diff --git a/daemons/attrd/attrd_alerts.c b/daemons/attrd/attrd_alerts.c index 32d808a22f..dffc8e1bc0 100644 --- a/daemons/attrd/attrd_alerts.c +++ b/daemons/attrd/attrd_alerts.c @@ -1,155 +1,137 @@ /* * Copyright 2015-2023 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 "pacemaker-attrd.h" static GList *attrd_alert_list = NULL; static void attrd_lrmd_callback(lrmd_event_data_t * op) { CRM_CHECK(op != NULL, return); switch (op->type) { case lrmd_event_disconnect: crm_info("Lost connection to executor"); attrd_lrmd_disconnect(); break; default: break; } } static lrmd_t * attrd_lrmd_connect(void) { if (the_lrmd == NULL) { the_lrmd = lrmd_api_new(); the_lrmd->cmds->set_callback(the_lrmd, attrd_lrmd_callback); } if (!the_lrmd->cmds->is_connected(the_lrmd)) { const unsigned int max_attempts = 10; int ret = -ENOTCONN; for (int fails = 0; fails < max_attempts; ++fails) { ret = the_lrmd->cmds->connect(the_lrmd, T_ATTRD, NULL); if (ret == pcmk_ok) { break; } crm_debug("Could not connect to executor, %d tries remaining", (max_attempts - fails)); /* @TODO We don't want to block here with sleep, but we should wait * some time between connection attempts. We could possibly add a * timer with a callback, but then we'd likely need an alert queue. */ } if (ret != pcmk_ok) { attrd_lrmd_disconnect(); } } return the_lrmd; } void attrd_lrmd_disconnect(void) { if (the_lrmd) { lrmd_t *conn = the_lrmd; the_lrmd = NULL; /* in case we're called recursively */ lrmd_api_delete(conn); /* will disconnect if necessary */ } } static void config_query_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { xmlNode *crmalerts = NULL; if (rc == -ENXIO) { crm_debug("Local CIB has no alerts section"); return; } else if (rc != pcmk_ok) { crm_notice("Could not query local CIB: %s", pcmk_strerror(rc)); return; } crmalerts = output; if ((crmalerts != NULL) && !pcmk__xe_is(crmalerts, XML_CIB_TAG_ALERTS)) { crmalerts = first_named_child(crmalerts, XML_CIB_TAG_ALERTS); } if (!crmalerts) { crm_notice("CIB query result has no " XML_CIB_TAG_ALERTS " section"); return; } pe_free_alert_list(attrd_alert_list); attrd_alert_list = pe_unpack_alerts(crmalerts); } #define XPATH_ALERTS \ "/" XML_TAG_CIB "/" XML_CIB_TAG_CONFIGURATION "/" XML_CIB_TAG_ALERTS gboolean attrd_read_options(gpointer user_data) { int call_id; CRM_CHECK(the_cib != NULL, return TRUE); call_id = the_cib->cmds->query(the_cib, XPATH_ALERTS, NULL, cib_xpath | cib_scope_local); the_cib->cmds->register_callback_full(the_cib, call_id, 120, FALSE, NULL, "config_query_callback", config_query_callback, free); crm_trace("Querying the CIB... call %d", call_id); return TRUE; } -void -attrd_cib_updated_cb(const char *event, xmlNode * msg) -{ - const xmlNode *patchset = NULL; - - if (attrd_shutting_down(true)) { - return; - } - - if (cib__get_notify_patchset(msg, &patchset) != pcmk_rc_ok) { - return; - } - - if (cib__element_in_patchset(patchset, XML_CIB_TAG_ALERTS)) { - mainloop_set_trigger(attrd_config_read); - } -} - int attrd_send_attribute_alert(const char *node, int nodeid, const char *attr, const char *value) { if (attrd_alert_list == NULL) { return pcmk_ok; } return lrmd_send_attribute_alert(attrd_lrmd_connect(), attrd_alert_list, node, nodeid, attr, value); } diff --git a/daemons/attrd/attrd_cib.c b/daemons/attrd/attrd_cib.c index 2c910b4c64..123ee1f1ba 100644 --- a/daemons/attrd/attrd_cib.c +++ b/daemons/attrd/attrd_cib.c @@ -1,397 +1,588 @@ /* * Copyright 2013-2023 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 "pacemaker-attrd.h" static int last_cib_op_done = 0; +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_replaced_cb(const char *event, xmlNode *msg) +{ + int change_section = cib_change_section_nodes + |cib_change_section_status + |cib_change_section_alerts; + + if (attrd_shutting_down(true)) { + return; + } + + crm_element_value_int(msg, F_CIB_CHANGE_SECTION, &change_section); + + if (attrd_election_won()) { + if (pcmk_any_flags_set(change_section, + cib_change_section_nodes + |cib_change_section_status)) { + + crm_notice("Updating all attributes after %s event", event); + attrd_write_attributes(attrd_write_all); + } + } + + if (pcmk_is_set(change_section, cib_change_section_alerts)) { + // Check for changes in alerts + mainloop_set_trigger(attrd_config_read); + } +} + +static void +attrd_cib_updated_cb(const char *event, xmlNode *msg) +{ + const xmlNode *patchset = NULL; + + if (attrd_shutting_down(true)) { + return; + } + + if (cib__get_notify_patchset(msg, &patchset) != pcmk_rc_ok) { + return; + } + + if (cib__element_in_patchset(patchset, XML_CIB_TAG_ALERTS)) { + mainloop_set_trigger(attrd_config_read); + } +} + +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, T_ATTRD, cib_command); + + } while ((rc != pcmk_ok) && (attempts < max_retry)); + + if (rc != pcmk_ok) { + crm_err("Connection to the CIB manager failed: %s " CRM_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, T_CIB_REPLACE_NOTIFY, + attrd_cib_replaced_cb); + if (rc != pcmk_ok) { + crm_err("Could not set CIB notification callback"); + goto cleanup; + } + + rc = the_cib->cmds->add_notify_callback(the_cib, T_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, T_CIB_REPLACE_NOTIFY, + attrd_cib_replaced_cb); + the_cib->cmds->del_notify_callback(the_cib, T_CIB_DIFF_NOTIFY, + attrd_cib_updated_cb); + cib__clean_up_connection(&the_cib); +} + +static void +attrd_erase_cb(xmlNode *msg, int call_id, int rc, xmlNode *output, + void *user_data) +{ + do_crm_log_unlikely(((rc != pcmk_ok)? LOG_NOTICE : LOG_DEBUG), + "Cleared transient attributes: %s " + CRM_XS " xpath=%s rc=%d", + pcmk_strerror(rc), (char *) user_data, rc); +} + +#define XPATH_TRANSIENT "//node_state[@uname='%s']/" XML_TAG_TRANSIENT_NODEATTRS + +/*! + * \internal + * \brief Wipe all transient attributes for this node from the CIB + * + * Clear any previous transient node attributes from the CIB. 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 pacemaker-attrd respawns after crashing (see PCMK_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(). + */ +static void +attrd_erase_attrs(void) +{ + int call_id = 0; + char *xpath = crm_strdup_printf(XPATH_TRANSIENT, attrd_cluster->uname); + + crm_info("Clearing transient attributes from CIB " CRM_XS " xpath=%s", + xpath); + + call_id = the_cib->cmds->remove(the_cib, xpath, NULL, cib_xpath); + the_cib->cmds->register_callback_full(the_cib, call_id, 120, FALSE, xpath, + "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, wipe the CIB to match + attrd_erase_attrs(); + + // 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 " CRM_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)) { do_crm_log(level, "* %s[%s]=%s", a->id, peer, v->requested); free(v->requested); v->requested = NULL; if (rc != pcmk_ok) { a->changed = true; /* Attempt write out again */ } } if (a->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. */ attrd_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); } } } static void build_update_element(xmlNode *parent, attribute_t *a, const char *nodeid, const char *value) { const char *set = NULL; xmlNode *xml_obj = NULL; xml_obj = create_xml_node(parent, XML_CIB_TAG_STATE); crm_xml_add(xml_obj, XML_ATTR_ID, nodeid); xml_obj = create_xml_node(xml_obj, XML_TAG_TRANSIENT_NODEATTRS); crm_xml_add(xml_obj, XML_ATTR_ID, nodeid); if (pcmk__str_eq(a->set_type, XML_TAG_ATTR_SETS, pcmk__str_null_matches)) { xml_obj = create_xml_node(xml_obj, XML_TAG_ATTR_SETS); } else if (pcmk__str_eq(a->set_type, XML_TAG_UTILIZATION, pcmk__str_none)) { xml_obj = create_xml_node(xml_obj, XML_TAG_UTILIZATION); } else { crm_err("Unknown set type attribute: %s", a->set_type); } if (a->set_id) { crm_xml_set_id(xml_obj, "%s", a->set_id); } else { crm_xml_set_id(xml_obj, "%s-%s", XML_CIB_TAG_STATUS, nodeid); } set = ID(xml_obj); xml_obj = create_xml_node(xml_obj, XML_CIB_TAG_NVPAIR); if (a->uuid) { crm_xml_set_id(xml_obj, "%s", a->uuid); } else { crm_xml_set_id(xml_obj, "%s-%s", set, a->id); } crm_xml_add(xml_obj, XML_NVPAIR_ATTR_NAME, a->id); if(value) { crm_xml_add(xml_obj, XML_NVPAIR_ATTR_VALUE, value); } else { crm_xml_add(xml_obj, XML_NVPAIR_ATTR_VALUE, ""); crm_xml_add(xml_obj, "__delete__", XML_NVPAIR_ATTR_VALUE); } } 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 = NULL; a_v = calloc(1, sizeof(attribute_value_t)); CRM_ASSERT(a_v != NULL); a_v->nodeid = v->nodeid; a_v->nodename = strdup(v->nodename); pcmk__str_update(&a_v->current, 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); } void attrd_write_attribute(attribute_t *a, bool ignore_delay) { int private_updates = 0, cib_updates = 0; xmlNode *xml_top = NULL; attribute_value_t *v = NULL; GHashTableIter iter; enum cib_call_options flags = cib_none; GHashTable *alert_attribute_value = NULL; if (a == NULL) { return; } /* If this attribute will be written to the CIB ... */ if (!stand_alone && !a->is_private) { /* Defer the write if now's not a good time */ CRM_CHECK(the_cib != NULL, return); 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); return; } else if (mainloop_timer_running(a->timer)) { if (ignore_delay) { /* 'refresh' forces a write of the current value of all attributes * Cancel any existing timers, we're writing it NOW */ mainloop_timer_stop(a->timer); crm_debug("Write out of '%s': timer is running but ignore delay", a->id); } else { crm_info("Write out of '%s' delayed: timer is running", a->id); return; } } /* Initialize the status update XML */ xml_top = create_xml_node(NULL, XML_CIB_TAG_STATUS); } /* Attribute will be written shortly, so clear changed flag */ a->changed = false; /* We will check all peers' uuids shortly, so initialize this to false */ a->unknown_peer_uuids = false; /* Attribute will be written shortly, so clear forced write flag */ a->force_write = FALSE; /* 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)) { crm_node_t *peer = crm_get_peer_full(v->nodeid, v->nodename, CRM_GET_PEER_ANY); /* If the value's peer info does not correspond to a peer, ignore it */ if (peer == NULL) { crm_notice("Cannot update %s[%s]=%s because peer not known", a->id, v->nodename, v->current); continue; } /* If we're just learning the peer's node id, remember it */ if (peer->id && (v->nodeid == 0)) { crm_trace("Learned ID %u for node %s", peer->id, v->nodename); v->nodeid = peer->id; } /* If this is a private attribute, no update needs to be sent */ if (stand_alone || a->is_private) { private_updates++; continue; } /* If the peer is found, but its uuid is unknown, defer write */ if (peer->uuid == NULL) { a->unknown_peer_uuids = true; crm_notice("Cannot update %s[%s]=%s because peer UUID not known " "(will retry if learned)", a->id, v->nodename, v->current); continue; } /* Add this value to status update XML */ crm_debug("Updating %s[%s]=%s (peer known as %s, UUID %s, ID %u/%u)", a->id, v->nodename, v->current, peer->uname, peer->uuid, peer->id, v->nodeid); build_update_element(xml_top, a, peer->uuid, v->current); cib_updates++; /* Preservation of the attribute to transmit alert */ set_alert_attribute_value(alert_attribute_value, v); free(v->requested); v->requested = NULL; if (v->current) { v->requested = strdup(v->current); } else { /* Older attrd versions don't know about the cib_mixed_update * flag so make sure it goes to the local cib which does */ cib__set_call_options(flags, crm_system_name, cib_mixed_update|cib_scope_local); } } if (private_updates) { crm_info("Processed %d private change%s for %s, id=%s, set=%s", private_updates, pcmk__plural_s(private_updates), a->id, pcmk__s(a->uuid, "n/a"), pcmk__s(a->set_id, "n/a")); } if (cib_updates) { crm_log_xml_trace(xml_top, __func__); a->update = cib_internal_op(the_cib, PCMK__CIB_REQUEST_MODIFY, NULL, XML_CIB_TAG_STATUS, xml_top, NULL, flags, a->user); crm_info("Sent CIB request %d with %d change%s for %s (id %s, set %s)", a->update, cib_updates, pcmk__plural_s(cib_updates), a->id, pcmk__s(a->uuid, "n/a"), pcmk__s(a->set_id, "n/a")); the_cib->cmds->register_callback_full(the_cib, a->update, CIB_OP_TIMEOUT_S, FALSE, strdup(a->id), "attrd_cib_callback", attrd_cib_callback, free); /* Transmit alert of the attribute */ send_alert_attributes_value(a, alert_attribute_value); } g_hash_table_destroy(alert_attribute_value); free_xml(xml_top); } /*! * \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_skip_shutdown) && pcmk__str_eq(a->id, XML_CIB_ATTR_SHUTDOWN, pcmk__str_none)) { continue; } if (!pcmk_is_set(options, attrd_write_all) && a->unknown_peer_uuids) { // Try writing this attribute again, in case peer ID was learned a->changed = true; } else if (a->force_write) { /* If the force_write flag is set, write the attribute. */ a->changed = true; } if (pcmk_is_set(options, attrd_write_all) || a->changed) { bool ignore_delay = pcmk_is_set(options, attrd_write_no_delay); if (a->force_write) { // Always ignore delay when forced write flag is set ignore_delay = true; } attrd_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()) { attrd_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 49631df6e4..86dc67b049 100644 --- a/daemons/attrd/attrd_corosync.c +++ b/daemons/attrd/attrd_corosync.c @@ -1,619 +1,617 @@ /* * Copyright 2013-2023 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" -extern crm_exit_t attrd_exit_status; - static xmlNode * attrd_confirmation(int callid) { xmlNode *node = create_xml_node(NULL, __func__); crm_xml_add(node, F_TYPE, T_ATTRD); crm_xml_add(node, F_ORIG, get_local_node_name()); crm_xml_add(node, PCMK__XA_TASK, PCMK__ATTRD_CMD_CONFIRM); crm_xml_add_int(node, XML_LRM_ATTR_CALLID, callid); return node; } static void attrd_peer_message(crm_node_t *peer, xmlNode *xml) { const char *election_op = crm_element_value(xml, F_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->uname, .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, XML_LRM_ATTR_CALLID, &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->uname); attrd_send_message(peer, reply, false); free_xml(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) { uint32_t kind = 0; xmlNode *xml = NULL; const char *from = NULL; char *data = pcmk_message_common_cs(handle, nodeid, pid, msg, &kind, &from); if(data == NULL) { return; } if (kind == crm_class_cluster) { xml = string2xml(data); } if (xml == NULL) { crm_err("Bad message of class %d received from %s[%u]: '%.120s'", kind, from, nodeid, data); } else { crm_node_t *peer = crm_get_peer(nodeid, from); attrd_peer_message(peer, xml); } free_xml(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 Override an attribute sync with a local value * * Broadcast the local node's value for an attribute that's different 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. * * \param[in] a Attribute entry to override * * \return Local instance of attribute value */ static attribute_value_t * broadcast_local_value(const attribute_t *a) { attribute_value_t *v = g_hash_table_lookup(a->values, attrd_cluster->uname); xmlNode *sync = create_xml_node(NULL, __func__); crm_xml_add(sync, PCMK__XA_TASK, PCMK__ATTRD_CMD_SYNC_RESPONSE); attrd_add_value_xml(sync, a, v, false); attrd_send_message(NULL, sync, false); free_xml(sync); return v; } /*! * \internal * \brief Ensure a Pacemaker Remote node is in the correct peer cache * * \param[in] node_name Name of Pacemaker Remote node to check */ static void cache_remote_node(const char *node_name) { /* If we previously assumed this node was an unseen cluster node, * remove its entry from the cluster peer cache. */ crm_node_t *dup = pcmk__search_cluster_node_cache(0, node_name, NULL); if (dup && (dup->uuid == NULL)) { reap_crm_member(0, node_name); } // Ensure node is in the remote peer cache CRM_ASSERT(crm_remote_peer_get(node_name) != NULL); } #define state_text(state) pcmk__s((state), "in unknown state") /*! * \internal * \brief Return host's hash table entry (creating one if needed) * * \param[in,out] values Hash table of values * \param[in] host Name of peer to look up * \param[in] xml XML describing the attribute * * \return Pointer to new or existing hash table entry */ static attribute_value_t * attrd_lookup_or_create_value(GHashTable *values, const char *host, const xmlNode *xml) { attribute_value_t *v = g_hash_table_lookup(values, host); int is_remote = 0; crm_element_value_int(xml, PCMK__XA_ATTR_IS_REMOTE, &is_remote); if (is_remote) { cache_remote_node(host); } if (v == NULL) { v = calloc(1, sizeof(attribute_value_t)); CRM_ASSERT(v != NULL); pcmk__str_update(&v->nodename, host); v->is_remote = is_remote; g_hash_table_replace(values, v->nodename, v); } return(v); } static void attrd_peer_change_cb(enum crm_status_type kind, crm_node_t *peer, const void *data) { bool gone = false; bool is_remote = pcmk_is_set(peer->flags, crm_remote_node); switch (kind) { case crm_status_uname: crm_debug("%s node %s is now %s", (is_remote? "Remote" : "Cluster"), peer->uname, state_text(peer->state)); break; case crm_status_processes: if (!pcmk_is_set(peer->processes, crm_get_cluster_proc())) { gone = true; } crm_debug("Node %s is %s a peer", peer->uname, (gone? "no longer" : "now")); break; case crm_status_nstate: crm_debug("%s node %s is now %s (was %s)", (is_remote? "Remote" : "Cluster"), peer->uname, state_text(peer->state), state_text(data)); if (pcmk__str_eq(peer->state, CRM_NODE_MEMBER, pcmk__str_casei)) { /* 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, crm_remote_node)) { attrd_peer_sync(peer, NULL); } } else { // Remove all attribute values associated with lost nodes attrd_peer_remove(peer->uname, 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->uname); attrd_do_not_expect_from_peer(peer->uname); // Ensure remote nodes that come up are in the remote node cache } else if (!gone && is_remote) { cache_remote_node(peer->uname); } } static void record_peer_nodeid(attribute_value_t *v, const char *host) { crm_node_t *known_peer = crm_get_peer(v->nodeid, host); crm_trace("Learned %s has node id %s", known_peer->uname, known_peer->uuid); if (attrd_election_won()) { attrd_write_attributes(attrd_write_changed); } } static void update_attr_on_host(attribute_t *a, const crm_node_t *peer, const xmlNode *xml, const char *attr, const char *value, const char *host, bool filter, int is_force_write) { attribute_value_t *v = NULL; v = attrd_lookup_or_create_value(a->values, host, xml); if (filter && !pcmk__str_eq(v->current, value, pcmk__str_casei) && pcmk__str_eq(host, attrd_cluster->uname, pcmk__str_casei)) { crm_notice("%s[%s]: local value '%s' takes priority over '%s' from %s", attr, host, v->current, value, peer->uname); v = broadcast_local_value(a); } else if (!pcmk__str_eq(v->current, value, pcmk__str_casei)) { crm_notice("Setting %s[%s]%s%s: %s -> %s " CRM_XS " from %s with %s write delay", attr, host, a->set_type ? " in " : "", pcmk__s(a->set_type, ""), pcmk__s(v->current, "(unset)"), pcmk__s(value, "(unset)"), peer->uname, (a->timeout_ms == 0)? "no" : pcmk__readable_interval(a->timeout_ms)); pcmk__str_update(&v->current, value); a->changed = true; if (pcmk__str_eq(host, attrd_cluster->uname, pcmk__str_casei) && pcmk__str_eq(attr, XML_CIB_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("Delayed write out (%dms) for %s", a->timeout_ms, attr); mainloop_timer_start(a->timer); } else { attrd_write_or_elect_attribute(a); } } else { 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("Unchanged %s[%s] from %s is %s(Set the forced write flag)", attr, host, peer->uname, value); a->force_write = TRUE; } else { crm_trace("Unchanged %s[%s] from %s is %s", attr, host, peer->uname, value); } } /* Set the seen flag for attribute processing held only in the own node. */ v->seen = TRUE; /* If this is a cluster node whose node ID we are learning, remember it */ if ((v->nodeid == 0) && (v->is_remote == FALSE) && (crm_element_value_int(xml, PCMK__XA_ATTR_NODE_ID, (int*)&v->nodeid) == 0) && (v->nodeid > 0)) { record_peer_nodeid(v, host); } } static void attrd_peer_update_one(const crm_node_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_NODE_NAME); int is_force_write = 0; if (attr == NULL) { crm_warn("Could not update attribute: peer did not specify name"); return; } crm_element_value_int(xml, PCMK__XA_ATTR_FORCE, &is_force_write); 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); xml_remove_prop(xml, PCMK__XA_ATTR_NODE_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, is_force_write); } } else { // Update attribute value for the given host update_attr_on_host(a, peer, xml, attr, value, host, filter, is_force_write); } /* 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->uname, 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 (!(v->seen) && pcmk__str_eq(v->nodename, attrd_cluster->uname, pcmk__str_casei)) { if (sync == NULL) { sync = create_xml_node(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); free_xml(sync); } } int attrd_cluster_connect(void) { attrd_cluster = pcmk_cluster_new(); attrd_cluster->destroy = attrd_cpg_destroy; attrd_cluster->cpg.cpg_deliver_fn = attrd_cpg_dispatch; attrd_cluster->cpg.cpg_confchg_fn = pcmk_cpg_membership; crm_set_status_callback(&attrd_peer_change_cb); if (crm_cluster_connect(attrd_cluster) == FALSE) { crm_err("Cluster connection failed"); return -ENOTCONN; } 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_NODE_NAME); const char *op = crm_element_value(xml, PCMK__XA_ATTR_OPERATION); const char *interval_spec = crm_element_value(xml, PCMK__XA_ATTR_INTERVAL); guint interval_ms = crm_parse_interval_spec(interval_spec); char *attr = NULL; GHashTableIter iter; regex_t regex; crm_node_t *peer = crm_get_peer(0, request->peer); 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 */ xml_remove_prop(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 clear request * \param[in] peer_won Whether peer is the attribute writer * \param[in,out] xml Request XML */ void attrd_peer_sync_response(const crm_node_t *peer, bool peer_won, xmlNode *xml) { crm_info("Processing " PCMK__ATTRD_CMD_SYNC_RESPONSE " from %s", peer->uname); 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__xml_first_child(xml); child != NULL; child = pcmk__xml_next(child)) { attrd_peer_update(peer, child, crm_element_value(child, PCMK__XA_ATTR_NODE_NAME), 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 peer %s", host, source); 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) { crm_remote_peer_cache_remove(host); reap_crm_member(0, host); } } void attrd_peer_sync(crm_node_t *peer, xmlNode *xml) { GHashTableIter aIter; GHashTableIter vIter; attribute_t *a = NULL; attribute_value_t *v = NULL; xmlNode *sync = create_xml_node(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, v->current, peer?peer->uname:"everyone"); attrd_add_value_xml(sync, a, v, false); } } crm_debug("Syncing values to %s", peer?peer->uname:"everyone"); attrd_send_message(peer, sync, false); free_xml(sync); } void attrd_peer_update(const crm_node_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 = first_named_child(xml, XML_ATTR_OP); child != NULL; child = crm_next_same_xml(child)) { attrd_copy_xml_attributes(xml, child); 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); } } diff --git a/daemons/attrd/attrd_utils.c b/daemons/attrd/attrd_utils.c index bfd5136889..341ee1aed0 100644 --- a/daemons/attrd/attrd_utils.c +++ b/daemons/attrd/attrd_utils.c @@ -1,358 +1,325 @@ /* * Copyright 2004-2023 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" cib_t *the_cib = NULL; static bool requesting_shutdown = false; static bool shutting_down = false; static GMainLoop *mloop = NULL; /* A hash table storing information on the protocol version of each peer attrd. * The key is the peer's uname, and the value is the protocol version number. */ GHashTable *peer_protocol_vers = NULL; /*! * \internal * \brief Set requesting_shutdown state */ void attrd_set_requesting_shutdown(void) { requesting_shutdown = true; } /*! * \internal * \brief Clear requesting_shutdown state */ void attrd_clear_requesting_shutdown(void) { requesting_shutdown = false; } /*! * \internal * \brief Check whether local attribute manager is shutting down * * \param[in] if_requested Also consider presence of "shutdown" attribute * * \return \c true if local attribute manager has begun shutdown sequence * or (if \p if_requested is \c true) whether local node has a nonzero * "shutdown" attribute set, otherwise \c false * \note Most callers should pass \c false for \p if_requested, because the * attribute manager needs to continue performing while the controller is * shutting down, and even needs to be eligible for election in case all * nodes are shutting down. */ bool attrd_shutting_down(bool if_requested) { return shutting_down || (if_requested && requesting_shutdown); } /*! * \internal * \brief Exit (using mainloop or not, as appropriate) * * \param[in] nsig Ignored */ void attrd_shutdown(int nsig) { // Tell various functions not to do anthing shutting_down = true; // Don't respond to signals while shutting down mainloop_destroy_signal(SIGTERM); mainloop_destroy_signal(SIGCHLD); mainloop_destroy_signal(SIGPIPE); mainloop_destroy_signal(SIGUSR1); mainloop_destroy_signal(SIGUSR2); mainloop_destroy_signal(SIGTRAP); attrd_free_waitlist(); attrd_free_confirmations(); if (peer_protocol_vers != NULL) { g_hash_table_destroy(peer_protocol_vers); peer_protocol_vers = NULL; } if ((mloop == NULL) || !g_main_loop_is_running(mloop)) { /* If there's no main loop active, just exit. This should be possible * only if we get SIGTERM in brief windows at start-up and shutdown. */ crm_exit(CRM_EX_OK); } else { g_main_loop_quit(mloop); g_main_loop_unref(mloop); } } /*! * \internal * \brief Create a main loop for attrd */ void attrd_init_mainloop(void) { mloop = g_main_loop_new(NULL, FALSE); } /*! * \internal * \brief Run attrd main loop */ void attrd_run_mainloop(void) { g_main_loop_run(mloop); } -void -attrd_cib_disconnect(void) -{ - CRM_CHECK(the_cib != NULL, return); - the_cib->cmds->del_notify_callback(the_cib, T_CIB_REPLACE_NOTIFY, attrd_cib_replaced_cb); - the_cib->cmds->del_notify_callback(the_cib, T_CIB_DIFF_NOTIFY, attrd_cib_updated_cb); - cib__clean_up_connection(&the_cib); -} - -void -attrd_cib_replaced_cb(const char *event, xmlNode * msg) -{ - int change_section = cib_change_section_nodes | cib_change_section_status | cib_change_section_alerts; - - if (attrd_shutting_down(true)) { - return; - } - - crm_element_value_int(msg, F_CIB_CHANGE_SECTION, &change_section); - - if (attrd_election_won()) { - if (change_section & (cib_change_section_nodes | cib_change_section_status)) { - crm_notice("Updating all attributes after %s event", event); - attrd_write_attributes(attrd_write_all); - } - } - - if (change_section & cib_change_section_alerts) { - // Check for changes in alerts - mainloop_set_trigger(attrd_config_read); - } -} - /* strlen("value") */ #define plus_plus_len (5) /*! * \internal * \brief Check whether an attribute value should be expanded * * \param[in] value Attribute value to check * * \return true if value needs expansion, false otherwise */ bool attrd_value_needs_expansion(const char *value) { return ((strlen(value) >= (plus_plus_len + 2)) && (value[plus_plus_len] == '+') && ((value[plus_plus_len + 1] == '+') || (value[plus_plus_len + 1] == '='))); } /*! * \internal * \brief Expand an increment expression into an integer * * \param[in] value Attribute increment expression to expand * \param[in] old_value Previous value of attribute * * \return Expanded value */ int attrd_expand_value(const char *value, const char *old_value) { int offset = 1; int int_value = char2score(old_value); if (value[plus_plus_len + 1] != '+') { const char *offset_s = value + (plus_plus_len + 2); offset = char2score(offset_s); } int_value += offset; if (int_value > INFINITY) { int_value = INFINITY; } return int_value; } /*! * \internal * \brief Create regular expression matching failure-related attributes * * \param[out] regex Where to store created regular expression * \param[in] rsc Name of resource to clear (or NULL for all) * \param[in] op Operation to clear if rsc is specified (or NULL for all) * \param[in] interval_ms Interval of operation to clear if op is specified * * \return pcmk_ok on success, -EINVAL if arguments are invalid * * \note The caller is responsible for freeing the result with regfree(). */ int attrd_failure_regex(regex_t *regex, const char *rsc, const char *op, guint interval_ms) { char *pattern = NULL; int rc; /* Create a pattern that matches desired attributes */ if (rsc == NULL) { pattern = strdup(ATTRD_RE_CLEAR_ALL); } else if (op == NULL) { pattern = crm_strdup_printf(ATTRD_RE_CLEAR_ONE, rsc); } else { pattern = crm_strdup_printf(ATTRD_RE_CLEAR_OP, rsc, op, interval_ms); } /* Compile pattern into regular expression */ crm_trace("Clearing attributes matching %s", pattern); rc = regcomp(regex, pattern, REG_EXTENDED|REG_NOSUB); free(pattern); return (rc == 0)? pcmk_ok : -EINVAL; } void attrd_free_attribute_value(gpointer data) { attribute_value_t *v = data; free(v->nodename); free(v->current); free(v->requested); free(v); } void attrd_free_attribute(gpointer data) { attribute_t *a = data; if(a) { free(a->id); free(a->set_id); free(a->set_type); free(a->uuid); free(a->user); mainloop_timer_del(a->timer); g_hash_table_destroy(a->values); free(a); } } /*! * \internal * \brief When a peer node leaves the cluster, stop tracking its protocol version. * * \param[in] host The peer node's uname to be removed */ void attrd_remove_peer_protocol_ver(const char *host) { if (peer_protocol_vers != NULL) { g_hash_table_remove(peer_protocol_vers, host); } } /*! * \internal * \brief When a peer node broadcasts a message with its protocol version, keep * track of that information. * * We keep track of each peer's protocol version so we know which peers to * expect confirmation messages from when handling cluster-wide sync points. * We additionally keep track of the lowest protocol version supported by all * peers so we know when we can send IPC messages containing more than one * request. * * \param[in] host The peer node's uname to be tracked * \param[in] value The peer node's protocol version */ void attrd_update_minimum_protocol_ver(const char *host, const char *value) { int ver; if (peer_protocol_vers == NULL) { peer_protocol_vers = pcmk__strkey_table(free, NULL); } pcmk__scan_min_int(value, &ver, 0); if (ver > 0) { char *host_name = strdup(host); /* Record the peer attrd's protocol version. */ CRM_ASSERT(host_name != NULL); g_hash_table_insert(peer_protocol_vers, host_name, GINT_TO_POINTER(ver)); /* If the protocol version is a new minimum, record it as such. */ if (minimum_protocol_version == -1 || ver < minimum_protocol_version) { minimum_protocol_version = ver; crm_trace("Set minimum attrd protocol version to %d", minimum_protocol_version); } } } void attrd_copy_xml_attributes(xmlNode *src, xmlNode *dest) { /* Copy attributes from the wrapper parent node into the child node. * We can't just use copy_in_properties because we want to skip any * attributes that are already set on the child. For instance, if * we were told to use a specific node, there will already be a node * attribute on the child. Copying the parent's node attribute over * could result in the wrong value. */ for (xmlAttrPtr a = pcmk__xe_first_attr(src); a != NULL; a = a->next) { const char *p_name = (const char *) a->name; const char *p_value = ((a == NULL) || (a->children == NULL)) ? NULL : (const char *) a->children->content; if (crm_element_value(dest, p_name) == NULL) { crm_xml_add(dest, p_name, p_value); } } } diff --git a/daemons/attrd/pacemaker-attrd.c b/daemons/attrd/pacemaker-attrd.c index 1ea89e1785..8091c5b1ac 100644 --- a/daemons/attrd/pacemaker-attrd.c +++ b/daemons/attrd/pacemaker-attrd.c @@ -1,360 +1,226 @@ /* * Copyright 2013-2023 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 #include #include #include #include #include #include #include #include #include #include "pacemaker-attrd.h" #define SUMMARY "daemon for managing Pacemaker node attributes" gboolean stand_alone = FALSE; gchar **log_files = NULL; static GOptionEntry entries[] = { { "stand-alone", 's', G_OPTION_FLAG_NONE, G_OPTION_ARG_NONE, &stand_alone, "(Advanced use only) Run in stand-alone mode", NULL }, { "logfile", 'l', G_OPTION_FLAG_NONE, G_OPTION_ARG_FILENAME_ARRAY, &log_files, "Send logs to the additional named logfile", NULL }, { NULL } }; static pcmk__output_t *out = NULL; static pcmk__supported_format_t formats[] = { PCMK__SUPPORTED_FORMAT_NONE, PCMK__SUPPORTED_FORMAT_TEXT, PCMK__SUPPORTED_FORMAT_XML, { NULL, NULL, NULL } }; lrmd_t *the_lrmd = NULL; crm_cluster_t *attrd_cluster = NULL; crm_trigger_t *attrd_config_read = NULL; crm_exit_t attrd_exit_status = CRM_EX_OK; -static void -attrd_cib_destroy_cb(gpointer user_data) -{ - cib_t *conn = user_data; - - conn->cmds->signoff(conn); /* Ensure IPC is cleaned up */ - - 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); - } - - return; -} - -static void -attrd_erase_cb(xmlNode *msg, int call_id, int rc, xmlNode *output, - void *user_data) -{ - do_crm_log_unlikely((rc? LOG_NOTICE : LOG_DEBUG), - "Cleared transient attributes: %s " - CRM_XS " xpath=%s rc=%d", - pcmk_strerror(rc), (char *) user_data, rc); -} - -#define XPATH_TRANSIENT "//node_state[@uname='%s']/" XML_TAG_TRANSIENT_NODEATTRS - -/*! - * \internal - * \brief Wipe all transient attributes for this node from the CIB - * - * Clear any previous transient node attributes from the CIB. 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 pacemaker-attrd respawns after crashing (see PCMK_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(). - */ -static void -attrd_erase_attrs(void) -{ - int call_id; - char *xpath = crm_strdup_printf(XPATH_TRANSIENT, attrd_cluster->uname); - - crm_info("Clearing transient attributes from CIB " CRM_XS " xpath=%s", - xpath); - - call_id = the_cib->cmds->remove(the_cib, xpath, NULL, cib_xpath); - the_cib->cmds->register_callback_full(the_cib, call_id, 120, FALSE, xpath, - "attrd_erase_cb", attrd_erase_cb, - free); -} - -static 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, T_ATTRD, cib_command); - - } while(rc != pcmk_ok && attempts < max_retry); - - if (rc != pcmk_ok) { - crm_err("Connection to the CIB manager failed: %s " CRM_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, T_CIB_REPLACE_NOTIFY, attrd_cib_replaced_cb); - if(rc != pcmk_ok) { - crm_err("Could not set CIB notification callback"); - goto cleanup; - } - - rc = the_cib->cmds->add_notify_callback(the_cib, T_CIB_DIFF_NOTIFY, attrd_cib_updated_cb); - if (rc != pcmk_ok) { - crm_err("Could not set CIB notification callback (update)"); - goto cleanup; - } - - return pcmk_ok; - - cleanup: - cib__clean_up_connection(&the_cib); - return -ENOTCONN; -} - -/*! - * \internal - * \brief Prepare the CIB after cluster is connected - */ -static void -attrd_cib_init(void) -{ - // We have no attribute values in memory, wipe the CIB to match - attrd_erase_attrs(); - - // 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 bool ipc_already_running(void) { pcmk_ipc_api_t *old_instance = NULL; int rc = pcmk_rc_ok; rc = pcmk_new_ipc_api(&old_instance, pcmk_ipc_attrd); if (rc != pcmk_rc_ok) { return false; } rc = pcmk__connect_ipc(old_instance, pcmk_ipc_dispatch_sync, 2); if (rc != pcmk_rc_ok) { crm_debug("No existing %s manager instance found: %s", pcmk_ipc_name(old_instance, true), pcmk_rc_str(rc)); pcmk_free_ipc_api(old_instance); return false; } pcmk_disconnect_ipc(old_instance); pcmk_free_ipc_api(old_instance); return true; } static GOptionContext * build_arg_context(pcmk__common_args_t *args, GOptionGroup **group) { GOptionContext *context = NULL; context = pcmk__build_arg_context(args, "text (default), xml", group, NULL); pcmk__add_main_args(context, entries); return context; } int main(int argc, char **argv) { int rc = pcmk_rc_ok; GError *error = NULL; bool initialized = false; GOptionGroup *output_group = NULL; pcmk__common_args_t *args = pcmk__new_common_args(SUMMARY); gchar **processed_args = pcmk__cmdline_preproc(argv, NULL); GOptionContext *context = build_arg_context(args, &output_group); attrd_init_mainloop(); crm_log_preinit(NULL, argc, argv); mainloop_add_signal(SIGTERM, attrd_shutdown); pcmk__register_formats(output_group, formats); if (!g_option_context_parse_strv(context, &processed_args, &error)) { attrd_exit_status = CRM_EX_USAGE; goto done; } rc = pcmk__output_new(&out, args->output_ty, args->output_dest, argv); if ((rc != pcmk_rc_ok) || (out == NULL)) { attrd_exit_status = CRM_EX_ERROR; g_set_error(&error, PCMK__EXITC_ERROR, attrd_exit_status, "Error creating output format %s: %s", args->output_ty, pcmk_rc_str(rc)); goto done; } if (args->version) { out->version(out, false); goto done; } // Open additional log files pcmk__add_logfiles(log_files, out); crm_log_init(T_ATTRD, LOG_INFO, TRUE, FALSE, argc, argv, FALSE); crm_notice("Starting Pacemaker node attribute manager%s", stand_alone ? " in standalone mode" : ""); if (ipc_already_running()) { const char *msg = "pacemaker-attrd is already active, aborting startup"; attrd_exit_status = CRM_EX_OK; g_set_error(&error, PCMK__EXITC_ERROR, attrd_exit_status, "%s", msg); crm_err("%s", msg); goto done; } initialized = true; attributes = pcmk__strkey_table(NULL, attrd_free_attribute); /* Connect to the CIB before connecting to the cluster or listening for IPC. * This allows us to assume the CIB is connected whenever we process a * cluster or IPC message (which also avoids start-up race conditions). */ if (!stand_alone) { if (attrd_cib_connect(30) != pcmk_ok) { attrd_exit_status = CRM_EX_FATAL; g_set_error(&error, PCMK__EXITC_ERROR, attrd_exit_status, "Could not connect to the CIB"); goto done; } crm_info("CIB connection active"); } if (attrd_cluster_connect() != pcmk_ok) { attrd_exit_status = CRM_EX_FATAL; g_set_error(&error, PCMK__EXITC_ERROR, attrd_exit_status, "Could not connect to the cluster"); goto done; } crm_info("Cluster connection active"); // Initialization that requires the cluster to be connected attrd_election_init(); if (!stand_alone) { attrd_cib_init(); } /* Set a private attribute for ourselves with the protocol version we * support. This lets all nodes determine the minimum supported version * across all nodes. It also ensures that the writer learns our node name, * so it can send our attributes to the CIB. */ attrd_broadcast_protocol(); attrd_init_ipc(); crm_notice("Pacemaker node attribute manager successfully started and accepting connections"); attrd_run_mainloop(); done: if (initialized) { crm_info("Shutting down attribute manager"); attrd_election_fini(); attrd_ipc_fini(); attrd_lrmd_disconnect(); if (!stand_alone) { attrd_cib_disconnect(); } attrd_free_waitlist(); pcmk_cluster_free(attrd_cluster); g_hash_table_destroy(attributes); } g_strfreev(processed_args); pcmk__free_arg_context(context); g_strfreev(log_files); pcmk__output_and_clear_error(&error, out); if (out != NULL) { out->finish(out, attrd_exit_status, true, NULL); pcmk__output_free(out); } pcmk__unregister_formats(); crm_exit(attrd_exit_status); } diff --git a/daemons/attrd/pacemaker-attrd.h b/daemons/attrd/pacemaker-attrd.h index 2e35bd7ec5..e3c369b5bc 100644 --- a/daemons/attrd/pacemaker-attrd.h +++ b/daemons/attrd/pacemaker-attrd.h @@ -1,222 +1,223 @@ /* * Copyright 2013-2023 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. */ #ifndef PACEMAKER_ATTRD__H # define PACEMAKER_ATTRD__H #include #include #include #include #include #include #include /* * Legacy attrd (all pre-1.1.11 Pacemaker versions, plus all versions when used * with the no-longer-supported CMAN or corosync-plugin stacks) is unversioned. * * With atomic attrd, each attrd will send ATTRD_PROTOCOL_VERSION with every * peer request and reply. As of Pacemaker 2.0.0, at start-up each attrd will * also set a private attribute for itself with its version, so any attrd can * determine the minimum version supported by all peers. * * Protocol Pacemaker Significant changes * -------- --------- ------------------- * 1 1.1.11 PCMK__ATTRD_CMD_UPDATE (PCMK__XA_ATTR_NAME only), * PCMK__ATTRD_CMD_PEER_REMOVE, PCMK__ATTRD_CMD_REFRESH, * PCMK__ATTRD_CMD_FLUSH, PCMK__ATTRD_CMD_SYNC, * PCMK__ATTRD_CMD_SYNC_RESPONSE * 1 1.1.13 PCMK__ATTRD_CMD_UPDATE (with PCMK__XA_ATTR_PATTERN), * PCMK__ATTRD_CMD_QUERY * 1 1.1.15 PCMK__ATTRD_CMD_UPDATE_BOTH, * PCMK__ATTRD_CMD_UPDATE_DELAY * 2 1.1.17 PCMK__ATTRD_CMD_CLEAR_FAILURE * 3 2.1.1 PCMK__ATTRD_CMD_SYNC_RESPONSE indicates remote nodes * 4 2.1.5 Multiple attributes can be updated in a single IPC * message * 5 2.1.5 Peers can request confirmation of a sent message */ #define ATTRD_PROTOCOL_VERSION "5" #define ATTRD_SUPPORTS_MULTI_MESSAGE(x) ((x) >= 4) #define ATTRD_SUPPORTS_CONFIRMATION(x) ((x) >= 5) #define attrd_send_ack(client, id, flags) \ pcmk__ipc_send_ack((client), (id), (flags), "ack", ATTRD_PROTOCOL_VERSION, CRM_EX_INDETERMINATE) void attrd_init_mainloop(void); void attrd_run_mainloop(void); void attrd_set_requesting_shutdown(void); void attrd_clear_requesting_shutdown(void); void attrd_free_waitlist(void); bool attrd_shutting_down(bool if_requested); void attrd_shutdown(int nsig); void attrd_init_ipc(void); void attrd_ipc_fini(void); +int attrd_cib_connect(int max_retry); void attrd_cib_disconnect(void); +void attrd_cib_init(void); bool attrd_value_needs_expansion(const char *value); int attrd_expand_value(const char *value, const char *old_value); /* regular expression to clear failures of all resources */ #define ATTRD_RE_CLEAR_ALL \ "^(" PCMK__FAIL_COUNT_PREFIX "|" PCMK__LAST_FAILURE_PREFIX ")-" /* regular expression to clear failure of all operations for one resource * (format takes resource name) * * @COMPAT attributes set < 1.1.17: * also match older attributes that do not have the operation part */ #define ATTRD_RE_CLEAR_ONE ATTRD_RE_CLEAR_ALL "%s(#.+_[0-9]+)?$" /* regular expression to clear failure of one operation for one resource * (format takes resource name, operation name, and interval) * * @COMPAT attributes set < 1.1.17: * also match older attributes that do not have the operation part */ #define ATTRD_RE_CLEAR_OP ATTRD_RE_CLEAR_ALL "%s(#%s_%u)?$" int attrd_failure_regex(regex_t *regex, const char *rsc, const char *op, guint interval_ms); extern cib_t *the_cib; +extern crm_exit_t attrd_exit_status; /* Alerts */ extern lrmd_t *the_lrmd; extern crm_trigger_t *attrd_config_read; void attrd_lrmd_disconnect(void); gboolean attrd_read_options(gpointer user_data); -void attrd_cib_replaced_cb(const char *event, xmlNode * msg); -void attrd_cib_updated_cb(const char *event, xmlNode *msg); int attrd_send_attribute_alert(const char *node, int nodeid, const char *attr, const char *value); // Elections void attrd_election_init(void); void attrd_election_fini(void); void attrd_start_election_if_needed(void); bool attrd_election_won(void); void attrd_handle_election_op(const crm_node_t *peer, xmlNode *xml); bool attrd_check_for_new_writer(const crm_node_t *peer, const xmlNode *xml); void attrd_declare_winner(void); void attrd_remove_voter(const crm_node_t *peer); void attrd_xml_add_writer(xmlNode *xml); typedef struct attribute_s { char *uuid; /* TODO: Remove if at all possible */ char *id; char *set_id; char *set_type; GHashTable *values; int update; int timeout_ms; /* TODO: refactor these three as a bitmask */ bool changed; /* whether attribute value has changed since last write */ bool unknown_peer_uuids; /* whether we know we're missing a peer uuid */ gboolean is_private; /* whether to keep this attribute out of the CIB */ mainloop_timer_t *timer; char *user; gboolean force_write; /* Flag for updating attribute by ignoring delay */ } attribute_t; typedef struct attribute_value_s { uint32_t nodeid; gboolean is_remote; char *nodename; char *current; char *requested; gboolean seen; } attribute_value_t; extern crm_cluster_t *attrd_cluster; extern GHashTable *attributes; extern GHashTable *peer_protocol_vers; #define CIB_OP_TIMEOUT_S 120 int attrd_cluster_connect(void); void attrd_peer_update(const crm_node_t *peer, xmlNode *xml, const char *host, bool filter); void attrd_peer_sync(crm_node_t *peer, xmlNode *xml); void attrd_peer_remove(const char *host, bool uncache, const char *source); void attrd_peer_clear_failure(pcmk__request_t *request); void attrd_peer_sync_response(const crm_node_t *peer, bool peer_won, xmlNode *xml); void attrd_broadcast_protocol(void); xmlNode *attrd_client_peer_remove(pcmk__request_t *request); xmlNode *attrd_client_clear_failure(pcmk__request_t *request); xmlNode *attrd_client_update(pcmk__request_t *request); xmlNode *attrd_client_refresh(pcmk__request_t *request); xmlNode *attrd_client_query(pcmk__request_t *request); gboolean attrd_send_message(crm_node_t *node, xmlNode *data, bool confirm); xmlNode *attrd_add_value_xml(xmlNode *parent, const attribute_t *a, const attribute_value_t *v, bool force_write); void attrd_clear_value_seen(void); void attrd_free_attribute(gpointer data); void attrd_free_attribute_value(gpointer data); attribute_t *attrd_populate_attribute(xmlNode *xml, const char *attr); enum attrd_write_options { attrd_write_changed = 0, attrd_write_all = (1 << 0), attrd_write_no_delay = (1 << 1), attrd_write_skip_shutdown = (1 << 2), }; void attrd_write_attribute(attribute_t *a, bool ignore_delay); void attrd_write_attributes(uint32_t options); void attrd_write_or_elect_attribute(attribute_t *a); extern int minimum_protocol_version; void attrd_remove_peer_protocol_ver(const char *host); void attrd_update_minimum_protocol_ver(const char *host, const char *value); mainloop_timer_t *attrd_add_timer(const char *id, int timeout_ms, attribute_t *attr); void attrd_unregister_handlers(void); void attrd_handle_request(pcmk__request_t *request); enum attrd_sync_point { attrd_sync_point_local, attrd_sync_point_cluster, }; typedef int (*attrd_confirmation_action_fn)(xmlNode *); void attrd_add_client_to_waitlist(pcmk__request_t *request); void attrd_ack_waitlist_clients(enum attrd_sync_point sync_point, const xmlNode *xml); int attrd_cluster_sync_point_update(xmlNode *xml); void attrd_do_not_expect_from_peer(const char *host); void attrd_do_not_wait_for_client(pcmk__client_t *client); void attrd_expect_confirmations(pcmk__request_t *request, attrd_confirmation_action_fn fn); void attrd_free_confirmations(void); void attrd_handle_confirmation(int callid, const char *host); void attrd_remove_client_from_waitlist(pcmk__client_t *client); const char *attrd_request_sync_point(xmlNode *xml); bool attrd_request_has_sync_point(xmlNode *xml); void attrd_copy_xml_attributes(xmlNode *src, xmlNode *dest); extern gboolean stand_alone; #endif /* PACEMAKER_ATTRD__H */