction (node XML ID %s)", a->id, v->nodename, pcmk__s(v->current, "(unset)"), node_xml_id); cib_updates++; /* Preservation of the attribute to transmit alert */ set_alert_attribute_value(alert_attribute_value, v); // Save this value so we can log it when write completes pcmk__str_update(&(v->requested), v->current); } if (private_updates) { crm_info("Processed %d private change%s for %s (set %s)", private_updates, pcmk__plural_s(private_updates), a->id, pcmk__s(a->set_id, "unspecified")); } if (cib_updates > 0) { char *id = pcmk__str_copy(a->id); // Commit transaction a->update = the_cib->cmds->end_transaction(the_cib, true, cib_none); crm_info("Sent CIB request %d with %d change%s for %s (set %s)", a->update, cib_updates, pcmk__plural_s(cib_updates), a->id, pcmk__s(a->set_id, "unspecified")); if (the_cib->cmds->register_callback_full(the_cib, a->update, CIB_OP_TIMEOUT_S, FALSE, id, "attrd_cib_callback", attrd_cib_callback, free)) { // Transmit alert of the attribute send_alert_attributes_value(a, alert_attribute_value); } } done: // Discard transaction (if any) if (the_cib != NULL) { the_cib->cmds->end_transaction(the_cib, false, cib_none); the_cib->cmds->set_user(the_cib, NULL); } if (alert_attribute_value != NULL) { g_hash_table_destroy(alert_attribute_value); } } /*! * \internal * \brief Write out attributes * * \param[in] options Group of enum attrd_write_options */ void attrd_write_attributes(uint32_t options) { GHashTableIter iter; attribute_t *a = NULL; crm_debug("Writing out %s attributes", pcmk_is_set(options, attrd_write_all)? "all" : "changed"); g_hash_table_iter_init(&iter, attributes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & a)) { if (!pcmk_is_set(options, attrd_write_all) && pcmk_is_set(a->flags, attrd_attr_uuid_missing)) { // Try writing this attribute again, in case peer ID was learned attrd_set_attr_flags(a, attrd_attr_changed); } else if (pcmk_is_set(a->flags, attrd_attr_force_write)) { /* If the force_write flag is set, write the attribute. */ attrd_set_attr_flags(a, attrd_attr_changed); } if (pcmk_is_set(options, attrd_write_all) || pcmk_is_set(a->flags, attrd_attr_changed)) { bool ignore_delay = pcmk_is_set(options, attrd_write_no_delay); if (pcmk_is_set(a->flags, attrd_attr_force_write)) { // Always ignore delay when forced write flag is set ignore_delay = true; } write_attribute(a, ignore_delay); } else { crm_trace("Skipping unchanged attribute %s", a->id); } } } void attrd_write_or_elect_attribute(attribute_t *a) { if (attrd_election_won()) { write_attribute(a, false); } else { attrd_start_election_if_needed(); } } diff --git a/daemons/attrd/attrd_corosync.c b/daemons/attrd/attrd_corosync.c index eeb2b9b1df..72ebc1843b 100644 --- a/daemons/attrd/attrd_corosync.c +++ b/daemons/attrd/attrd_corosync.c @@ -1,612 +1,612 @@ /* * 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 #include #include #include #include "pacemaker-attrd.h" static xmlNode * attrd_confirmation(int callid) { xmlNode *node = pcmk__xe_create(NULL, __func__); crm_xml_add(node, PCMK__XA_T, PCMK__VALUE_ATTRD); crm_xml_add(node, PCMK__XA_SRC, pcmk__cluster_local_node_name()); crm_xml_add(node, PCMK_XA_TASK, PCMK__ATTRD_CMD_CONFIRM); crm_xml_add_int(node, PCMK__XA_CALL_ID, callid); return node; } static void attrd_peer_message(pcmk__node_status_t *peer, xmlNode *xml) { const char *election_op = crm_element_value(xml, PCMK__XA_CRM_TASK); if (election_op) { attrd_handle_election_op(peer, xml); return; } if (attrd_shutting_down(false)) { /* If we're shutting down, we want to continue responding to election * ops as long as we're a cluster member (because our vote may be * needed). Ignore all other messages. */ return; } else { pcmk__request_t request = { .ipc_client = NULL, .ipc_id = 0, .ipc_flags = 0, .peer = peer->name, .xml = xml, .call_options = 0, .result = PCMK__UNKNOWN_RESULT, }; request.op = crm_element_value_copy(request.xml, PCMK_XA_TASK); CRM_CHECK(request.op != NULL, return); attrd_handle_request(&request); /* Having finished handling the request, check to see if the originating * peer requested confirmation. If so, send that confirmation back now. */ if (pcmk__xe_attr_is_true(xml, PCMK__XA_CONFIRM) && !pcmk__str_eq(request.op, PCMK__ATTRD_CMD_CONFIRM, pcmk__str_none)) { int callid = 0; xmlNode *reply = NULL; /* Add the confirmation ID for the message we are confirming to the * response so the originating peer knows what they're a confirmation * for. */ crm_element_value_int(xml, PCMK__XA_CALL_ID, &callid); reply = attrd_confirmation(callid); /* And then send the confirmation back to the originating peer. This * ends up right back in this same function (attrd_peer_message) on the * peer where it will have to do something with a PCMK__XA_CONFIRM type * message. */ crm_debug("Sending %s a confirmation", peer->name); attrd_send_message(peer, reply, false); pcmk__xml_free(reply); } pcmk__reset_request(&request); } } static void attrd_cpg_dispatch(cpg_handle_t handle, const struct cpg_name *groupName, uint32_t nodeid, uint32_t pid, void *msg, size_t msg_len) { xmlNode *xml = NULL; const char *from = NULL; char *data = pcmk__cpg_message_data(handle, nodeid, pid, msg, &from); if(data == NULL) { return; } xml = pcmk__xml_parse(data); if (xml == NULL) { crm_err("Bad message received from %s[%u]: '%.120s'", from, nodeid, data); } else { attrd_peer_message(pcmk__get_node(nodeid, from, NULL, pcmk__node_search_cluster_member), xml); } pcmk__xml_free(xml); free(data); } static void attrd_cpg_destroy(gpointer unused) { if (attrd_shutting_down(false)) { crm_info("Disconnected from Corosync process group"); } else { crm_crit("Lost connection to Corosync process group, shutting down"); attrd_exit_status = CRM_EX_DISCONNECT; attrd_shutdown(0); } } /*! * \internal * \brief Broadcast an update for a single attribute value * * \param[in] a Attribute to broadcast * \param[in] v Attribute value to broadcast */ void attrd_broadcast_value(const attribute_t *a, const attribute_value_t *v) { xmlNode *op = pcmk__xe_create(NULL, PCMK_XE_OP); crm_xml_add(op, PCMK_XA_TASK, PCMK__ATTRD_CMD_UPDATE); attrd_add_value_xml(op, a, v, false); attrd_send_message(NULL, op, false); pcmk__xml_free(op); } #define state_text(state) pcmk__s((state), "in unknown state") static void attrd_peer_change_cb(enum pcmk__node_update kind, pcmk__node_status_t *peer, const void *data) { bool gone = false; bool is_remote = pcmk_is_set(peer->flags, pcmk__node_status_remote); switch (kind) { case pcmk__node_update_name: crm_debug("%s node %s is now %s", (is_remote? "Remote" : "Cluster"), peer->name, state_text(peer->state)); break; case pcmk__node_update_processes: if (!pcmk_is_set(peer->processes, crm_get_cluster_proc())) { gone = true; } crm_debug("Node %s is %s a peer", peer->name, (gone? "no longer" : "now")); break; case pcmk__node_update_state: crm_debug("%s node %s is now %s (was %s)", (is_remote? "Remote" : "Cluster"), peer->name, state_text(peer->state), state_text(data)); if (pcmk__str_eq(peer->state, PCMK_VALUE_MEMBER, pcmk__str_none)) { /* If we're the writer, send new peers a list of all attributes * (unless it's a remote node, which doesn't run its own attrd) */ if (attrd_election_won() && !pcmk_is_set(peer->flags, pcmk__node_status_remote)) { attrd_peer_sync(peer); } } else { // Remove all attribute values associated with lost nodes attrd_peer_remove(peer->name, false, "loss"); gone = true; } break; } // Remove votes from cluster nodes that leave, in case election in progress if (gone && !is_remote) { attrd_remove_voter(peer); attrd_remove_peer_protocol_ver(peer->name); attrd_do_not_expect_from_peer(peer->name); } } static void record_peer_nodeid(attribute_value_t *v, const char *host) { pcmk__node_status_t *known_peer = pcmk__get_node(v->nodeid, host, NULL, pcmk__node_search_cluster_member); - crm_trace("Learned %s has node id %s", - known_peer->name, known_peer->xml_id); + crm_trace("Learned %s has XML ID %s", + known_peer->name, pcmk__cluster_node_uuid(known_peer)); if (attrd_election_won()) { attrd_write_attributes(attrd_write_changed); } } #define readable_value(rv_v) pcmk__s((rv_v)->current, "(unset)") #define readable_peer(p) \ (((p) == NULL)? "all peers" : pcmk__s((p)->name, "unknown peer")) static void update_attr_on_host(attribute_t *a, const pcmk__node_status_t *peer, const xmlNode *xml, const char *attr, const char *value, const char *host, bool filter) { int is_remote = 0; bool changed = false; attribute_value_t *v = NULL; // Create entry for value if not already existing v = g_hash_table_lookup(a->values, host); if (v == NULL) { v = pcmk__assert_alloc(1, sizeof(attribute_value_t)); v->nodename = pcmk__str_copy(host); g_hash_table_replace(a->values, v->nodename, v); } // If value is for a Pacemaker Remote node, remember that crm_element_value_int(xml, PCMK__XA_ATTR_IS_REMOTE, &is_remote); if (is_remote) { attrd_set_value_flags(v, attrd_value_remote); pcmk__assert(pcmk__cluster_lookup_remote_node(host) != NULL); } // Check whether the value changed changed = !pcmk__str_eq(v->current, value, pcmk__str_casei); if (changed && filter && pcmk__str_eq(host, attrd_cluster->priv->node_name, pcmk__str_casei)) { /* Broadcast the local value for an attribute that differs from the * value provided in a peer's attribute synchronization response. This * ensures a node's values for itself take precedence and all peers are * kept in sync. */ v = g_hash_table_lookup(a->values, attrd_cluster->priv->node_name); crm_notice("%s[%s]: local value '%s' takes priority over '%s' from %s", attr, host, readable_value(v), value, peer->name); attrd_broadcast_value(a, v); } else if (changed) { crm_notice("Setting %s[%s]%s%s: %s -> %s " QB_XS " from %s with %s write delay", attr, host, a->set_type ? " in " : "", pcmk__s(a->set_type, ""), readable_value(v), pcmk__s(value, "(unset)"), peer->name, (a->timeout_ms == 0)? "no" : pcmk__readable_interval(a->timeout_ms)); pcmk__str_update(&v->current, value); attrd_set_attr_flags(a, attrd_attr_changed); if (pcmk__str_eq(host, attrd_cluster->priv->node_name, pcmk__str_casei) && pcmk__str_eq(attr, PCMK__NODE_ATTR_SHUTDOWN, pcmk__str_none)) { if (!pcmk__str_eq(value, "0", pcmk__str_null_matches)) { attrd_set_requesting_shutdown(); } else { attrd_clear_requesting_shutdown(); } } // Write out new value or start dampening timer if (a->timeout_ms && a->timer) { crm_trace("Delaying write of %s %s for dampening", attr, pcmk__readable_interval(a->timeout_ms)); mainloop_timer_start(a->timer); } else { attrd_write_or_elect_attribute(a); } } else { int is_force_write = 0; crm_element_value_int(xml, PCMK__XA_ATTRD_IS_FORCE_WRITE, &is_force_write); if (is_force_write == 1 && a->timeout_ms && a->timer) { /* Save forced writing and set change flag. */ /* The actual attribute is written by Writer after election. */ crm_trace("%s[%s] from %s is unchanged (%s), forcing write", attr, host, peer->name, pcmk__s(value, "unset")); attrd_set_attr_flags(a, attrd_attr_force_write); } else { crm_trace("%s[%s] from %s is unchanged (%s)", attr, host, peer->name, pcmk__s(value, "unset")); } } // This allows us to later detect local values that peer doesn't know about attrd_set_value_flags(v, attrd_value_from_peer); /* If this is a cluster node whose node ID we are learning, remember it */ if ((v->nodeid == 0) && !pcmk_is_set(v->flags, attrd_value_remote) && (crm_element_value_int(xml, PCMK__XA_ATTR_HOST_ID, (int*)&v->nodeid) == 0) && (v->nodeid > 0)) { record_peer_nodeid(v, host); } } static void attrd_peer_update_one(const pcmk__node_status_t *peer, xmlNode *xml, bool filter) { attribute_t *a = NULL; const char *attr = crm_element_value(xml, PCMK__XA_ATTR_NAME); const char *value = crm_element_value(xml, PCMK__XA_ATTR_VALUE); const char *host = crm_element_value(xml, PCMK__XA_ATTR_HOST); if (attr == NULL) { crm_warn("Could not update attribute: peer did not specify name"); return; } a = attrd_populate_attribute(xml, attr); if (a == NULL) { return; } if (host == NULL) { // If no host was specified, update all hosts GHashTableIter vIter; crm_debug("Setting %s for all hosts to %s", attr, value); pcmk__xe_remove_attr(xml, PCMK__XA_ATTR_HOST_ID); g_hash_table_iter_init(&vIter, a->values); while (g_hash_table_iter_next(&vIter, (gpointer *) & host, NULL)) { update_attr_on_host(a, peer, xml, attr, value, host, filter); } } else { // Update attribute value for the given host update_attr_on_host(a, peer, xml, attr, value, host, filter); } /* If this is a message from some attrd instance broadcasting its protocol * version, check to see if it's a new minimum version. */ if (pcmk__str_eq(attr, CRM_ATTR_PROTOCOL, pcmk__str_none)) { attrd_update_minimum_protocol_ver(peer->name, value); } } static void broadcast_unseen_local_values(void) { GHashTableIter aIter; GHashTableIter vIter; attribute_t *a = NULL; attribute_value_t *v = NULL; xmlNode *sync = NULL; g_hash_table_iter_init(&aIter, attributes); while (g_hash_table_iter_next(&aIter, NULL, (gpointer *) & a)) { g_hash_table_iter_init(&vIter, a->values); while (g_hash_table_iter_next(&vIter, NULL, (gpointer *) & v)) { if (!pcmk_is_set(v->flags, attrd_value_from_peer) && pcmk__str_eq(v->nodename, attrd_cluster->priv->node_name, pcmk__str_casei)) { crm_trace("* %s[%s]='%s' is local-only", a->id, v->nodename, readable_value(v)); if (sync == NULL) { sync = pcmk__xe_create(NULL, __func__); crm_xml_add(sync, PCMK_XA_TASK, PCMK__ATTRD_CMD_SYNC_RESPONSE); } attrd_add_value_xml(sync, a, v, a->timeout_ms && a->timer); } } } if (sync != NULL) { crm_debug("Broadcasting local-only values"); attrd_send_message(NULL, sync, false); pcmk__xml_free(sync); } } int attrd_cluster_connect(void) { int rc = pcmk_rc_ok; attrd_cluster = pcmk_cluster_new(); pcmk_cluster_set_destroy_fn(attrd_cluster, attrd_cpg_destroy); pcmk_cpg_set_deliver_fn(attrd_cluster, attrd_cpg_dispatch); pcmk_cpg_set_confchg_fn(attrd_cluster, pcmk__cpg_confchg_cb); pcmk__cluster_set_status_callback(&attrd_peer_change_cb); rc = pcmk_cluster_connect(attrd_cluster); rc = pcmk_rc2legacy(rc); if (rc != pcmk_ok) { crm_err("Cluster connection failed"); return rc; } return pcmk_ok; } void attrd_peer_clear_failure(pcmk__request_t *request) { xmlNode *xml = request->xml; const char *rsc = crm_element_value(xml, PCMK__XA_ATTR_RESOURCE); const char *host = crm_element_value(xml, PCMK__XA_ATTR_HOST); const char *op = crm_element_value(xml, PCMK__XA_ATTR_CLEAR_OPERATION); const char *interval_spec = crm_element_value(xml, PCMK__XA_ATTR_CLEAR_INTERVAL); guint interval_ms = 0U; char *attr = NULL; GHashTableIter iter; regex_t regex; pcmk__node_status_t *peer = pcmk__get_node(0, request->peer, NULL, pcmk__node_search_cluster_member); pcmk_parse_interval_spec(interval_spec, &interval_ms); if (attrd_failure_regex(®ex, rsc, op, interval_ms) != pcmk_ok) { crm_info("Ignoring invalid request to clear failures for %s", pcmk__s(rsc, "all resources")); return; } crm_xml_add(xml, PCMK_XA_TASK, PCMK__ATTRD_CMD_UPDATE); /* Make sure value is not set, so we delete */ pcmk__xe_remove_attr(xml, PCMK__XA_ATTR_VALUE); g_hash_table_iter_init(&iter, attributes); while (g_hash_table_iter_next(&iter, (gpointer *) &attr, NULL)) { if (regexec(®ex, attr, 0, NULL, 0) == 0) { crm_trace("Matched %s when clearing %s", attr, pcmk__s(rsc, "all resources")); crm_xml_add(xml, PCMK__XA_ATTR_NAME, attr); attrd_peer_update(peer, xml, host, false); } } regfree(®ex); } /*! * \internal * \brief Load attributes from a peer sync response * * \param[in] peer Peer that sent sync response * \param[in] peer_won Whether peer is the attribute writer * \param[in,out] xml Request XML */ void attrd_peer_sync_response(const pcmk__node_status_t *peer, bool peer_won, xmlNode *xml) { crm_info("Processing " PCMK__ATTRD_CMD_SYNC_RESPONSE " from %s", peer->name); if (peer_won) { /* Initialize the "seen" flag for all attributes to cleared, so we can * detect attributes that local node has but the writer doesn't. */ attrd_clear_value_seen(); } // Process each attribute update in the sync response for (xmlNode *child = pcmk__xe_first_child(xml, NULL, NULL, NULL); child != NULL; child = pcmk__xe_next(child, NULL)) { attrd_peer_update(peer, child, crm_element_value(child, PCMK__XA_ATTR_HOST), true); } if (peer_won) { /* If any attributes are still not marked as seen, the writer doesn't * know about them, so send all peers an update with them. */ broadcast_unseen_local_values(); } } /*! * \internal * \brief Remove all attributes and optionally peer cache entries for a node * * \param[in] host Name of node to purge * \param[in] uncache If true, remove node from peer caches * \param[in] source Who requested removal (only used for logging) */ void attrd_peer_remove(const char *host, bool uncache, const char *source) { attribute_t *a = NULL; GHashTableIter aIter; CRM_CHECK(host != NULL, return); crm_notice("Removing all %s attributes for node %s " QB_XS " %s reaping node from cache", host, source, (uncache? "and" : "without")); g_hash_table_iter_init(&aIter, attributes); while (g_hash_table_iter_next(&aIter, NULL, (gpointer *) & a)) { if(g_hash_table_remove(a->values, host)) { crm_debug("Removed %s[%s] for peer %s", a->id, host, source); } } if (uncache) { pcmk__purge_node_from_cache(host, 0); } } /*! * \internal * \brief Send all known attributes and values to a peer * * \param[in] peer Peer to send sync to (if NULL, broadcast to all peers) */ void attrd_peer_sync(pcmk__node_status_t *peer) { GHashTableIter aIter; GHashTableIter vIter; attribute_t *a = NULL; attribute_value_t *v = NULL; xmlNode *sync = pcmk__xe_create(NULL, __func__); crm_xml_add(sync, PCMK_XA_TASK, PCMK__ATTRD_CMD_SYNC_RESPONSE); g_hash_table_iter_init(&aIter, attributes); while (g_hash_table_iter_next(&aIter, NULL, (gpointer *) & a)) { g_hash_table_iter_init(&vIter, a->values); while (g_hash_table_iter_next(&vIter, NULL, (gpointer *) & v)) { crm_debug("Syncing %s[%s]='%s' to %s", a->id, v->nodename, readable_value(v), readable_peer(peer)); attrd_add_value_xml(sync, a, v, false); } } crm_debug("Syncing values to %s", readable_peer(peer)); attrd_send_message(peer, sync, false); pcmk__xml_free(sync); } void attrd_peer_update(const pcmk__node_status_t *peer, xmlNode *xml, const char *host, bool filter) { bool handle_sync_point = false; CRM_CHECK((peer != NULL) && (xml != NULL), return); if (xml->children != NULL) { for (xmlNode *child = pcmk__xe_first_child(xml, PCMK_XE_OP, NULL, NULL); child != NULL; child = pcmk__xe_next(child, PCMK_XE_OP)) { pcmk__xe_copy_attrs(child, xml, pcmk__xaf_no_overwrite); attrd_peer_update_one(peer, child, filter); if (attrd_request_has_sync_point(child)) { handle_sync_point = true; } } } else { attrd_peer_update_one(peer, xml, filter); if (attrd_request_has_sync_point(xml)) { handle_sync_point = true; } } /* If the update XML specified that the client wanted to wait for a sync * point, process that now. */ if (handle_sync_point) { crm_trace("Hit local sync point for attribute update"); attrd_ack_waitlist_clients(attrd_sync_point_local, xml); } }