diff --git a/lib/pacemaker/pcmk_sched_bundle.c b/lib/pacemaker/pcmk_sched_bundle.c index 0220cb2c91..11df54a7fd 100644 --- a/lib/pacemaker/pcmk_sched_bundle.c +++ b/lib/pacemaker/pcmk_sched_bundle.c @@ -1,1104 +1,1121 @@ /* * Copyright 2004-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include "libpacemaker_private.h" #define PE__VARIANT_BUNDLE 1 #include static bool is_bundle_node(pe__bundle_variant_data_t *data, pe_node_t *node) { for (GList *gIter = data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; if (node->details == replica->node->details) { return TRUE; } } return FALSE; } gint sort_clone_instance(gconstpointer a, gconstpointer b, gpointer data_set); void distribute_children(pe_resource_t *rsc, GList *children, GList *nodes, int max, int per_host_max, pe_working_set_t * data_set); static GList * get_container_list(pe_resource_t *rsc) { GList *containers = NULL; if (rsc->variant == pe_container) { pe__bundle_variant_data_t *data = NULL; get_bundle_variant_data(data, rsc); for (GList *gIter = data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; containers = g_list_append(containers, replica->container); } } return containers; } static inline GList * get_containers_or_children(pe_resource_t *rsc) { return (rsc->variant == pe_container)? get_container_list(rsc) : rsc->children; } pe_node_t * pcmk__bundle_allocate(pe_resource_t *rsc, pe_node_t *prefer, pe_working_set_t *data_set) { GList *containers = NULL; GList *nodes = NULL; pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(rsc != NULL, return NULL); get_bundle_variant_data(bundle_data, rsc); pe__set_resource_flags(rsc, pe_rsc_allocating); containers = get_container_list(rsc); pe__show_node_weights(!pcmk_is_set(data_set->flags, pe_flag_show_scores), rsc, __func__, rsc->allowed_nodes, data_set); nodes = g_hash_table_get_values(rsc->allowed_nodes); nodes = pcmk__sort_nodes(nodes, NULL, data_set); containers = g_list_sort_with_data(containers, sort_clone_instance, data_set); distribute_children(rsc, containers, nodes, bundle_data->nreplicas, bundle_data->nreplicas_per_host, data_set); g_list_free(nodes); g_list_free(containers); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; pe_node_t *container_host = NULL; CRM_ASSERT(replica); if (replica->ip) { pe_rsc_trace(rsc, "Allocating bundle %s IP %s", rsc->id, replica->ip->id); replica->ip->cmds->allocate(replica->ip, prefer, data_set); } container_host = replica->container->allocated_to; if (replica->remote && pe__is_guest_or_remote_node(container_host)) { /* We need 'nested' connection resources to be on the same * host because pacemaker-remoted only supports a single * active connection */ pcmk__new_colocation("child-remote-with-docker-remote", NULL, INFINITY, replica->remote, container_host->details->remote_rsc, NULL, NULL, true, data_set); } if (replica->remote) { pe_rsc_trace(rsc, "Allocating bundle %s connection %s", rsc->id, replica->remote->id); replica->remote->cmds->allocate(replica->remote, prefer, data_set); } // Explicitly allocate replicas' children before bundle child if (replica->child) { pe_node_t *node = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, replica->child->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & node)) { if (node->details != replica->node->details) { node->weight = -INFINITY; } else if (!pcmk__threshold_reached(replica->child, node, data_set, NULL)) { node->weight = INFINITY; } } pe__set_resource_flags(replica->child->parent, pe_rsc_allocating); pe_rsc_trace(rsc, "Allocating bundle %s replica child %s", rsc->id, replica->child->id); replica->child->cmds->allocate(replica->child, replica->node, data_set); pe__clear_resource_flags(replica->child->parent, pe_rsc_allocating); } } if (bundle_data->child) { pe_node_t *node = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, bundle_data->child->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & node)) { if (is_bundle_node(bundle_data, node)) { node->weight = 0; } else { node->weight = -INFINITY; } } pe_rsc_trace(rsc, "Allocating bundle %s child %s", rsc->id, bundle_data->child->id); bundle_data->child->cmds->allocate(bundle_data->child, prefer, data_set); } pe__clear_resource_flags(rsc, pe_rsc_allocating|pe_rsc_provisional); return NULL; } void pcmk__bundle_create_actions(pe_resource_t *rsc, pe_working_set_t *data_set) { pe_action_t *action = NULL; GList *containers = NULL; pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(rsc != NULL, return); containers = get_container_list(rsc); get_bundle_variant_data(bundle_data, rsc); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica); if (replica->ip) { replica->ip->cmds->create_actions(replica->ip, data_set); } if (replica->container) { replica->container->cmds->create_actions(replica->container, data_set); } if (replica->remote) { replica->remote->cmds->create_actions(replica->remote, data_set); } } clone_create_pseudo_actions(rsc, containers, NULL, NULL, data_set); if (bundle_data->child) { bundle_data->child->cmds->create_actions(bundle_data->child, data_set); if (pcmk_is_set(bundle_data->child->flags, pe_rsc_promotable)) { /* promote */ pcmk__new_rsc_pseudo_action(rsc, RSC_PROMOTE, true, true); action = pcmk__new_rsc_pseudo_action(rsc, RSC_PROMOTED, true, true); action->priority = INFINITY; /* demote */ pcmk__new_rsc_pseudo_action(rsc, RSC_DEMOTE, true, true); action = pcmk__new_rsc_pseudo_action(rsc, RSC_DEMOTED, true, true); action->priority = INFINITY; } } g_list_free(containers); } void pcmk__bundle_internal_constraints(pe_resource_t *rsc, pe_working_set_t *data_set) { pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(rsc != NULL, return); get_bundle_variant_data(bundle_data, rsc); if (bundle_data->child) { pcmk__order_resource_actions(rsc, RSC_START, bundle_data->child, RSC_START, pe_order_implies_first_printed, data_set); pcmk__order_resource_actions(rsc, RSC_STOP, bundle_data->child, RSC_STOP, pe_order_implies_first_printed, data_set); if (bundle_data->child->children) { pcmk__order_resource_actions(bundle_data->child, RSC_STARTED, rsc, RSC_STARTED, pe_order_implies_then_printed, data_set); pcmk__order_resource_actions(bundle_data->child, RSC_STOPPED, rsc, RSC_STOPPED, pe_order_implies_then_printed, data_set); } else { pcmk__order_resource_actions(bundle_data->child, RSC_START, rsc, RSC_STARTED, pe_order_implies_then_printed, data_set); pcmk__order_resource_actions(bundle_data->child, RSC_STOP, rsc, RSC_STOPPED, pe_order_implies_then_printed, data_set); } } for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica); CRM_ASSERT(replica->container); replica->container->cmds->internal_constraints(replica->container, data_set); pcmk__order_starts(rsc, replica->container, pe_order_runnable_left|pe_order_implies_first_printed, data_set); if (replica->child) { pcmk__order_stops(rsc, replica->child, pe_order_implies_first_printed, data_set); } pcmk__order_stops(rsc, replica->container, pe_order_implies_first_printed, data_set); pcmk__order_resource_actions(replica->container, RSC_START, rsc, RSC_STARTED, pe_order_implies_then_printed, data_set); pcmk__order_resource_actions(replica->container, RSC_STOP, rsc, RSC_STOPPED, pe_order_implies_then_printed, data_set); if (replica->ip) { replica->ip->cmds->internal_constraints(replica->ip, data_set); // Start IP then container pcmk__order_starts(replica->ip, replica->container, pe_order_runnable_left|pe_order_preserve, data_set); pcmk__order_stops(replica->container, replica->ip, pe_order_implies_first|pe_order_preserve, data_set); pcmk__new_colocation("ip-with-docker", NULL, INFINITY, replica->ip, replica->container, NULL, NULL, true, data_set); } if (replica->remote) { /* This handles ordering and colocating remote relative to container * (via "resource-with-container"). Since IP is also ordered and * colocated relative to the container, we don't need to do anything * explicit here with IP. */ replica->remote->cmds->internal_constraints(replica->remote, data_set); } if (replica->child) { CRM_ASSERT(replica->remote); // "Start remote then child" is implicit in scheduler's remote logic } } if (bundle_data->child) { bundle_data->child->cmds->internal_constraints(bundle_data->child, data_set); if (pcmk_is_set(bundle_data->child->flags, pe_rsc_promotable)) { promote_demote_constraints(rsc, data_set); /* child demoted before global demoted */ pcmk__order_resource_actions(bundle_data->child, RSC_DEMOTED, rsc, RSC_DEMOTED, pe_order_implies_then_printed, data_set); /* global demote before child demote */ pcmk__order_resource_actions(rsc, RSC_DEMOTE, bundle_data->child, RSC_DEMOTE, pe_order_implies_first_printed, data_set); /* child promoted before global promoted */ pcmk__order_resource_actions(bundle_data->child, RSC_PROMOTED, rsc, RSC_PROMOTED, pe_order_implies_then_printed, data_set); /* global promote before child promote */ pcmk__order_resource_actions(rsc, RSC_PROMOTE, bundle_data->child, RSC_PROMOTE, pe_order_implies_first_printed, data_set); } } } static pe_resource_t * compatible_replica_for_node(pe_resource_t *rsc_lh, pe_node_t *candidate, pe_resource_t *rsc, enum rsc_role_e filter, gboolean current) { pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(candidate != NULL, return NULL); get_bundle_variant_data(bundle_data, rsc); crm_trace("Looking for compatible child from %s for %s on %s", rsc_lh->id, rsc->id, candidate->details->uname); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; if (is_child_compatible(replica->container, candidate, filter, current)) { crm_trace("Pairing %s with %s on %s", rsc_lh->id, replica->container->id, candidate->details->uname); return replica->container; } } crm_trace("Can't pair %s with %s", rsc_lh->id, rsc->id); return NULL; } static pe_resource_t * compatible_replica(pe_resource_t *rsc_lh, pe_resource_t *rsc, enum rsc_role_e filter, gboolean current, pe_working_set_t *data_set) { GList *scratch = NULL; pe_resource_t *pair = NULL; pe_node_t *active_node_lh = NULL; active_node_lh = rsc_lh->fns->location(rsc_lh, NULL, current); if (active_node_lh) { return compatible_replica_for_node(rsc_lh, active_node_lh, rsc, filter, current); } scratch = g_hash_table_get_values(rsc_lh->allowed_nodes); scratch = pcmk__sort_nodes(scratch, NULL, data_set); for (GList *gIter = scratch; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; pair = compatible_replica_for_node(rsc_lh, node, rsc, filter, current); if (pair) { goto done; } } pe_rsc_debug(rsc, "Can't pair %s with %s", rsc_lh->id, (rsc? rsc->id : "none")); done: g_list_free(scratch); return pair; } void pcmk__bundle_rsc_colocation_lh(pe_resource_t *dependent, pe_resource_t *primary, pcmk__colocation_t *constraint, pe_working_set_t *data_set) { /* -- Never called -- * * Instead we add the colocation constraints to the child and call from there */ CRM_ASSERT(FALSE); } int copies_per_node(pe_resource_t * rsc) { /* Strictly speaking, there should be a 'copies_per_node' addition * to the resource function table and each case would be a * function. However that would be serious overkill to return an * int. In fact, it seems to me that both function tables * could/should be replaced by resources.{c,h} full of * rsc_{some_operation} functions containing a switch as below * which calls out to functions named {variant}_{some_operation} * as needed. */ switch(rsc->variant) { case pe_unknown: return 0; case pe_native: case pe_group: return 1; case pe_clone: { const char *max_clones_node = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_INCARNATION_NODEMAX); if (max_clones_node == NULL) { return 1; } else { int max_i; pcmk__scan_min_int(max_clones_node, &max_i, 0); return max_i; } } case pe_container: { pe__bundle_variant_data_t *data = NULL; get_bundle_variant_data(data, rsc); return data->nreplicas_per_host; } } return 0; } void pcmk__bundle_rsc_colocation_rh(pe_resource_t *dependent, pe_resource_t *primary, pcmk__colocation_t *constraint, pe_working_set_t *data_set) { GList *allocated_primaries = NULL; pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(constraint != NULL, return); CRM_CHECK(dependent != NULL, pe_err("dependent was NULL for %s", constraint->id); return); CRM_CHECK(primary != NULL, pe_err("primary was NULL for %s", constraint->id); return); CRM_ASSERT(dependent->variant == pe_native); if (pcmk_is_set(primary->flags, pe_rsc_provisional)) { pe_rsc_trace(primary, "%s is still provisional", primary->id); return; } else if(constraint->dependent->variant > pe_group) { pe_resource_t *primary_replica = compatible_replica(dependent, primary, RSC_ROLE_UNKNOWN, FALSE, data_set); if (primary_replica) { pe_rsc_debug(primary, "Pairing %s with %s", dependent->id, primary_replica->id); dependent->cmds->rsc_colocation_lh(dependent, primary_replica, constraint, data_set); } else if (constraint->score >= INFINITY) { crm_notice("Cannot pair %s with instance of %s", dependent->id, primary->id); pcmk__assign_resource(dependent, NULL, true); } else { pe_rsc_debug(primary, "Cannot pair %s with instance of %s", dependent->id, primary->id); } return; } get_bundle_variant_data(bundle_data, primary); pe_rsc_trace(primary, "Processing constraint %s: %s -> %s %d", constraint->id, dependent->id, primary->id, constraint->score); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; if (constraint->score < INFINITY) { replica->container->cmds->rsc_colocation_rh(dependent, replica->container, constraint, data_set); } else { pe_node_t *chosen = replica->container->fns->location(replica->container, NULL, FALSE); if ((chosen == NULL) || is_set_recursive(replica->container, pe_rsc_block, TRUE)) { continue; } if ((constraint->primary_role >= RSC_ROLE_PROMOTED) && (replica->child == NULL)) { continue; } if ((constraint->primary_role >= RSC_ROLE_PROMOTED) && (replica->child->next_role < RSC_ROLE_PROMOTED)) { continue; } pe_rsc_trace(primary, "Allowing %s: %s %d", constraint->id, chosen->details->uname, chosen->weight); allocated_primaries = g_list_prepend(allocated_primaries, chosen); } } if (constraint->score >= INFINITY) { node_list_exclude(dependent->allowed_nodes, allocated_primaries, FALSE); } g_list_free(allocated_primaries); } enum pe_action_flags pcmk__bundle_action_flags(pe_action_t *action, pe_node_t *node) { GList *containers = NULL; enum pe_action_flags flags = 0; pe__bundle_variant_data_t *data = NULL; get_bundle_variant_data(data, action->rsc); if(data->child) { enum action_tasks task = get_complex_task(data->child, action->task, TRUE); switch(task) { case no_action: case action_notify: case action_notified: case action_promote: case action_promoted: case action_demote: case action_demoted: return summary_action_flags(action, data->child->children, node); default: break; } } containers = get_container_list(action->rsc); flags = summary_action_flags(action, containers, node); g_list_free(containers); return flags; } pe_resource_t * find_compatible_child_by_node(pe_resource_t * local_child, pe_node_t * local_node, pe_resource_t * rsc, enum rsc_role_e filter, gboolean current) { GList *gIter = NULL; GList *children = NULL; if (local_node == NULL) { crm_err("Can't colocate unrunnable child %s with %s", local_child->id, rsc->id); return NULL; } crm_trace("Looking for compatible child from %s for %s on %s", local_child->id, rsc->id, local_node->details->uname); children = get_containers_or_children(rsc); for (gIter = children; gIter != NULL; gIter = gIter->next) { pe_resource_t *child_rsc = (pe_resource_t *) gIter->data; if(is_child_compatible(child_rsc, local_node, filter, current)) { crm_trace("Pairing %s with %s on %s", local_child->id, child_rsc->id, local_node->details->uname); return child_rsc; } } crm_trace("Can't pair %s with %s", local_child->id, rsc->id); if(children != rsc->children) { g_list_free(children); } return NULL; } static pe__bundle_replica_t * replica_for_container(pe_resource_t *rsc, pe_resource_t *container, pe_node_t *node) { if (rsc->variant == pe_container) { pe__bundle_variant_data_t *data = NULL; get_bundle_variant_data(data, rsc); for (GList *gIter = data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; if (replica->child && (container == replica->container) && (node->details == replica->node->details)) { return replica; } } } return NULL; } static enum pe_graph_flags multi_update_interleave_actions(pe_action_t *first, pe_action_t *then, pe_node_t *node, enum pe_action_flags flags, enum pe_action_flags filter, enum pe_ordering type, pe_working_set_t *data_set) { GList *gIter = NULL; GList *children = NULL; gboolean current = FALSE; enum pe_graph_flags changed = pe_graph_none; /* Fix this - lazy */ if (pcmk__ends_with(first->uuid, "_stopped_0") || pcmk__ends_with(first->uuid, "_demoted_0")) { current = TRUE; } children = get_containers_or_children(then->rsc); for (gIter = children; gIter != NULL; gIter = gIter->next) { pe_resource_t *then_child = gIter->data; pe_resource_t *first_child = find_compatible_child(then_child, first->rsc, RSC_ROLE_UNKNOWN, current, data_set); if (first_child == NULL && current) { crm_trace("Ignore"); } else if (first_child == NULL) { crm_debug("No match found for %s (%d / %s / %s)", then_child->id, current, first->uuid, then->uuid); /* Me no like this hack - but what else can we do? * * If there is no-one active or about to be active * on the same node as then_child, then they must * not be allowed to start */ if (type & (pe_order_runnable_left | pe_order_implies_then) /* Mandatory */ ) { pe_rsc_info(then->rsc, "Inhibiting %s from being active", then_child->id); if (pcmk__assign_resource(then_child, NULL, true)) { pe__set_graph_flags(changed, first, pe_graph_updated_then); } } } else { pe_action_t *first_action = NULL; pe_action_t *then_action = NULL; enum action_tasks task = clone_child_action(first); const char *first_task = task2text(task); pe__bundle_replica_t *first_replica = NULL; pe__bundle_replica_t *then_replica = NULL; first_replica = replica_for_container(first->rsc, first_child, node); if (strstr(first->task, "stop") && first_replica && first_replica->child) { /* Except for 'stopped' we should be looking at the * in-container resource, actions for the child will * happen later and are therefor more likely to align * with the user's intent. */ first_action = find_first_action(first_replica->child->actions, NULL, task2text(task), node); } else { first_action = find_first_action(first_child->actions, NULL, task2text(task), node); } then_replica = replica_for_container(then->rsc, then_child, node); if (strstr(then->task, "mote") && then_replica && then_replica->child) { /* Promote/demote actions will never be found for the * container resource, look in the child instead * * Alternatively treat: * 'XXXX then promote YYYY' as 'XXXX then start container for YYYY', and * 'demote XXXX then stop YYYY' as 'stop container for XXXX then stop YYYY' */ then_action = find_first_action(then_replica->child->actions, NULL, then->task, node); } else { then_action = find_first_action(then_child->actions, NULL, then->task, node); } if (first_action == NULL) { if (!pcmk_is_set(first_child->flags, pe_rsc_orphan) && !pcmk__str_any_of(first_task, RSC_STOP, RSC_DEMOTE, NULL)) { crm_err("Internal error: No action found for %s in %s (first)", first_task, first_child->id); } else { crm_trace("No action found for %s in %s%s (first)", first_task, first_child->id, pcmk_is_set(first_child->flags, pe_rsc_orphan)? " (ORPHAN)" : ""); } continue; } /* We're only interested if 'then' is neither stopping nor being demoted */ if (then_action == NULL) { if (!pcmk_is_set(then_child->flags, pe_rsc_orphan) && !pcmk__str_any_of(then->task, RSC_STOP, RSC_DEMOTE, NULL)) { crm_err("Internal error: No action found for %s in %s (then)", then->task, then_child->id); } else { crm_trace("No action found for %s in %s%s (then)", then->task, then_child->id, pcmk_is_set(then_child->flags, pe_rsc_orphan)? " (ORPHAN)" : ""); } continue; } if (order_actions(first_action, then_action, type)) { crm_debug("Created constraint for %s (%d) -> %s (%d) %.6x", first_action->uuid, pcmk_is_set(first_action->flags, pe_action_optional), then_action->uuid, pcmk_is_set(then_action->flags, pe_action_optional), type); pe__set_graph_flags(changed, first, pe_graph_updated_first|pe_graph_updated_then); } if(first_action && then_action) { changed |= then_child->cmds->update_actions(first_action, then_action, node, first_child->cmds->action_flags(first_action, node), filter, type, data_set); } else { crm_err("Nothing found either for %s (%p) or %s (%p) %s", first_child->id, first_action, then_child->id, then_action, task2text(task)); } } } if(children != then->rsc->children) { g_list_free(children); } return changed; } static bool can_interleave_actions(pe_action_t *first, pe_action_t *then) { bool interleave = FALSE; pe_resource_t *rsc = NULL; const char *interleave_s = NULL; if(first->rsc == NULL || then->rsc == NULL) { crm_trace("Not interleaving %s with %s (both must be resources)", first->uuid, then->uuid); return FALSE; } else if(first->rsc == then->rsc) { crm_trace("Not interleaving %s with %s (must belong to different resources)", first->uuid, then->uuid); return FALSE; } else if(first->rsc->variant < pe_clone || then->rsc->variant < pe_clone) { crm_trace("Not interleaving %s with %s (both sides must be clones or bundles)", first->uuid, then->uuid); return FALSE; } if (pcmk__ends_with(then->uuid, "_stop_0") || pcmk__ends_with(then->uuid, "_demote_0")) { rsc = first->rsc; } else { rsc = then->rsc; } interleave_s = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_INTERLEAVE); interleave = crm_is_true(interleave_s); crm_trace("Interleave %s -> %s: %s (based on %s)", first->uuid, then->uuid, interleave ? "yes" : "no", rsc->id); return interleave; } enum pe_graph_flags pcmk__multi_update_actions(pe_action_t *first, pe_action_t *then, pe_node_t *node, enum pe_action_flags flags, enum pe_action_flags filter, enum pe_ordering type, pe_working_set_t *data_set) { enum pe_graph_flags changed = pe_graph_none; crm_trace("%s -> %s", first->uuid, then->uuid); if(can_interleave_actions(first, then)) { changed = multi_update_interleave_actions(first, then, node, flags, filter, type, data_set); } else if(then->rsc) { GList *gIter = NULL; GList *children = NULL; // Handle the 'primitive' ordering case changed |= native_update_actions(first, then, node, flags, filter, type, data_set); // Now any children (or containers in the case of a bundle) children = get_containers_or_children(then->rsc); for (gIter = children; gIter != NULL; gIter = gIter->next) { pe_resource_t *then_child = (pe_resource_t *) gIter->data; enum pe_graph_flags then_child_changed = pe_graph_none; pe_action_t *then_child_action = find_first_action(then_child->actions, NULL, then->task, node); if (then_child_action) { enum pe_action_flags then_child_flags = then_child->cmds->action_flags(then_child_action, node); if (pcmk_is_set(then_child_flags, pe_action_runnable)) { then_child_changed |= then_child->cmds->update_actions(first, then_child_action, node, flags, filter, type, data_set); } changed |= then_child_changed; if (then_child_changed & pe_graph_updated_then) { for (GList *lpc = then_child_action->actions_after; lpc != NULL; lpc = lpc->next) { pe_action_wrapper_t *next = (pe_action_wrapper_t *) lpc->data; pcmk__update_action_for_orderings(next->action, data_set); } } } } if(children != then->rsc->children) { g_list_free(children); } } return changed; } void pcmk__bundle_rsc_location(pe_resource_t *rsc, pe__location_t *constraint) { pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, rsc); pcmk__apply_location(constraint, rsc); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; if (replica->container) { replica->container->cmds->rsc_location(replica->container, constraint); } if (replica->ip) { replica->ip->cmds->rsc_location(replica->ip, constraint); } } if (bundle_data->child && ((constraint->role_filter == RSC_ROLE_UNPROMOTED) || (constraint->role_filter == RSC_ROLE_PROMOTED))) { bundle_data->child->cmds->rsc_location(bundle_data->child, constraint); bundle_data->child->rsc_location = g_list_prepend(bundle_data->child->rsc_location, constraint); } } void pcmk__bundle_expand(pe_resource_t *rsc, pe_working_set_t * data_set) { pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(rsc != NULL, return); get_bundle_variant_data(bundle_data, rsc); if (bundle_data->child) { bundle_data->child->cmds->expand(bundle_data->child, data_set); } for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica); if (replica->remote && replica->container && pe__bundle_needs_remote_name(replica->remote, data_set)) { /* REMOTE_CONTAINER_HACK: Allow remote nodes to run containers that * run pacemaker-remoted inside, without needing a separate IP for * the container. This is done by configuring the inner remote's * connection host as the magic string "#uname", then * replacing it with the underlying host when needed. */ xmlNode *nvpair = get_xpath_object("//nvpair[@name='" XML_RSC_ATTR_REMOTE_RA_ADDR "']", replica->remote->xml, LOG_ERR); const char *calculated_addr = NULL; // Replace the value in replica->remote->xml (if appropriate) calculated_addr = pe__add_bundle_remote_name(replica->remote, data_set, nvpair, "value"); if (calculated_addr) { /* Since this is for the bundle as a resource, and not any * particular action, replace the value in the default * parameters (not evaluated for node). action2xml() will grab * it from there to replace it in node-evaluated parameters. */ GHashTable *params = pe_rsc_params(replica->remote, NULL, data_set); crm_trace("Set address for bundle connection %s to bundle host %s", replica->remote->id, calculated_addr); g_hash_table_replace(params, strdup(XML_RSC_ATTR_REMOTE_RA_ADDR), strdup(calculated_addr)); } else { /* The only way to get here is if the remote connection is * neither currently running nor scheduled to run. That means we * won't be doing any operations that require addr (only start * requires it; we additionally use it to compare digests when * unpacking status, promote, and migrate_from history, but * that's already happened by this point). */ crm_info("Unable to determine address for bundle %s remote connection", rsc->id); } } if (replica->ip) { replica->ip->cmds->expand(replica->ip, data_set); } if (replica->container) { replica->container->cmds->expand(replica->container, data_set); } if (replica->remote) { replica->remote->cmds->expand(replica->remote, data_set); } } } gboolean pcmk__bundle_create_probe(pe_resource_t *rsc, pe_node_t *node, pe_action_t *complete, gboolean force, pe_working_set_t * data_set) { bool any_created = FALSE; pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(rsc != NULL, return FALSE); get_bundle_variant_data(bundle_data, rsc); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica); if (replica->ip) { any_created |= replica->ip->cmds->create_probe(replica->ip, node, complete, force, data_set); } if (replica->child && (node->details == replica->node->details)) { any_created |= replica->child->cmds->create_probe(replica->child, node, complete, force, data_set); } if (replica->container) { bool created = replica->container->cmds->create_probe(replica->container, node, complete, force, data_set); if(created) { any_created = TRUE; /* If we're limited to one replica per host (due to * the lack of an IP range probably), then we don't * want any of our peer containers starting until * we've established that no other copies are already * running. * * Partly this is to ensure that nreplicas_per_host is * observed, but also to ensure that the containers * don't fail to start because the necessary port * mappings (which won't include an IP for uniqueness) * are already taken */ for (GList *tIter = bundle_data->replicas; tIter && (bundle_data->nreplicas_per_host == 1); tIter = tIter->next) { pe__bundle_replica_t *other = tIter->data; if ((other != replica) && (other != NULL) && (other->container != NULL)) { pcmk__new_ordering(replica->container, pcmk__op_key(replica->container->id, RSC_STATUS, 0), NULL, other->container, pcmk__op_key(other->container->id, RSC_START, 0), NULL, pe_order_optional|pe_order_same_node, data_set); } } } } if (replica->container && replica->remote && replica->remote->cmds->create_probe(replica->remote, node, complete, force, data_set)) { /* Do not probe the remote resource until we know where the * container is running. This is required for REMOTE_CONTAINER_HACK * to correctly probe remote resources. */ char *probe_uuid = pcmk__op_key(replica->remote->id, RSC_STATUS, 0); pe_action_t *probe = find_first_action(replica->remote->actions, probe_uuid, NULL, node); free(probe_uuid); if (probe) { any_created = TRUE; crm_trace("Ordering %s probe on %s", replica->remote->id, node->details->uname); pcmk__new_ordering(replica->container, pcmk__op_key(replica->container->id, RSC_START, 0), NULL, replica->remote, NULL, probe, pe_order_probe, data_set); } } } return any_created; } void pcmk__bundle_append_meta(pe_resource_t *rsc, xmlNode *xml) { } void pcmk__output_bundle_actions(pe_resource_t *rsc) { pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(rsc != NULL, return); get_bundle_variant_data(bundle_data, rsc); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica); if (replica->ip != NULL) { replica->ip->cmds->output_actions(replica->ip); } if (replica->container != NULL) { replica->container->cmds->output_actions(replica->container); } if (replica->remote != NULL) { replica->remote->cmds->output_actions(replica->remote); } if (replica->child != NULL) { replica->child->cmds->output_actions(replica->child); } } } // Bundle implementation of resource_alloc_functions_t:add_utilization() void pcmk__bundle_add_utilization(pe_resource_t *rsc, pe_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization) { + pe__bundle_variant_data_t *bundle_data = NULL; + pe__bundle_replica_t *replica = NULL; + if (!pcmk_is_set(rsc->flags, pe_rsc_provisional)) { return; } - // @TODO + + get_bundle_variant_data(bundle_data, rsc); + if (bundle_data->replicas == NULL) { + return; + } + + /* All bundle replicas are identical, so using the utilization of the first + * is sufficient for any. Only the implicit container resource can have + * utilization values. + */ + replica = (pe__bundle_replica_t *) bundle_data->replicas->data; + if (replica->container != NULL) { + replica->container->cmds->add_utilization(replica->container, orig_rsc, + all_rscs, utilization); + } } diff --git a/lib/pengine/bundle.c b/lib/pengine/bundle.c index 2c63a946e3..6e2949d1d0 100644 --- a/lib/pengine/bundle.c +++ b/lib/pengine/bundle.c @@ -1,2066 +1,2089 @@ /* * Copyright 2004-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #define PE__VARIANT_BUNDLE 1 #include "./variant.h" static char * next_ip(const char *last_ip) { unsigned int oct1 = 0; unsigned int oct2 = 0; unsigned int oct3 = 0; unsigned int oct4 = 0; int rc = sscanf(last_ip, "%u.%u.%u.%u", &oct1, &oct2, &oct3, &oct4); if (rc != 4) { /*@ TODO check for IPv6 */ return NULL; } else if (oct3 > 253) { return NULL; } else if (oct4 > 253) { ++oct3; oct4 = 1; } else { ++oct4; } return crm_strdup_printf("%u.%u.%u.%u", oct1, oct2, oct3, oct4); } static int allocate_ip(pe__bundle_variant_data_t *data, pe__bundle_replica_t *replica, char *buffer, int max) { if(data->ip_range_start == NULL) { return 0; } else if(data->ip_last) { replica->ipaddr = next_ip(data->ip_last); } else { replica->ipaddr = strdup(data->ip_range_start); } data->ip_last = replica->ipaddr; switch (data->agent_type) { case PE__CONTAINER_AGENT_DOCKER: case PE__CONTAINER_AGENT_PODMAN: if (data->add_host) { return snprintf(buffer, max, " --add-host=%s-%d:%s", data->prefix, replica->offset, replica->ipaddr); } case PE__CONTAINER_AGENT_RKT: return snprintf(buffer, max, " --hosts-entry=%s=%s-%d", replica->ipaddr, data->prefix, replica->offset); default: // PE__CONTAINER_AGENT_UNKNOWN break; } return 0; } static xmlNode * create_resource(const char *name, const char *provider, const char *kind) { xmlNode *rsc = create_xml_node(NULL, XML_CIB_TAG_RESOURCE); crm_xml_add(rsc, XML_ATTR_ID, name); crm_xml_add(rsc, XML_AGENT_ATTR_CLASS, PCMK_RESOURCE_CLASS_OCF); crm_xml_add(rsc, XML_AGENT_ATTR_PROVIDER, provider); crm_xml_add(rsc, XML_ATTR_TYPE, kind); return rsc; } /*! * \internal * \brief Check whether cluster can manage resource inside container * * \param[in] data Container variant data * * \return TRUE if networking configuration is acceptable, FALSE otherwise * * \note The resource is manageable if an IP range or control port has been * specified. If a control port is used without an IP range, replicas per * host must be 1. */ static bool valid_network(pe__bundle_variant_data_t *data) { if(data->ip_range_start) { return TRUE; } if(data->control_port) { if(data->nreplicas_per_host > 1) { pe_err("Specifying the 'control-port' for %s requires 'replicas-per-host=1'", data->prefix); data->nreplicas_per_host = 1; // @TODO to be sure: pe__clear_resource_flags(rsc, pe_rsc_unique); } return TRUE; } return FALSE; } static bool create_ip_resource(pe_resource_t *parent, pe__bundle_variant_data_t *data, pe__bundle_replica_t *replica, pe_working_set_t *data_set) { if(data->ip_range_start) { char *id = NULL; xmlNode *xml_ip = NULL; xmlNode *xml_obj = NULL; id = crm_strdup_printf("%s-ip-%s", data->prefix, replica->ipaddr); crm_xml_sanitize_id(id); xml_ip = create_resource(id, "heartbeat", "IPaddr2"); free(id); xml_obj = create_xml_node(xml_ip, XML_TAG_ATTR_SETS); crm_xml_set_id(xml_obj, "%s-attributes-%d", data->prefix, replica->offset); crm_create_nvpair_xml(xml_obj, NULL, "ip", replica->ipaddr); if(data->host_network) { crm_create_nvpair_xml(xml_obj, NULL, "nic", data->host_network); } if(data->host_netmask) { crm_create_nvpair_xml(xml_obj, NULL, "cidr_netmask", data->host_netmask); } else { crm_create_nvpair_xml(xml_obj, NULL, "cidr_netmask", "32"); } xml_obj = create_xml_node(xml_ip, "operations"); crm_create_op_xml(xml_obj, ID(xml_ip), "monitor", "60s", NULL); // TODO: Other ops? Timeouts and intervals from underlying resource? if (!common_unpack(xml_ip, &replica->ip, parent, data_set)) { return FALSE; } parent->children = g_list_append(parent->children, replica->ip); } return TRUE; } static bool create_docker_resource(pe_resource_t *parent, pe__bundle_variant_data_t *data, pe__bundle_replica_t *replica, pe_working_set_t *data_set) { int offset = 0, max = 4096; char *buffer = calloc(1, max+1); int doffset = 0, dmax = 1024; char *dbuffer = calloc(1, dmax+1); char *id = NULL; xmlNode *xml_container = NULL; xmlNode *xml_obj = NULL; id = crm_strdup_printf("%s-docker-%d", data->prefix, replica->offset); crm_xml_sanitize_id(id); xml_container = create_resource(id, "heartbeat", PE__CONTAINER_AGENT_DOCKER_S); free(id); xml_obj = create_xml_node(xml_container, XML_TAG_ATTR_SETS); crm_xml_set_id(xml_obj, "%s-attributes-%d", data->prefix, replica->offset); crm_create_nvpair_xml(xml_obj, NULL, "image", data->image); crm_create_nvpair_xml(xml_obj, NULL, "allow_pull", XML_BOOLEAN_TRUE); crm_create_nvpair_xml(xml_obj, NULL, "force_kill", XML_BOOLEAN_FALSE); crm_create_nvpair_xml(xml_obj, NULL, "reuse", XML_BOOLEAN_FALSE); offset += snprintf(buffer+offset, max-offset, " --restart=no"); /* Set a container hostname only if we have an IP to map it to. * The user can set -h or --uts=host themselves if they want a nicer * name for logs, but this makes applications happy who need their * hostname to match the IP they bind to. */ if (data->ip_range_start != NULL) { offset += snprintf(buffer+offset, max-offset, " -h %s-%d", data->prefix, replica->offset); } offset += snprintf(buffer+offset, max-offset, " -e PCMK_stderr=1"); if (data->container_network) { #if 0 offset += snprintf(buffer+offset, max-offset, " --link-local-ip=%s", replica->ipaddr); #endif offset += snprintf(buffer+offset, max-offset, " --net=%s", data->container_network); } if(data->control_port) { offset += snprintf(buffer+offset, max-offset, " -e PCMK_remote_port=%s", data->control_port); } else { offset += snprintf(buffer+offset, max-offset, " -e PCMK_remote_port=%d", DEFAULT_REMOTE_PORT); } for(GList *pIter = data->mounts; pIter != NULL; pIter = pIter->next) { pe__bundle_mount_t *mount = pIter->data; if (pcmk_is_set(mount->flags, pe__bundle_mount_subdir)) { char *source = crm_strdup_printf( "%s/%s-%d", mount->source, data->prefix, replica->offset); if(doffset > 0) { doffset += snprintf(dbuffer+doffset, dmax-doffset, ","); } doffset += snprintf(dbuffer+doffset, dmax-doffset, "%s", source); offset += snprintf(buffer+offset, max-offset, " -v %s:%s", source, mount->target); free(source); } else { offset += snprintf(buffer+offset, max-offset, " -v %s:%s", mount->source, mount->target); } if(mount->options) { offset += snprintf(buffer+offset, max-offset, ":%s", mount->options); } } for(GList *pIter = data->ports; pIter != NULL; pIter = pIter->next) { pe__bundle_port_t *port = pIter->data; if (replica->ipaddr) { offset += snprintf(buffer+offset, max-offset, " -p %s:%s:%s", replica->ipaddr, port->source, port->target); } else if(!pcmk__str_eq(data->container_network, "host", pcmk__str_casei)) { // No need to do port mapping if net=host offset += snprintf(buffer+offset, max-offset, " -p %s:%s", port->source, port->target); } } if (data->launcher_options) { offset += snprintf(buffer+offset, max-offset, " %s", data->launcher_options); } if (data->container_host_options) { offset += snprintf(buffer + offset, max - offset, " %s", data->container_host_options); } crm_create_nvpair_xml(xml_obj, NULL, "run_opts", buffer); free(buffer); crm_create_nvpair_xml(xml_obj, NULL, "mount_points", dbuffer); free(dbuffer); if (replica->child) { if (data->container_command) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", data->container_command); } else { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", SBIN_DIR "/pacemaker-remoted"); } /* TODO: Allow users to specify their own? * * We just want to know if the container is alive, we'll * monitor the child independently */ crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", "/bin/true"); #if 0 /* @TODO Consider supporting the use case where we can start and stop * resources, but not proxy local commands (such as setting node * attributes), by running the local executor in stand-alone mode. * However, this would probably be better done via ACLs as with other * Pacemaker Remote nodes. */ } else if ((child != NULL) && data->untrusted) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", CRM_DAEMON_DIR "/pacemaker-execd"); crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", CRM_DAEMON_DIR "/pacemaker/cts-exec-helper -c poke"); #endif } else { if (data->container_command) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", data->container_command); } /* TODO: Allow users to specify their own? * * We don't know what's in the container, so we just want * to know if it is alive */ crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", "/bin/true"); } xml_obj = create_xml_node(xml_container, "operations"); crm_create_op_xml(xml_obj, ID(xml_container), "monitor", "60s", NULL); // TODO: Other ops? Timeouts and intervals from underlying resource? if (!common_unpack(xml_container, &replica->container, parent, data_set)) { return FALSE; } parent->children = g_list_append(parent->children, replica->container); return TRUE; } static bool create_podman_resource(pe_resource_t *parent, pe__bundle_variant_data_t *data, pe__bundle_replica_t *replica, pe_working_set_t *data_set) { int offset = 0, max = 4096; char *buffer = calloc(1, max+1); int doffset = 0, dmax = 1024; char *dbuffer = calloc(1, dmax+1); char *id = NULL; xmlNode *xml_container = NULL; xmlNode *xml_obj = NULL; id = crm_strdup_printf("%s-podman-%d", data->prefix, replica->offset); crm_xml_sanitize_id(id); xml_container = create_resource(id, "heartbeat", PE__CONTAINER_AGENT_PODMAN_S); free(id); xml_obj = create_xml_node(xml_container, XML_TAG_ATTR_SETS); crm_xml_set_id(xml_obj, "%s-attributes-%d", data->prefix, replica->offset); crm_create_nvpair_xml(xml_obj, NULL, "image", data->image); crm_create_nvpair_xml(xml_obj, NULL, "allow_pull", XML_BOOLEAN_TRUE); crm_create_nvpair_xml(xml_obj, NULL, "force_kill", XML_BOOLEAN_FALSE); crm_create_nvpair_xml(xml_obj, NULL, "reuse", XML_BOOLEAN_FALSE); // FIXME: (bandini 2018-08) podman has no restart policies //offset += snprintf(buffer+offset, max-offset, " --restart=no"); /* Set a container hostname only if we have an IP to map it to. * The user can set -h or --uts=host themselves if they want a nicer * name for logs, but this makes applications happy who need their * hostname to match the IP they bind to. */ if (data->ip_range_start != NULL) { offset += snprintf(buffer+offset, max-offset, " -h %s-%d", data->prefix, replica->offset); } offset += snprintf(buffer+offset, max-offset, " -e PCMK_stderr=1"); if (data->container_network) { #if 0 // podman has no support for --link-local-ip offset += snprintf(buffer+offset, max-offset, " --link-local-ip=%s", replica->ipaddr); #endif offset += snprintf(buffer+offset, max-offset, " --net=%s", data->container_network); } if(data->control_port) { offset += snprintf(buffer+offset, max-offset, " -e PCMK_remote_port=%s", data->control_port); } else { offset += snprintf(buffer+offset, max-offset, " -e PCMK_remote_port=%d", DEFAULT_REMOTE_PORT); } for(GList *pIter = data->mounts; pIter != NULL; pIter = pIter->next) { pe__bundle_mount_t *mount = pIter->data; if (pcmk_is_set(mount->flags, pe__bundle_mount_subdir)) { char *source = crm_strdup_printf( "%s/%s-%d", mount->source, data->prefix, replica->offset); if(doffset > 0) { doffset += snprintf(dbuffer+doffset, dmax-doffset, ","); } doffset += snprintf(dbuffer+doffset, dmax-doffset, "%s", source); offset += snprintf(buffer+offset, max-offset, " -v %s:%s", source, mount->target); free(source); } else { offset += snprintf(buffer+offset, max-offset, " -v %s:%s", mount->source, mount->target); } if(mount->options) { offset += snprintf(buffer+offset, max-offset, ":%s", mount->options); } } for(GList *pIter = data->ports; pIter != NULL; pIter = pIter->next) { pe__bundle_port_t *port = pIter->data; if (replica->ipaddr) { offset += snprintf(buffer+offset, max-offset, " -p %s:%s:%s", replica->ipaddr, port->source, port->target); } else if(!pcmk__str_eq(data->container_network, "host", pcmk__str_casei)) { // No need to do port mapping if net=host offset += snprintf(buffer+offset, max-offset, " -p %s:%s", port->source, port->target); } } if (data->launcher_options) { offset += snprintf(buffer+offset, max-offset, " %s", data->launcher_options); } if (data->container_host_options) { offset += snprintf(buffer + offset, max - offset, " %s", data->container_host_options); } crm_create_nvpair_xml(xml_obj, NULL, "run_opts", buffer); free(buffer); crm_create_nvpair_xml(xml_obj, NULL, "mount_points", dbuffer); free(dbuffer); if (replica->child) { if (data->container_command) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", data->container_command); } else { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", SBIN_DIR "/pacemaker-remoted"); } /* TODO: Allow users to specify their own? * * We just want to know if the container is alive, we'll * monitor the child independently */ crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", "/bin/true"); #if 0 /* @TODO Consider supporting the use case where we can start and stop * resources, but not proxy local commands (such as setting node * attributes), by running the local executor in stand-alone mode. * However, this would probably be better done via ACLs as with other * Pacemaker Remote nodes. */ } else if ((child != NULL) && data->untrusted) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", CRM_DAEMON_DIR "/pacemaker-execd"); crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", CRM_DAEMON_DIR "/pacemaker/cts-exec-helper -c poke"); #endif } else { if (data->container_command) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", data->container_command); } /* TODO: Allow users to specify their own? * * We don't know what's in the container, so we just want * to know if it is alive */ crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", "/bin/true"); } xml_obj = create_xml_node(xml_container, "operations"); crm_create_op_xml(xml_obj, ID(xml_container), "monitor", "60s", NULL); // TODO: Other ops? Timeouts and intervals from underlying resource? if (!common_unpack(xml_container, &replica->container, parent, data_set)) { return FALSE; } parent->children = g_list_append(parent->children, replica->container); return TRUE; } static bool create_rkt_resource(pe_resource_t *parent, pe__bundle_variant_data_t *data, pe__bundle_replica_t *replica, pe_working_set_t *data_set) { int offset = 0, max = 4096; char *buffer = calloc(1, max+1); int doffset = 0, dmax = 1024; char *dbuffer = calloc(1, dmax+1); char *id = NULL; xmlNode *xml_container = NULL; xmlNode *xml_obj = NULL; int volid = 0; id = crm_strdup_printf("%s-rkt-%d", data->prefix, replica->offset); crm_xml_sanitize_id(id); xml_container = create_resource(id, "heartbeat", PE__CONTAINER_AGENT_RKT_S); free(id); xml_obj = create_xml_node(xml_container, XML_TAG_ATTR_SETS); crm_xml_set_id(xml_obj, "%s-attributes-%d", data->prefix, replica->offset); crm_create_nvpair_xml(xml_obj, NULL, "image", data->image); crm_create_nvpair_xml(xml_obj, NULL, "allow_pull", "true"); crm_create_nvpair_xml(xml_obj, NULL, "force_kill", "false"); crm_create_nvpair_xml(xml_obj, NULL, "reuse", "false"); /* Set a container hostname only if we have an IP to map it to. * The user can set -h or --uts=host themselves if they want a nicer * name for logs, but this makes applications happy who need their * hostname to match the IP they bind to. */ if (data->ip_range_start != NULL) { offset += snprintf(buffer+offset, max-offset, " --hostname=%s-%d", data->prefix, replica->offset); } offset += snprintf(buffer+offset, max-offset, " --environment=PCMK_stderr=1"); if (data->container_network) { #if 0 offset += snprintf(buffer+offset, max-offset, " --link-local-ip=%s", replica->ipaddr); #endif offset += snprintf(buffer+offset, max-offset, " --net=%s", data->container_network); } if(data->control_port) { offset += snprintf(buffer+offset, max-offset, " --environment=PCMK_remote_port=%s", data->control_port); } else { offset += snprintf(buffer+offset, max-offset, " --environment=PCMK_remote_port=%d", DEFAULT_REMOTE_PORT); } for(GList *pIter = data->mounts; pIter != NULL; pIter = pIter->next) { pe__bundle_mount_t *mount = pIter->data; if (pcmk_is_set(mount->flags, pe__bundle_mount_subdir)) { char *source = crm_strdup_printf( "%s/%s-%d", mount->source, data->prefix, replica->offset); if(doffset > 0) { doffset += snprintf(dbuffer+doffset, dmax-doffset, ","); } doffset += snprintf(dbuffer+doffset, dmax-doffset, "%s", source); offset += snprintf(buffer+offset, max-offset, " --volume vol%d,kind=host,source=%s", volid, source); if(mount->options) { offset += snprintf(buffer+offset, max-offset, ",%s", mount->options); } offset += snprintf(buffer+offset, max-offset, " --mount volume=vol%d,target=%s", volid, mount->target); free(source); } else { offset += snprintf(buffer+offset, max-offset, " --volume vol%d,kind=host,source=%s", volid, mount->source); if(mount->options) { offset += snprintf(buffer+offset, max-offset, ",%s", mount->options); } offset += snprintf(buffer+offset, max-offset, " --mount volume=vol%d,target=%s", volid, mount->target); } volid++; } for(GList *pIter = data->ports; pIter != NULL; pIter = pIter->next) { pe__bundle_port_t *port = pIter->data; if (replica->ipaddr) { offset += snprintf(buffer+offset, max-offset, " --port=%s:%s:%s", port->target, replica->ipaddr, port->source); } else { offset += snprintf(buffer+offset, max-offset, " --port=%s:%s", port->target, port->source); } } if (data->launcher_options) { offset += snprintf(buffer+offset, max-offset, " %s", data->launcher_options); } if (data->container_host_options) { offset += snprintf(buffer + offset, max - offset, " %s", data->container_host_options); } crm_create_nvpair_xml(xml_obj, NULL, "run_opts", buffer); free(buffer); crm_create_nvpair_xml(xml_obj, NULL, "mount_points", dbuffer); free(dbuffer); if (replica->child) { if (data->container_command) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", data->container_command); } else { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", SBIN_DIR "/pacemaker-remoted"); } /* TODO: Allow users to specify their own? * * We just want to know if the container is alive, we'll * monitor the child independently */ crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", "/bin/true"); #if 0 /* @TODO Consider supporting the use case where we can start and stop * resources, but not proxy local commands (such as setting node * attributes), by running the local executor in stand-alone mode. * However, this would probably be better done via ACLs as with other * Pacemaker Remote nodes. */ } else if ((child != NULL) && data->untrusted) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", CRM_DAEMON_DIR "/pacemaker-execd"); crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", CRM_DAEMON_DIR "/pacemaker/cts-exec-helper -c poke"); #endif } else { if (data->container_command) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", data->container_command); } /* TODO: Allow users to specify their own? * * We don't know what's in the container, so we just want * to know if it is alive */ crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", "/bin/true"); } xml_obj = create_xml_node(xml_container, "operations"); crm_create_op_xml(xml_obj, ID(xml_container), "monitor", "60s", NULL); // TODO: Other ops? Timeouts and intervals from underlying resource? if (!common_unpack(xml_container, &replica->container, parent, data_set)) { return FALSE; } parent->children = g_list_append(parent->children, replica->container); return TRUE; } /*! * \brief Ban a node from a resource's (and its children's) allowed nodes list * * \param[in,out] rsc Resource to modify * \param[in] uname Name of node to ban */ static void disallow_node(pe_resource_t *rsc, const char *uname) { gpointer match = g_hash_table_lookup(rsc->allowed_nodes, uname); if (match) { ((pe_node_t *) match)->weight = -INFINITY; ((pe_node_t *) match)->rsc_discover_mode = pe_discover_never; } if (rsc->children) { g_list_foreach(rsc->children, (GFunc) disallow_node, (gpointer) uname); } } static bool create_remote_resource(pe_resource_t *parent, pe__bundle_variant_data_t *data, pe__bundle_replica_t *replica, pe_working_set_t *data_set) { if (replica->child && valid_network(data)) { GHashTableIter gIter; pe_node_t *node = NULL; xmlNode *xml_remote = NULL; char *id = crm_strdup_printf("%s-%d", data->prefix, replica->offset); char *port_s = NULL; const char *uname = NULL; const char *connect_name = NULL; if (pe_find_resource(data_set->resources, id) != NULL) { free(id); // The biggest hammer we have id = crm_strdup_printf("pcmk-internal-%s-remote-%d", replica->child->id, replica->offset); //@TODO return false instead of asserting? CRM_ASSERT(pe_find_resource(data_set->resources, id) == NULL); } /* REMOTE_CONTAINER_HACK: Using "#uname" as the server name when the * connection does not have its own IP is a magic string that we use to * support nested remotes (i.e. a bundle running on a remote node). */ connect_name = (replica->ipaddr? replica->ipaddr : "#uname"); if (data->control_port == NULL) { port_s = pcmk__itoa(DEFAULT_REMOTE_PORT); } /* This sets replica->container as replica->remote's container, which is * similar to what happens with guest nodes. This is how the scheduler * knows that the bundle node is fenced by recovering the container, and * that remote should be ordered relative to the container. */ xml_remote = pe_create_remote_xml(NULL, id, replica->container->id, NULL, NULL, NULL, connect_name, (data->control_port? data->control_port : port_s)); free(port_s); /* Abandon our created ID, and pull the copy from the XML, because we * need something that will get freed during data set cleanup to use as * the node ID and uname. */ free(id); id = NULL; uname = ID(xml_remote); /* Ensure a node has been created for the guest (it may have already * been, if it has a permanent node attribute), and ensure its weight is * -INFINITY so no other resources can run on it. */ node = pe_find_node(data_set->nodes, uname); if (node == NULL) { node = pe_create_node(uname, uname, "remote", "-INFINITY", data_set); } else { node->weight = -INFINITY; } node->rsc_discover_mode = pe_discover_never; /* unpack_remote_nodes() ensures that each remote node and guest node * has a pe_node_t entry. Ideally, it would do the same for bundle nodes. * Unfortunately, a bundle has to be mostly unpacked before it's obvious * what nodes will be needed, so we do it just above. * * Worse, that means that the node may have been utilized while * unpacking other resources, without our weight correction. The most * likely place for this to happen is when common_unpack() calls * resource_location() to set a default score in symmetric clusters. * This adds a node *copy* to each resource's allowed nodes, and these * copies will have the wrong weight. * * As a hacky workaround, fix those copies here. * * @TODO Possible alternative: ensure bundles are unpacked before other * resources, so the weight is correct before any copies are made. */ g_list_foreach(data_set->resources, (GFunc) disallow_node, (gpointer) uname); replica->node = pe__copy_node(node); replica->node->weight = 500; replica->node->rsc_discover_mode = pe_discover_exclusive; /* Ensure the node shows up as allowed and with the correct discovery set */ if (replica->child->allowed_nodes != NULL) { g_hash_table_destroy(replica->child->allowed_nodes); } replica->child->allowed_nodes = pcmk__strkey_table(NULL, free); g_hash_table_insert(replica->child->allowed_nodes, (gpointer) replica->node->details->id, pe__copy_node(replica->node)); { pe_node_t *copy = pe__copy_node(replica->node); copy->weight = -INFINITY; g_hash_table_insert(replica->child->parent->allowed_nodes, (gpointer) replica->node->details->id, copy); } if (!common_unpack(xml_remote, &replica->remote, parent, data_set)) { return FALSE; } g_hash_table_iter_init(&gIter, replica->remote->allowed_nodes); while (g_hash_table_iter_next(&gIter, NULL, (void **)&node)) { if (pe__is_guest_or_remote_node(node)) { /* Remote resources can only run on 'normal' cluster node */ node->weight = -INFINITY; } } replica->node->details->remote_rsc = replica->remote; // Ensure pe__is_guest_node() functions correctly immediately replica->remote->container = replica->container; /* A bundle's #kind is closer to "container" (guest node) than the * "remote" set by pe_create_node(). */ g_hash_table_insert(replica->node->details->attrs, strdup(CRM_ATTR_KIND), strdup("container")); /* One effect of this is that setup_container() will add * replica->remote to replica->container's fillers, which will make * pe__resource_contains_guest_node() true for replica->container. * * replica->child does NOT get added to replica->container's fillers. * The only noticeable effect if it did would be for its fail count to * be taken into account when checking replica->container's migration * threshold. */ parent->children = g_list_append(parent->children, replica->remote); } return TRUE; } static bool create_container(pe_resource_t *parent, pe__bundle_variant_data_t *data, pe__bundle_replica_t *replica, pe_working_set_t *data_set) { switch (data->agent_type) { case PE__CONTAINER_AGENT_DOCKER: if (!create_docker_resource(parent, data, replica, data_set)) { return FALSE; } break; case PE__CONTAINER_AGENT_PODMAN: if (!create_podman_resource(parent, data, replica, data_set)) { return FALSE; } break; case PE__CONTAINER_AGENT_RKT: if (!create_rkt_resource(parent, data, replica, data_set)) { return FALSE; } break; default: // PE__CONTAINER_AGENT_UNKNOWN return FALSE; } if (create_ip_resource(parent, data, replica, data_set) == FALSE) { return FALSE; } if(create_remote_resource(parent, data, replica, data_set) == FALSE) { return FALSE; } if (replica->child && replica->ipaddr) { add_hash_param(replica->child->meta, "external-ip", replica->ipaddr); } if (replica->remote) { /* * Allow the remote connection resource to be allocated to a * different node than the one on which the container is active. * * This makes it possible to have Pacemaker Remote nodes running * containers with pacemaker-remoted inside in order to start * services inside those containers. */ pe__set_resource_flags(replica->remote, pe_rsc_allow_remote_remotes); } return TRUE; } static void mount_add(pe__bundle_variant_data_t *bundle_data, const char *source, const char *target, const char *options, uint32_t flags) { pe__bundle_mount_t *mount = calloc(1, sizeof(pe__bundle_mount_t)); mount->source = strdup(source); mount->target = strdup(target); if (options) { mount->options = strdup(options); } mount->flags = flags; bundle_data->mounts = g_list_append(bundle_data->mounts, mount); } static void mount_free(pe__bundle_mount_t *mount) { free(mount->source); free(mount->target); free(mount->options); free(mount); } static void port_free(pe__bundle_port_t *port) { free(port->source); free(port->target); free(port); } static pe__bundle_replica_t * replica_for_remote(pe_resource_t *remote) { pe_resource_t *top = remote; pe__bundle_variant_data_t *bundle_data = NULL; if (top == NULL) { return NULL; } while (top->parent != NULL) { top = top->parent; } get_bundle_variant_data(bundle_data, top); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; if (replica->remote == remote) { return replica; } } CRM_LOG_ASSERT(FALSE); return NULL; } bool pe__bundle_needs_remote_name(pe_resource_t *rsc, pe_working_set_t *data_set) { const char *value; GHashTable *params = NULL; if (rsc == NULL) { return false; } // Use NULL node since pcmk__bundle_expand() uses that to set value params = pe_rsc_params(rsc, NULL, data_set); value = g_hash_table_lookup(params, XML_RSC_ATTR_REMOTE_RA_ADDR); return pcmk__str_eq(value, "#uname", pcmk__str_casei) && xml_contains_remote_node(rsc->xml); } const char * pe__add_bundle_remote_name(pe_resource_t *rsc, pe_working_set_t *data_set, xmlNode *xml, const char *field) { // REMOTE_CONTAINER_HACK: Allow remote nodes that start containers with pacemaker remote inside pe_node_t *node = NULL; pe__bundle_replica_t *replica = NULL; if (!pe__bundle_needs_remote_name(rsc, data_set)) { return NULL; } replica = replica_for_remote(rsc); if (replica == NULL) { return NULL; } node = replica->container->allocated_to; if (node == NULL) { /* If it won't be running anywhere after the * transition, go with where it's running now. */ node = pe__current_node(replica->container); } if(node == NULL) { crm_trace("Cannot determine address for bundle connection %s", rsc->id); return NULL; } crm_trace("Setting address for bundle connection %s to bundle host %s", rsc->id, node->details->uname); if(xml != NULL && field != NULL) { crm_xml_add(xml, field, node->details->uname); } return node->details->uname; } #define pe__set_bundle_mount_flags(mount_xml, flags, flags_to_set) do { \ flags = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, \ "Bundle mount", ID(mount_xml), flags, \ (flags_to_set), #flags_to_set); \ } while (0) gboolean pe__unpack_bundle(pe_resource_t *rsc, pe_working_set_t *data_set) { const char *value = NULL; xmlNode *xml_obj = NULL; xmlNode *xml_resource = NULL; pe__bundle_variant_data_t *bundle_data = NULL; bool need_log_mount = TRUE; CRM_ASSERT(rsc != NULL); pe_rsc_trace(rsc, "Processing resource %s...", rsc->id); bundle_data = calloc(1, sizeof(pe__bundle_variant_data_t)); rsc->variant_opaque = bundle_data; bundle_data->prefix = strdup(rsc->id); xml_obj = first_named_child(rsc->xml, PE__CONTAINER_AGENT_DOCKER_S); if (xml_obj != NULL) { bundle_data->agent_type = PE__CONTAINER_AGENT_DOCKER; } else { xml_obj = first_named_child(rsc->xml, PE__CONTAINER_AGENT_RKT_S); if (xml_obj != NULL) { bundle_data->agent_type = PE__CONTAINER_AGENT_RKT; } else { xml_obj = first_named_child(rsc->xml, PE__CONTAINER_AGENT_PODMAN_S); if (xml_obj != NULL) { bundle_data->agent_type = PE__CONTAINER_AGENT_PODMAN; } else { return FALSE; } } } // Use 0 for default, minimum, and invalid promoted-max value = crm_element_value(xml_obj, XML_RSC_ATTR_PROMOTED_MAX); if (value == NULL) { // @COMPAT deprecated since 2.0.0 value = crm_element_value(xml_obj, "masters"); } pcmk__scan_min_int(value, &bundle_data->promoted_max, 0); // Default replicas to promoted-max if it was specified and 1 otherwise value = crm_element_value(xml_obj, "replicas"); if ((value == NULL) && (bundle_data->promoted_max > 0)) { bundle_data->nreplicas = bundle_data->promoted_max; } else { pcmk__scan_min_int(value, &bundle_data->nreplicas, 1); } /* * Communication between containers on the same host via the * floating IPs only works if the container is started with: * --userland-proxy=false --ip-masq=false */ value = crm_element_value(xml_obj, "replicas-per-host"); pcmk__scan_min_int(value, &bundle_data->nreplicas_per_host, 1); if (bundle_data->nreplicas_per_host == 1) { pe__clear_resource_flags(rsc, pe_rsc_unique); } bundle_data->container_command = crm_element_value_copy(xml_obj, "run-command"); bundle_data->launcher_options = crm_element_value_copy(xml_obj, "options"); bundle_data->image = crm_element_value_copy(xml_obj, "image"); bundle_data->container_network = crm_element_value_copy(xml_obj, "network"); xml_obj = first_named_child(rsc->xml, "network"); if(xml_obj) { bundle_data->ip_range_start = crm_element_value_copy(xml_obj, "ip-range-start"); bundle_data->host_netmask = crm_element_value_copy(xml_obj, "host-netmask"); bundle_data->host_network = crm_element_value_copy(xml_obj, "host-interface"); bundle_data->control_port = crm_element_value_copy(xml_obj, "control-port"); value = crm_element_value(xml_obj, "add-host"); if (crm_str_to_boolean(value, &bundle_data->add_host) != 1) { bundle_data->add_host = TRUE; } for (xmlNode *xml_child = pcmk__xe_first_child(xml_obj); xml_child != NULL; xml_child = pcmk__xe_next(xml_child)) { pe__bundle_port_t *port = calloc(1, sizeof(pe__bundle_port_t)); port->source = crm_element_value_copy(xml_child, "port"); if(port->source == NULL) { port->source = crm_element_value_copy(xml_child, "range"); } else { port->target = crm_element_value_copy(xml_child, "internal-port"); } if(port->source != NULL && strlen(port->source) > 0) { if(port->target == NULL) { port->target = strdup(port->source); } bundle_data->ports = g_list_append(bundle_data->ports, port); } else { pe_err("Invalid port directive %s", ID(xml_child)); port_free(port); } } } xml_obj = first_named_child(rsc->xml, "storage"); for (xmlNode *xml_child = pcmk__xe_first_child(xml_obj); xml_child != NULL; xml_child = pcmk__xe_next(xml_child)) { const char *source = crm_element_value(xml_child, "source-dir"); const char *target = crm_element_value(xml_child, "target-dir"); const char *options = crm_element_value(xml_child, "options"); int flags = pe__bundle_mount_none; if (source == NULL) { source = crm_element_value(xml_child, "source-dir-root"); pe__set_bundle_mount_flags(xml_child, flags, pe__bundle_mount_subdir); } if (source && target) { mount_add(bundle_data, source, target, options, flags); if (strcmp(target, "/var/log") == 0) { need_log_mount = FALSE; } } else { pe_err("Invalid mount directive %s", ID(xml_child)); } } xml_obj = first_named_child(rsc->xml, "primitive"); if (xml_obj && valid_network(bundle_data)) { char *value = NULL; xmlNode *xml_set = NULL; xml_resource = create_xml_node(NULL, XML_CIB_TAG_INCARNATION); /* @COMPAT We no longer use the tag, but we need to keep it as * part of the resource name, so that bundles don't restart in a rolling * upgrade. (It also avoids needing to change regression tests.) */ crm_xml_set_id(xml_resource, "%s-%s", bundle_data->prefix, (bundle_data->promoted_max? "master" : (const char *)xml_resource->name)); xml_set = create_xml_node(xml_resource, XML_TAG_META_SETS); crm_xml_set_id(xml_set, "%s-%s-meta", bundle_data->prefix, xml_resource->name); crm_create_nvpair_xml(xml_set, NULL, XML_RSC_ATTR_ORDERED, XML_BOOLEAN_TRUE); value = pcmk__itoa(bundle_data->nreplicas); crm_create_nvpair_xml(xml_set, NULL, XML_RSC_ATTR_INCARNATION_MAX, value); free(value); value = pcmk__itoa(bundle_data->nreplicas_per_host); crm_create_nvpair_xml(xml_set, NULL, XML_RSC_ATTR_INCARNATION_NODEMAX, value); free(value); crm_create_nvpair_xml(xml_set, NULL, XML_RSC_ATTR_UNIQUE, pcmk__btoa(bundle_data->nreplicas_per_host > 1)); if (bundle_data->promoted_max) { crm_create_nvpair_xml(xml_set, NULL, XML_RSC_ATTR_PROMOTABLE, XML_BOOLEAN_TRUE); value = pcmk__itoa(bundle_data->promoted_max); crm_create_nvpair_xml(xml_set, NULL, XML_RSC_ATTR_PROMOTED_MAX, value); free(value); } //crm_xml_add(xml_obj, XML_ATTR_ID, bundle_data->prefix); add_node_copy(xml_resource, xml_obj); } else if(xml_obj) { pe_err("Cannot control %s inside %s without either ip-range-start or control-port", rsc->id, ID(xml_obj)); return FALSE; } if(xml_resource) { int lpc = 0; GList *childIter = NULL; pe_resource_t *new_rsc = NULL; pe__bundle_port_t *port = NULL; int offset = 0, max = 1024; char *buffer = NULL; if (common_unpack(xml_resource, &new_rsc, rsc, data_set) == FALSE) { pe_err("Failed unpacking resource %s", ID(rsc->xml)); if (new_rsc != NULL && new_rsc->fns != NULL) { new_rsc->fns->free(new_rsc); } return FALSE; } bundle_data->child = new_rsc; /* Currently, we always map the default authentication key location * into the same location inside the container. * * Ideally, we would respect the host's PCMK_authkey_location, but: * - it may be different on different nodes; * - the actual connection will do extra checking to make sure the key * file exists and is readable, that we can't do here on the DC * - tools such as crm_resource and crm_simulate may not have the same * environment variables as the cluster, causing operation digests to * differ * * Always using the default location inside the container is fine, * because we control the pacemaker_remote environment, and it avoids * having to pass another environment variable to the container. * * @TODO A better solution may be to have only pacemaker_remote use the * environment variable, and have the cluster nodes use a new * cluster option for key location. This would introduce the limitation * of the location being the same on all cluster nodes, but that's * reasonable. */ mount_add(bundle_data, DEFAULT_REMOTE_KEY_LOCATION, DEFAULT_REMOTE_KEY_LOCATION, NULL, pe__bundle_mount_none); if (need_log_mount) { mount_add(bundle_data, CRM_BUNDLE_DIR, "/var/log", NULL, pe__bundle_mount_subdir); } port = calloc(1, sizeof(pe__bundle_port_t)); if(bundle_data->control_port) { port->source = strdup(bundle_data->control_port); } else { /* If we wanted to respect PCMK_remote_port, we could use * crm_default_remote_port() here and elsewhere in this file instead * of DEFAULT_REMOTE_PORT. * * However, it gains nothing, since we control both the container * environment and the connection resource parameters, and the user * can use a different port if desired by setting control-port. */ port->source = pcmk__itoa(DEFAULT_REMOTE_PORT); } port->target = strdup(port->source); bundle_data->ports = g_list_append(bundle_data->ports, port); buffer = calloc(1, max+1); for (childIter = bundle_data->child->children; childIter != NULL; childIter = childIter->next) { pe__bundle_replica_t *replica = calloc(1, sizeof(pe__bundle_replica_t)); replica->child = childIter->data; replica->child->exclusive_discover = TRUE; replica->offset = lpc++; // Ensure the child's notify gets set based on the underlying primitive's value if (pcmk_is_set(replica->child->flags, pe_rsc_notify)) { pe__set_resource_flags(bundle_data->child, pe_rsc_notify); } offset += allocate_ip(bundle_data, replica, buffer+offset, max-offset); bundle_data->replicas = g_list_append(bundle_data->replicas, replica); bundle_data->attribute_target = g_hash_table_lookup(replica->child->meta, XML_RSC_ATTR_TARGET); } bundle_data->container_host_options = buffer; if (bundle_data->attribute_target) { g_hash_table_replace(rsc->meta, strdup(XML_RSC_ATTR_TARGET), strdup(bundle_data->attribute_target)); g_hash_table_replace(bundle_data->child->meta, strdup(XML_RSC_ATTR_TARGET), strdup(bundle_data->attribute_target)); } } else { // Just a naked container, no pacemaker-remote int offset = 0, max = 1024; char *buffer = calloc(1, max+1); for (int lpc = 0; lpc < bundle_data->nreplicas; lpc++) { pe__bundle_replica_t *replica = calloc(1, sizeof(pe__bundle_replica_t)); replica->offset = lpc; offset += allocate_ip(bundle_data, replica, buffer+offset, max-offset); bundle_data->replicas = g_list_append(bundle_data->replicas, replica); } bundle_data->container_host_options = buffer; } for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; if (!create_container(rsc, bundle_data, replica, data_set)) { pe_err("Failed unpacking resource %s", rsc->id); rsc->fns->free(rsc); return FALSE; } + + /* Utilization needs special handling for bundles. It makes no sense for + * the inner primitive to have utilization, because it is tied + * one-to-one to the guest node created by the container resource -- and + * there's no way to set capacities for that guest node anyway. + * + * What the user really wants is to configure utilization for the + * container. However, the schema only allows utilization for + * primitives, and the container resource is implicit anyway, so the + * user can *only* configure utilization for the inner primitive. If + * they do, move the primitive's utilization values to the container. + * + * @TODO This means that bundles without an inner primitive can't have + * utilization. An alternative might be to allow utilization values in + * the top-level bundle XML in the schema, and copy those to each + * container. + */ + if (replica->child != NULL) { + GHashTable *empty = replica->container->utilization; + + replica->container->utilization = replica->child->utilization; + replica->child->utilization = empty; + } } if (bundle_data->child) { rsc->children = g_list_append(rsc->children, bundle_data->child); } return TRUE; } static int replica_resource_active(pe_resource_t *rsc, gboolean all) { if (rsc) { gboolean child_active = rsc->fns->active(rsc, all); if (child_active && !all) { return TRUE; } else if (!child_active && all) { return FALSE; } } return -1; } gboolean pe__bundle_active(pe_resource_t *rsc, gboolean all) { pe__bundle_variant_data_t *bundle_data = NULL; GList *iter = NULL; get_bundle_variant_data(bundle_data, rsc); for (iter = bundle_data->replicas; iter != NULL; iter = iter->next) { pe__bundle_replica_t *replica = iter->data; int rsc_active; rsc_active = replica_resource_active(replica->ip, all); if (rsc_active >= 0) { return (gboolean) rsc_active; } rsc_active = replica_resource_active(replica->child, all); if (rsc_active >= 0) { return (gboolean) rsc_active; } rsc_active = replica_resource_active(replica->container, all); if (rsc_active >= 0) { return (gboolean) rsc_active; } rsc_active = replica_resource_active(replica->remote, all); if (rsc_active >= 0) { return (gboolean) rsc_active; } } /* If "all" is TRUE, we've already checked that no resources were inactive, * so return TRUE; if "all" is FALSE, we didn't find any active resources, * so return FALSE. */ return all; } /*! * \internal * \brief Find the bundle replica corresponding to a given node * * \param[in] bundle Top-level bundle resource * \param[in] node Node to search for * * \return Bundle replica if found, NULL otherwise */ pe_resource_t * pe__find_bundle_replica(const pe_resource_t *bundle, const pe_node_t *node) { pe__bundle_variant_data_t *bundle_data = NULL; CRM_ASSERT(bundle && node); get_bundle_variant_data(bundle_data, bundle); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica && replica->node); if (replica->node->details == node->details) { return replica->child; } } return NULL; } static void print_rsc_in_list(pe_resource_t *rsc, const char *pre_text, long options, void *print_data) { if (rsc != NULL) { if (options & pe_print_html) { status_print("
  • "); } rsc->fns->print(rsc, pre_text, options, print_data); if (options & pe_print_html) { status_print("
    • \n"); } } } static const char* container_agent_str(enum pe__container_agent t) { switch (t) { case PE__CONTAINER_AGENT_DOCKER: return PE__CONTAINER_AGENT_DOCKER_S; case PE__CONTAINER_AGENT_RKT: return PE__CONTAINER_AGENT_RKT_S; case PE__CONTAINER_AGENT_PODMAN: return PE__CONTAINER_AGENT_PODMAN_S; default: // PE__CONTAINER_AGENT_UNKNOWN break; } return PE__CONTAINER_AGENT_UNKNOWN_S; } static void bundle_print_xml(pe_resource_t *rsc, const char *pre_text, long options, void *print_data) { pe__bundle_variant_data_t *bundle_data = NULL; char *child_text = NULL; CRM_CHECK(rsc != NULL, return); if (pre_text == NULL) { pre_text = ""; } child_text = crm_strdup_printf("%s ", pre_text); get_bundle_variant_data(bundle_data, rsc); status_print("%sid); status_print("type=\"%s\" ", container_agent_str(bundle_data->agent_type)); status_print("image=\"%s\" ", bundle_data->image); status_print("unique=\"%s\" ", pe__rsc_bool_str(rsc, pe_rsc_unique)); status_print("managed=\"%s\" ", pe__rsc_bool_str(rsc, pe_rsc_managed)); status_print("failed=\"%s\" ", pe__rsc_bool_str(rsc, pe_rsc_failed)); status_print(">\n"); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica); status_print("%s \n", pre_text, replica->offset); print_rsc_in_list(replica->ip, child_text, options, print_data); print_rsc_in_list(replica->child, child_text, options, print_data); print_rsc_in_list(replica->container, child_text, options, print_data); print_rsc_in_list(replica->remote, child_text, options, print_data); status_print("%s \n", pre_text); } status_print("%s\n", pre_text); free(child_text); } PCMK__OUTPUT_ARGS("bundle", "unsigned int", "pe_resource_t *", "GList *", "GList *") int pe__bundle_xml(pcmk__output_t *out, va_list args) { unsigned int show_opts = va_arg(args, unsigned int); pe_resource_t *rsc = va_arg(args, pe_resource_t *); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); pe__bundle_variant_data_t *bundle_data = NULL; int rc = pcmk_rc_no_output; gboolean printed_header = FALSE; gboolean print_everything = TRUE; CRM_ASSERT(rsc != NULL); get_bundle_variant_data(bundle_data, rsc); if (rsc->fns->is_filtered(rsc, only_rsc, TRUE)) { return rc; } print_everything = pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; char *id = NULL; gboolean print_ip, print_child, print_ctnr, print_remote; CRM_ASSERT(replica); if (pcmk__rsc_filtered_by_node(replica->container, only_node)) { continue; } print_ip = replica->ip != NULL && !replica->ip->fns->is_filtered(replica->ip, only_rsc, print_everything); print_child = replica->child != NULL && !replica->child->fns->is_filtered(replica->child, only_rsc, print_everything); print_ctnr = !replica->container->fns->is_filtered(replica->container, only_rsc, print_everything); print_remote = replica->remote != NULL && !replica->remote->fns->is_filtered(replica->remote, only_rsc, print_everything); if (!print_everything && !print_ip && !print_child && !print_ctnr && !print_remote) { continue; } if (!printed_header) { printed_header = TRUE; rc = pe__name_and_nvpairs_xml(out, true, "bundle", 6, "id", rsc->id, "type", container_agent_str(bundle_data->agent_type), "image", bundle_data->image, "unique", pe__rsc_bool_str(rsc, pe_rsc_unique), "managed", pe__rsc_bool_str(rsc, pe_rsc_managed), "failed", pe__rsc_bool_str(rsc, pe_rsc_failed)); CRM_ASSERT(rc == pcmk_rc_ok); } id = pcmk__itoa(replica->offset); rc = pe__name_and_nvpairs_xml(out, true, "replica", 1, "id", id); free(id); CRM_ASSERT(rc == pcmk_rc_ok); if (print_ip) { out->message(out, crm_map_element_name(replica->ip->xml), show_opts, replica->ip, only_node, only_rsc); } if (print_child) { out->message(out, crm_map_element_name(replica->child->xml), show_opts, replica->child, only_node, only_rsc); } if (print_ctnr) { out->message(out, crm_map_element_name(replica->container->xml), show_opts, replica->container, only_node, only_rsc); } if (print_remote) { out->message(out, crm_map_element_name(replica->remote->xml), show_opts, replica->remote, only_node, only_rsc); } pcmk__output_xml_pop_parent(out); // replica } if (printed_header) { pcmk__output_xml_pop_parent(out); // bundle } return rc; } static void pe__bundle_replica_output_html(pcmk__output_t *out, pe__bundle_replica_t *replica, pe_node_t *node, unsigned int show_opts) { pe_resource_t *rsc = replica->child; int offset = 0; char buffer[LINE_MAX]; if(rsc == NULL) { rsc = replica->container; } if (replica->remote) { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->remote)); } else { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->container)); } if (replica->ipaddr) { offset += snprintf(buffer + offset, LINE_MAX - offset, " (%s)", replica->ipaddr); } pe__common_output_html(out, rsc, buffer, node, show_opts); } PCMK__OUTPUT_ARGS("bundle", "unsigned int", "pe_resource_t *", "GList *", "GList *") int pe__bundle_html(pcmk__output_t *out, va_list args) { unsigned int show_opts = va_arg(args, unsigned int); pe_resource_t *rsc = va_arg(args, pe_resource_t *); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); pe__bundle_variant_data_t *bundle_data = NULL; int rc = pcmk_rc_no_output; gboolean print_everything = TRUE; CRM_ASSERT(rsc != NULL); get_bundle_variant_data(bundle_data, rsc); if (rsc->fns->is_filtered(rsc, only_rsc, TRUE)) { return rc; } print_everything = pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; gboolean print_ip, print_child, print_ctnr, print_remote; CRM_ASSERT(replica); if (pcmk__rsc_filtered_by_node(replica->container, only_node)) { continue; } print_ip = replica->ip != NULL && !replica->ip->fns->is_filtered(replica->ip, only_rsc, print_everything); print_child = replica->child != NULL && !replica->child->fns->is_filtered(replica->child, only_rsc, print_everything); print_ctnr = !replica->container->fns->is_filtered(replica->container, only_rsc, print_everything); print_remote = replica->remote != NULL && !replica->remote->fns->is_filtered(replica->remote, only_rsc, print_everything); if (pcmk_is_set(show_opts, pcmk_show_implicit_rscs) || (print_everything == FALSE && (print_ip || print_child || print_ctnr || print_remote))) { /* The text output messages used below require pe_print_implicit to * be set to do anything. */ unsigned int new_show_opts = show_opts | pcmk_show_implicit_rscs; PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Container bundle%s: %s [%s]%s%s", (bundle_data->nreplicas > 1)? " set" : "", rsc->id, bundle_data->image, pcmk_is_set(rsc->flags, pe_rsc_unique) ? " (unique)" : "", pcmk_is_set(rsc->flags, pe_rsc_managed) ? "" : " (unmanaged)"); if (pcmk__list_of_multiple(bundle_data->replicas)) { out->begin_list(out, NULL, NULL, "Replica[%d]", replica->offset); } if (print_ip) { out->message(out, crm_map_element_name(replica->ip->xml), new_show_opts, replica->ip, only_node, only_rsc); } if (print_child) { out->message(out, crm_map_element_name(replica->child->xml), new_show_opts, replica->child, only_node, only_rsc); } if (print_ctnr) { out->message(out, crm_map_element_name(replica->container->xml), new_show_opts, replica->container, only_node, only_rsc); } if (print_remote) { out->message(out, crm_map_element_name(replica->remote->xml), new_show_opts, replica->remote, only_node, only_rsc); } if (pcmk__list_of_multiple(bundle_data->replicas)) { out->end_list(out); } } else if (print_everything == FALSE && !(print_ip || print_child || print_ctnr || print_remote)) { continue; } else { PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Container bundle%s: %s [%s]%s%s", (bundle_data->nreplicas > 1)? " set" : "", rsc->id, bundle_data->image, pcmk_is_set(rsc->flags, pe_rsc_unique) ? " (unique)" : "", pcmk_is_set(rsc->flags, pe_rsc_managed) ? "" : " (unmanaged)"); pe__bundle_replica_output_html(out, replica, pe__current_node(replica->container), show_opts); } } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } static void pe__bundle_replica_output_text(pcmk__output_t *out, pe__bundle_replica_t *replica, pe_node_t *node, unsigned int show_opts) { pe_resource_t *rsc = replica->child; int offset = 0; char buffer[LINE_MAX]; if(rsc == NULL) { rsc = replica->container; } if (replica->remote) { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->remote)); } else { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->container)); } if (replica->ipaddr) { offset += snprintf(buffer + offset, LINE_MAX - offset, " (%s)", replica->ipaddr); } pe__common_output_text(out, rsc, buffer, node, show_opts); } PCMK__OUTPUT_ARGS("bundle", "unsigned int", "pe_resource_t *", "GList *", "GList *") int pe__bundle_text(pcmk__output_t *out, va_list args) { unsigned int show_opts = va_arg(args, unsigned int); pe_resource_t *rsc = va_arg(args, pe_resource_t *); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); pe__bundle_variant_data_t *bundle_data = NULL; int rc = pcmk_rc_no_output; gboolean print_everything = TRUE; get_bundle_variant_data(bundle_data, rsc); CRM_ASSERT(rsc != NULL); if (rsc->fns->is_filtered(rsc, only_rsc, TRUE)) { return rc; } print_everything = pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; gboolean print_ip, print_child, print_ctnr, print_remote; CRM_ASSERT(replica); if (pcmk__rsc_filtered_by_node(replica->container, only_node)) { continue; } print_ip = replica->ip != NULL && !replica->ip->fns->is_filtered(replica->ip, only_rsc, print_everything); print_child = replica->child != NULL && !replica->child->fns->is_filtered(replica->child, only_rsc, print_everything); print_ctnr = !replica->container->fns->is_filtered(replica->container, only_rsc, print_everything); print_remote = replica->remote != NULL && !replica->remote->fns->is_filtered(replica->remote, only_rsc, print_everything); if (pcmk_is_set(show_opts, pcmk_show_implicit_rscs) || (print_everything == FALSE && (print_ip || print_child || print_ctnr || print_remote))) { /* The text output messages used below require pe_print_implicit to * be set to do anything. */ unsigned int new_show_opts = show_opts | pcmk_show_implicit_rscs; PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Container bundle%s: %s [%s]%s%s", (bundle_data->nreplicas > 1)? " set" : "", rsc->id, bundle_data->image, pcmk_is_set(rsc->flags, pe_rsc_unique) ? " (unique)" : "", pcmk_is_set(rsc->flags, pe_rsc_managed) ? "" : " (unmanaged)"); if (pcmk__list_of_multiple(bundle_data->replicas)) { out->list_item(out, NULL, "Replica[%d]", replica->offset); } out->begin_list(out, NULL, NULL, NULL); if (print_ip) { out->message(out, crm_map_element_name(replica->ip->xml), new_show_opts, replica->ip, only_node, only_rsc); } if (print_child) { out->message(out, crm_map_element_name(replica->child->xml), new_show_opts, replica->child, only_node, only_rsc); } if (print_ctnr) { out->message(out, crm_map_element_name(replica->container->xml), new_show_opts, replica->container, only_node, only_rsc); } if (print_remote) { out->message(out, crm_map_element_name(replica->remote->xml), new_show_opts, replica->remote, only_node, only_rsc); } out->end_list(out); } else if (print_everything == FALSE && !(print_ip || print_child || print_ctnr || print_remote)) { continue; } else { PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Container bundle%s: %s [%s]%s%s", (bundle_data->nreplicas > 1)? " set" : "", rsc->id, bundle_data->image, pcmk_is_set(rsc->flags, pe_rsc_unique) ? " (unique)" : "", pcmk_is_set(rsc->flags, pe_rsc_managed) ? "" : " (unmanaged)"); pe__bundle_replica_output_text(out, replica, pe__current_node(replica->container), show_opts); } } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } static void print_bundle_replica(pe__bundle_replica_t *replica, const char *pre_text, long options, void *print_data) { pe_node_t *node = NULL; pe_resource_t *rsc = replica->child; int offset = 0; char buffer[LINE_MAX]; if(rsc == NULL) { rsc = replica->container; } if (replica->remote) { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->remote)); } else { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->container)); } if (replica->ipaddr) { offset += snprintf(buffer + offset, LINE_MAX - offset, " (%s)", replica->ipaddr); } node = pe__current_node(replica->container); common_print(rsc, pre_text, buffer, node, options, print_data); } void pe__print_bundle(pe_resource_t *rsc, const char *pre_text, long options, void *print_data) { pe__bundle_variant_data_t *bundle_data = NULL; char *child_text = NULL; CRM_CHECK(rsc != NULL, return); if (options & pe_print_xml) { bundle_print_xml(rsc, pre_text, options, print_data); return; } get_bundle_variant_data(bundle_data, rsc); if (pre_text == NULL) { pre_text = " "; } status_print("%sContainer bundle%s: %s [%s]%s%s\n", pre_text, ((bundle_data->nreplicas > 1)? " set" : ""), rsc->id, bundle_data->image, pcmk_is_set(rsc->flags, pe_rsc_unique) ? " (unique)" : "", pcmk_is_set(rsc->flags, pe_rsc_managed) ? "" : " (unmanaged)"); if (options & pe_print_html) { status_print("
    \n
      \n"); } for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica); if (options & pe_print_html) { status_print("
    • "); } if (pcmk_is_set(options, pe_print_implicit)) { child_text = crm_strdup_printf(" %s", pre_text); if (pcmk__list_of_multiple(bundle_data->replicas)) { status_print(" %sReplica[%d]\n", pre_text, replica->offset); } if (options & pe_print_html) { status_print("
      \n
        \n"); } print_rsc_in_list(replica->ip, child_text, options, print_data); print_rsc_in_list(replica->container, child_text, options, print_data); print_rsc_in_list(replica->remote, child_text, options, print_data); print_rsc_in_list(replica->child, child_text, options, print_data); if (options & pe_print_html) { status_print("
      \n"); } } else { child_text = crm_strdup_printf("%s ", pre_text); print_bundle_replica(replica, child_text, options, print_data); } free(child_text); if (options & pe_print_html) { status_print("
      • \n"); } } if (options & pe_print_html) { status_print("
    \n"); } } static void free_bundle_replica(pe__bundle_replica_t *replica) { if (replica == NULL) { return; } if (replica->node) { free(replica->node); replica->node = NULL; } if (replica->ip) { free_xml(replica->ip->xml); replica->ip->xml = NULL; replica->ip->fns->free(replica->ip); replica->ip = NULL; } if (replica->container) { free_xml(replica->container->xml); replica->container->xml = NULL; replica->container->fns->free(replica->container); replica->container = NULL; } if (replica->remote) { free_xml(replica->remote->xml); replica->remote->xml = NULL; replica->remote->fns->free(replica->remote); replica->remote = NULL; } free(replica->ipaddr); free(replica); } void pe__free_bundle(pe_resource_t *rsc) { pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(rsc != NULL, return); get_bundle_variant_data(bundle_data, rsc); pe_rsc_trace(rsc, "Freeing %s", rsc->id); free(bundle_data->prefix); free(bundle_data->image); free(bundle_data->control_port); free(bundle_data->host_network); free(bundle_data->host_netmask); free(bundle_data->ip_range_start); free(bundle_data->container_network); free(bundle_data->launcher_options); free(bundle_data->container_command); free(bundle_data->container_host_options); g_list_free_full(bundle_data->replicas, (GDestroyNotify) free_bundle_replica); g_list_free_full(bundle_data->mounts, (GDestroyNotify)mount_free); g_list_free_full(bundle_data->ports, (GDestroyNotify)port_free); g_list_free(rsc->children); if(bundle_data->child) { free_xml(bundle_data->child->xml); bundle_data->child->xml = NULL; bundle_data->child->fns->free(bundle_data->child); } common_free(rsc); } enum rsc_role_e pe__bundle_resource_state(const pe_resource_t *rsc, gboolean current) { enum rsc_role_e container_role = RSC_ROLE_UNKNOWN; return container_role; } /*! * \brief Get the number of configured replicas in a bundle * * \param[in] rsc Bundle resource * * \return Number of configured replicas, or 0 on error */ int pe_bundle_replicas(const pe_resource_t *rsc) { if ((rsc == NULL) || (rsc->variant != pe_container)) { return 0; } else { pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, rsc); return bundle_data->nreplicas; } } void pe__count_bundle(pe_resource_t *rsc) { pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, rsc); for (GList *item = bundle_data->replicas; item != NULL; item = item->next) { pe__bundle_replica_t *replica = item->data; if (replica->ip) { replica->ip->fns->count(replica->ip); } if (replica->child) { replica->child->fns->count(replica->child); } if (replica->container) { replica->container->fns->count(replica->container); } if (replica->remote) { replica->remote->fns->count(replica->remote); } } } gboolean pe__bundle_is_filtered(pe_resource_t *rsc, GList *only_rsc, gboolean check_parent) { gboolean passes = FALSE; pe__bundle_variant_data_t *bundle_data = NULL; if (pcmk__str_in_list(rsc_printable_id(rsc), only_rsc, pcmk__str_star_matches)) { passes = TRUE; } else { get_bundle_variant_data(bundle_data, rsc); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; if (replica->ip != NULL && !replica->ip->fns->is_filtered(replica->ip, only_rsc, FALSE)) { passes = TRUE; break; } else if (replica->child != NULL && !replica->child->fns->is_filtered(replica->child, only_rsc, FALSE)) { passes = TRUE; break; } else if (!replica->container->fns->is_filtered(replica->container, only_rsc, FALSE)) { passes = TRUE; break; } else if (replica->remote != NULL && !replica->remote->fns->is_filtered(replica->remote, only_rsc, FALSE)) { passes = TRUE; break; } } } return !passes; }