diff --git a/lib/pacemaker/pcmk_sched_probes.c b/lib/pacemaker/pcmk_sched_probes.c index 197c7ff3cc..08b077c9dc 100644 --- a/lib/pacemaker/pcmk_sched_probes.c +++ b/lib/pacemaker/pcmk_sched_probes.c @@ -1,823 +1,823 @@ /* * Copyright 2004-2022 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 "libpacemaker_private.h" /*! * \internal * \brief Add the expected result to a newly created probe * * \param[in] probe Probe action to add expected result to * \param[in] rsc Resource that probe is for * \param[in] node Node that probe will run on */ static void add_expected_result(pe_action_t *probe, pe_resource_t *rsc, pe_node_t *node) { // Check whether resource is currently active on node pe_node_t *running = pe_find_node_id(rsc->running_on, node->details->id); // The expected result is what we think the resource's current state is if (running == NULL) { pe__add_action_expected_result(probe, CRM_EX_NOT_RUNNING); } else if (rsc->role == RSC_ROLE_PROMOTED) { pe__add_action_expected_result(probe, CRM_EX_PROMOTED); } } /*! * \internal * \brief Create any needed robes on a node for a list of resources * * \param[in] rscs List of resources to create probes for * \param[in] node Node to create probes on * * \return true if any probe was created, otherwise false */ bool pcmk__probe_resource_list(GList *rscs, pe_node_t *node) { bool any_created = false; for (GList *iter = rscs; iter != NULL; iter = iter->next) { pe_resource_t *rsc = (pe_resource_t *) iter->data; if (rsc->cmds->create_probe(rsc, node)) { any_created = true; } } return any_created; } /*! * \internal * \brief Order one resource's start after another's start-up probe * * \param[in] rsc1 Resource that might get start-up probe * \param[in] rsc2 Resource that might be started */ static void probe_then_start(pe_resource_t *rsc1, pe_resource_t *rsc2) { if ((rsc1->allocated_to != NULL) && (g_hash_table_lookup(rsc1->known_on, rsc1->allocated_to->details->id) == NULL)) { pcmk__new_ordering(rsc1, pcmk__op_key(rsc1->id, RSC_STATUS, 0), NULL, rsc2, pcmk__op_key(rsc2->id, RSC_START, 0), NULL, pe_order_optional, rsc1->cluster); } } /*! * \internal * \brief Check whether a guest resource will stop * * \param[in] node Guest node to check * * \return true if guest resource will likely stop, otherwise false */ static bool guest_resource_will_stop(pe_node_t *node) { pe_resource_t *guest_rsc = node->details->remote_rsc->container; /* Ideally, we'd check whether the guest has a required stop, but that * information doesn't exist yet, so approximate it ... */ return node->details->remote_requires_reset || node->details->unclean || pcmk_is_set(guest_rsc->flags, pe_rsc_failed) || (guest_rsc->next_role == RSC_ROLE_STOPPED) // Guest is moving || ((guest_rsc->role > RSC_ROLE_STOPPED) && (guest_rsc->allocated_to != NULL) && (pe_find_node(guest_rsc->running_on, guest_rsc->allocated_to->details->uname) == NULL)); } /*! * \internal * \brief Create a probe action for a resource on a node * * \param[in] rsc Resource to create probe for * \param[in[ node Node to create probe on * * \return Newly created probe action */ static pe_action_t * probe_action(pe_resource_t *rsc, pe_node_t *node) { pe_action_t *probe = NULL; char *key = pcmk__op_key(rsc->id, RSC_STATUS, 0); crm_debug("Scheduling probe of %s %s on %s", role2text(rsc->role), rsc->id, pe__node_name(node)); probe = custom_action(rsc, key, RSC_STATUS, node, FALSE, TRUE, rsc->cluster); pe__clear_action_flags(probe, pe_action_optional); pcmk__order_vs_unfence(rsc, node, probe, pe_order_optional); add_expected_result(probe, rsc, node); return probe; } /*! * \internal * \brief Create probes for a resource on a node, if needed * * \brief Schedule any probes needed for a resource on a node * * \param[in] rsc Resource to create probe for * \param[in] node Node to create probe on * * \return true if any probe was created, otherwise false */ bool pcmk__probe_rsc_on_node(pe_resource_t *rsc, pe_node_t *node) { uint32_t flags = pe_order_optional; pe_action_t *probe = NULL; pe_node_t *allowed = NULL; pe_resource_t *top = uber_parent(rsc); const char *reason = NULL; CRM_CHECK((rsc != NULL) && (node != NULL), return false); if (!pcmk_is_set(rsc->cluster->flags, pe_flag_startup_probes)) { reason = "start-up probes are disabled"; goto no_probe; } if (pe__is_guest_or_remote_node(node)) { const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS); if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_none)) { reason = "Pacemaker Remote nodes cannot run stonith agents"; goto no_probe; } else if (pe__is_guest_node(node) && pe__resource_contains_guest_node(rsc->cluster, rsc)) { reason = "guest nodes cannot run resources containing guest nodes"; goto no_probe; } else if (rsc->is_remote_node) { reason = "Pacemaker Remote nodes cannot host remote connections"; goto no_probe; } } // If this is a collective resource, probes are created for its children if (rsc->children != NULL) { return pcmk__probe_resource_list(rsc->children, node); } if ((rsc->container != NULL) && !rsc->is_remote_node) { reason = "resource is inside a container"; goto no_probe; } else if (pcmk_is_set(rsc->flags, pe_rsc_orphan)) { reason = "resource is orphaned"; goto no_probe; } else if (g_hash_table_lookup(rsc->known_on, node->details->id) != NULL) { reason = "resource state is already known"; goto no_probe; } allowed = g_hash_table_lookup(rsc->allowed_nodes, node->details->id); if (rsc->exclusive_discover || top->exclusive_discover) { // Exclusive discovery is enabled ... if (allowed == NULL) { // ... but this node is not allowed to run the resource reason = "resource has exclusive discovery but is not allowed " "on node"; goto no_probe; } else if (allowed->rsc_discover_mode != pe_discover_exclusive) { // ... but no constraint marks this node for discovery of resource reason = "resource has exclusive discovery but is not enabled " "on node"; goto no_probe; } } if (allowed == NULL) { allowed = node; } if (allowed->rsc_discover_mode == pe_discover_never) { reason = "node has discovery disabled"; goto no_probe; } if (pe__is_guest_node(node)) { pe_resource_t *guest = node->details->remote_rsc->container; if (guest->role == RSC_ROLE_STOPPED) { // The guest is stopped, so we know no resource is active there reason = "node's guest is stopped"; probe_then_start(guest, top); goto no_probe; } else if (guest_resource_will_stop(node)) { reason = "node's guest will stop"; // Order resource start after guest stop (in case it's restarting) pcmk__new_ordering(guest, pcmk__op_key(guest->id, RSC_STOP, 0), NULL, top, pcmk__op_key(top->id, RSC_START, 0), NULL, pe_order_optional, rsc->cluster); goto no_probe; } } // We've eliminated all cases where a probe is not needed, so now it is probe = probe_action(rsc, node); /* Order the probe relative to the parent -- or the resource itself if * cloned or a fence device when unfencing is used. */ if ((pcmk_is_set(rsc->flags, pe_rsc_fence_device) && pcmk_is_set(rsc->cluster->flags, pe_flag_enable_unfencing)) || !pe_rsc_is_clone(top)) { top = rsc; } if (!pcmk_is_set(probe->flags, pe_action_runnable) - && (rsc->running_on == NULL)) { + && (top->running_on == NULL)) { /* Prevent the parent from starting if the resource can't, but don't * cause the parent to stop if already active. */ pe__set_order_flags(flags, pe_order_runnable_left); } // Start or reload the parent after probing the resource pcmk__new_ordering(rsc, NULL, probe, top, pcmk__op_key(top->id, RSC_START, 0), NULL, flags, rsc->cluster); pcmk__new_ordering(rsc, NULL, probe, top, reload_key(rsc), NULL, pe_order_optional, rsc->cluster); return true; no_probe: pe_rsc_trace(rsc, "Skipping probe for %s on %s because %s", rsc->id, node->details->id, reason); return false; } /*! * \internal * \brief Check whether a probe should be ordered before another action * * \param[in] probe Probe action to check * \param[in] then Other action to check * * \return true if \p probe should be ordered before \p then, otherwise false */ static bool probe_needed_before_action(pe_action_t *probe, pe_action_t *then) { // Probes on a node are performed after unfencing it, not before if (pcmk__str_eq(then->task, CRM_OP_FENCE, pcmk__str_casei) && (probe->node != NULL) && (then->node != NULL) && (probe->node->details == then->node->details)) { const char *op = g_hash_table_lookup(then->meta, "stonith_action"); if (pcmk__str_eq(op, "on", pcmk__str_casei)) { return false; } } // Probes should be done on a node before shutting it down if (pcmk__str_eq(then->task, CRM_OP_SHUTDOWN, pcmk__str_none) && (probe->node != NULL) && (then->node != NULL) && (probe->node->details != then->node->details)) { return false; } // Otherwise probes should always be done before any other action return true; } /*! * \internal * \brief Add implicit "probe then X" orderings for "stop then X" orderings * * If the state of a resource is not known yet, a probe will be scheduled, * expecting a "not running" result. If the probe fails, a stop will not be * scheduled until the next transition. Thus, if there are ordering constraints * like "stop this resource then do something else that's not for the same * resource", add implicit "probe this resource then do something" equivalents * so the relation is upheld until we know whether a stop is needed. * * \param[in] data_set Cluster working set */ static void add_probe_orderings_for_stops(pe_working_set_t *data_set) { for (GList *iter = data_set->ordering_constraints; iter != NULL; iter = iter->next) { pe__ordering_t *order = iter->data; uint32_t order_flags = pe_order_optional; GList *probes = NULL; GList *then_actions = NULL; // Skip disabled orderings if (order->flags == pe_order_none) { continue; } // Skip non-resource orderings, and orderings for the same resource if ((order->lh_rsc == NULL) || (order->lh_rsc == order->rh_rsc)) { continue; } // Skip invalid orderings (shouldn't be possible) if (((order->lh_action == NULL) && (order->lh_action_task == NULL)) || ((order->rh_action == NULL) && (order->rh_action_task == NULL))) { continue; } // Skip orderings for first actions other than stop if ((order->lh_action != NULL) && !pcmk__str_eq(order->lh_action->task, RSC_STOP, pcmk__str_none)) { continue; } else if ((order->lh_action == NULL) && !pcmk__ends_with(order->lh_action_task, "_" RSC_STOP "_0")) { continue; } /* Do not imply a probe ordering for a resource inside of a stopping * container. Otherwise, it might introduce a transition loop, since a * probe could be scheduled after the container starts again. */ if ((order->rh_rsc != NULL) && (order->lh_rsc->container == order->rh_rsc)) { if ((order->rh_action != NULL) && pcmk__str_eq(order->rh_action->task, RSC_STOP, pcmk__str_none)) { continue; } else if ((order->rh_action == NULL) && pcmk__ends_with(order->rh_action_task, "_" RSC_STOP "_0")) { continue; } } // Preserve certain order options for future filtering if (pcmk_is_set(order->flags, pe_order_apply_first_non_migratable)) { pe__set_order_flags(order_flags, pe_order_apply_first_non_migratable); } if (pcmk_is_set(order->flags, pe_order_same_node)) { pe__set_order_flags(order_flags, pe_order_same_node); } // Preserve certain order types for future filtering if ((order->flags == pe_order_anti_colocation) || (order->flags == pe_order_load)) { order_flags = order->flags; } // List all scheduled probes for the first resource probes = pe__resource_actions(order->lh_rsc, NULL, RSC_STATUS, FALSE); if (probes == NULL) { // There aren't any continue; } // List all relevant "then" actions if (order->rh_action != NULL) { then_actions = g_list_prepend(NULL, order->rh_action); } else if (order->rh_rsc != NULL) { then_actions = find_actions(order->rh_rsc->actions, order->rh_action_task, NULL); if (then_actions == NULL) { // There aren't any g_list_free(probes); continue; } } crm_trace("Implying 'probe then' orderings for '%s then %s' " "(id=%d, type=%.6x)", order->lh_action? order->lh_action->uuid : order->lh_action_task, order->rh_action? order->rh_action->uuid : order->rh_action_task, order->id, order->flags); for (GList *probe_iter = probes; probe_iter != NULL; probe_iter = probe_iter->next) { pe_action_t *probe = (pe_action_t *) probe_iter->data; for (GList *then_iter = then_actions; then_iter != NULL; then_iter = then_iter->next) { pe_action_t *then = (pe_action_t *) then_iter->data; if (probe_needed_before_action(probe, then)) { order_actions(probe, then, order_flags); } } } g_list_free(then_actions); g_list_free(probes); } } /*! * \internal * \brief Order probes before restarts and re-promotes * * If a given ordering is a "probe then start" or "probe then promote" ordering, * add an implicit "probe then stop/demote" ordering in case the action is part * of a restart/re-promote, and do the same recursively for all actions ordered * after the "then" action. * * \param[in] probe Probe as 'first' action in an ordering * \param[in] after 'then' action in the ordering * \param[in] data_set Cluster working set */ static void add_restart_orderings_for_probe(pe_action_t *probe, pe_action_t *after, pe_working_set_t *data_set) { GList *iter = NULL; bool interleave = false; pe_resource_t *compatible_rsc = NULL; // Validate that this is a resource probe followed by some action if ((after == NULL) || (probe == NULL) || (probe->rsc == NULL) || (probe->rsc->variant != pe_native) || !pcmk__str_eq(probe->task, RSC_STATUS, pcmk__str_casei)) { return; } // Avoid running into any possible loop if (pcmk_is_set(after->flags, pe_action_tracking)) { return; } pe__set_action_flags(after, pe_action_tracking); crm_trace("Adding probe restart orderings for '%s@%s then %s@%s'", probe->uuid, pe__node_name(probe->node), after->uuid, pe__node_name(after->node)); /* Add restart orderings if "then" is for a different primitive. * Orderings for collective resources will be added later. */ if ((after->rsc != NULL) && (after->rsc->variant == pe_native) && (probe->rsc != after->rsc)) { GList *then_actions = NULL; if (pcmk__str_eq(after->task, RSC_START, pcmk__str_casei)) { then_actions = pe__resource_actions(after->rsc, NULL, RSC_STOP, FALSE); } else if (pcmk__str_eq(after->task, RSC_PROMOTE, pcmk__str_casei)) { then_actions = pe__resource_actions(after->rsc, NULL, RSC_DEMOTE, FALSE); } for (iter = then_actions; iter != NULL; iter = iter->next) { pe_action_t *then = (pe_action_t *) iter->data; // Skip pseudo-actions (for example, those implied by fencing) if (!pcmk_is_set(then->flags, pe_action_pseudo)) { order_actions(probe, then, pe_order_optional); } } g_list_free(then_actions); } /* Detect whether "then" is an interleaved clone action. For these, we want * to add orderings only for the relevant instance. */ if ((after->rsc != NULL) && (after->rsc->variant > pe_group)) { const char *interleave_s = g_hash_table_lookup(after->rsc->meta, XML_RSC_ATTR_INTERLEAVE); interleave = crm_is_true(interleave_s); if (interleave) { compatible_rsc = find_compatible_child(probe->rsc, after->rsc, RSC_ROLE_UNKNOWN, FALSE); } } /* Now recursively do the same for all actions ordered after "then". This * also handles collective resources since the collective action will be * ordered before its individual instances' actions. */ for (iter = after->actions_after; iter != NULL; iter = iter->next) { pe_action_wrapper_t *after_wrapper = (pe_action_wrapper_t *) iter->data; /* pe_order_implies_then is the reason why a required A.start * implies/enforces B.start to be required too, which is the cause of * B.restart/re-promote. * * Not sure about pe_order_implies_then_on_node though. It's now only * used for unfencing case, which tends to introduce transition * loops... */ if (!pcmk_is_set(after_wrapper->type, pe_order_implies_then)) { /* The order type between a group/clone and its child such as * B.start-> B_child.start is: * pe_order_implies_first_printed | pe_order_runnable_left * * Proceed through the ordering chain and build dependencies with * its children. */ if ((after->rsc == NULL) || (after->rsc->variant < pe_group) || (probe->rsc->parent == after->rsc) || (after_wrapper->action->rsc == NULL) || (after_wrapper->action->rsc->variant > pe_group) || (after->rsc != after_wrapper->action->rsc->parent)) { continue; } /* Proceed to the children of a group or a non-interleaved clone. * For an interleaved clone, proceed only to the relevant child. */ if ((after->rsc->variant > pe_group) && interleave && ((compatible_rsc == NULL) || (compatible_rsc != after_wrapper->action->rsc))) { continue; } } crm_trace("Recursively adding probe restart orderings for " "'%s@%s then %s@%s' (type=%#.6x)", after->uuid, pe__node_name(after->node), after_wrapper->action->uuid, pe__node_name(after_wrapper->action->node), after_wrapper->type); add_restart_orderings_for_probe(probe, after_wrapper->action, data_set); } } /*! * \internal * \brief Clear the tracking flag on all scheduled actions * * \param[in] data_set Cluster working set */ static void clear_actions_tracking_flag(pe_working_set_t *data_set) { GList *gIter = NULL; for (gIter = data_set->actions; gIter != NULL; gIter = gIter->next) { pe_action_t *action = (pe_action_t *) gIter->data; pe__clear_action_flags(action, pe_action_tracking); } } /*! * \internal * \brief Add restart orderings for any scheduled probes for a given resource * * \param[in] rsc Resource whose probes should be ordered * \param[in] data_set Cluster working set */ static void add_restart_orderings_for_rsc(pe_resource_t *rsc, pe_working_set_t *data_set) { GList *probes = NULL; // For collective resources, order each instance recursively if (rsc->variant != pe_native) { g_list_foreach(rsc->children, (GFunc) add_restart_orderings_for_rsc, data_set); return; } // Find all probes for given resource probes = pe__resource_actions(rsc, NULL, RSC_STATUS, FALSE); // Add probe restart orderings for each probe found for (GList *iter = probes; iter != NULL; iter = iter->next) { pe_action_t *probe = (pe_action_t *) iter->data; for (GList *then_iter = probe->actions_after; then_iter != NULL; then_iter = then_iter->next) { pe_action_wrapper_t *then = (pe_action_wrapper_t *) then_iter->data; add_restart_orderings_for_probe(probe, then->action, data_set); clear_actions_tracking_flag(data_set); } } g_list_free(probes); } /*! * \internal * \brief Add "A then probe B" orderings for "A then B" orderings * * \param[in] data_set Cluster working set * * \note This function is currently disabled (see next comment). */ static void order_then_probes(pe_working_set_t *data_set) { #if 0 /* Given an ordering "A then B", we would prefer to wait for A to be started * before probing B. * * For example, if A is a filesystem which B can't even run without, it * would be helpful if the author of B's agent could assume that A is * running before B.monitor will be called. * * However, we can't _only_ probe after A is running, otherwise we wouldn't * detect the state of B if A could not be started. We can't even do an * opportunistic version of this, because B may be moving: * * A.stop -> A.start -> B.probe -> B.stop -> B.start * * and if we add B.stop -> A.stop here, we get a loop: * * A.stop -> A.start -> B.probe -> B.stop -> A.stop * * We could kill the "B.probe -> B.stop" dependency, but that could mean * stopping B "too" soon, because B.start must wait for the probe, and * we don't want to stop B if we can't start it. * * We could add the ordering only if A is an anonymous clone with * clone-max == node-max (since we'll never be moving it). However, we could * still be stopping one instance at the same time as starting another. * * The complexity of checking for allowed conditions combined with the ever * narrowing use case suggests that this code should remain disabled until * someone gets smarter. */ for (GList *iter = data_set->resources; iter != NULL; iter = iter->next) { pe_resource_t *rsc = (pe_resource_t *) iter->data; pe_action_t *start = NULL; GList *actions = NULL; GList *probes = NULL; actions = pe__resource_actions(rsc, NULL, RSC_START, FALSE); if (actions) { start = actions->data; g_list_free(actions); } if (start == NULL) { crm_err("No start action for %s", rsc->id); continue; } probes = pe__resource_actions(rsc, NULL, RSC_STATUS, FALSE); for (actions = start->actions_before; actions != NULL; actions = actions->next) { pe_action_wrapper_t *before = (pe_action_wrapper_t *) actions->data; pe_action_t *first = before->action; pe_resource_t *first_rsc = first->rsc; if (first->required_runnable_before) { for (GList *clone_actions = first->actions_before; clone_actions != NULL; clone_actions = clone_actions->next) { before = (pe_action_wrapper_t *) clone_actions->data; crm_trace("Testing '%s then %s' for %s", first->uuid, before->action->uuid, start->uuid); CRM_ASSERT(before->action->rsc != NULL); first_rsc = before->action->rsc; break; } } else if (!pcmk__str_eq(first->task, RSC_START, pcmk__str_none)) { crm_trace("Not a start op %s for %s", first->uuid, start->uuid); } if (first_rsc == NULL) { continue; } else if (uber_parent(first_rsc) == uber_parent(start->rsc)) { crm_trace("Same parent %s for %s", first_rsc->id, start->uuid); continue; } else if (!pe_rsc_is_clone(uber_parent(first_rsc))) { crm_trace("Not a clone %s for %s", first_rsc->id, start->uuid); continue; } crm_err("Applying %s before %s %d", first->uuid, start->uuid, uber_parent(first_rsc)->variant); for (GList *probe_iter = probes; probe_iter != NULL; probe_iter = probe_iter->next) { pe_action_t *probe = (pe_action_t *) probe_iter->data; crm_err("Ordering %s before %s", first->uuid, probe->uuid); order_actions(first, probe, pe_order_optional); } } } #endif } void pcmk__order_probes(pe_working_set_t *data_set) { // Add orderings for "probe then X" g_list_foreach(data_set->resources, (GFunc) add_restart_orderings_for_rsc, data_set); add_probe_orderings_for_stops(data_set); order_then_probes(data_set); } /*! * \internal * \brief Schedule any probes needed * * \param[in] data_set Cluster working set * * \note This may also schedule fencing of failed remote nodes. */ void pcmk__schedule_probes(pe_working_set_t *data_set) { // Schedule probes on each node in the cluster as needed for (GList *iter = data_set->nodes; iter != NULL; iter = iter->next) { pe_node_t *node = (pe_node_t *) iter->data; const char *probed = NULL; if (!node->details->online) { // Don't probe offline nodes if (pcmk__is_failed_remote_node(node)) { pe_fence_node(data_set, node, "the connection is unrecoverable", FALSE); } continue; } else if (node->details->unclean) { // ... or nodes that need fencing continue; } else if (!node->details->rsc_discovery_enabled) { // The user requested that probes not be done on this node continue; } /* This is no longer needed for live clusters, since the probe_complete * node attribute will never be in the CIB. However this is still useful * for processing old saved CIBs (< 1.1.14), including the * reprobe-target_rc regression test. */ probed = pe_node_attribute_raw(node, CRM_OP_PROBED); if (probed != NULL && crm_is_true(probed) == FALSE) { pe_action_t *probe_op = NULL; probe_op = custom_action(NULL, crm_strdup_printf("%s-%s", CRM_OP_REPROBE, node->details->uname), CRM_OP_REPROBE, node, FALSE, TRUE, data_set); add_hash_param(probe_op->meta, XML_ATTR_TE_NOWAIT, XML_BOOLEAN_TRUE); continue; } // Probe each resource in the cluster on this node, as needed pcmk__probe_resource_list(data_set->resources, node); } }