From f5d88938955f63935058b7cc2d706a12e6ea1121 Mon Sep 17 00:00:00 2001
From: Ken Gaillot <kgaillot@redhat.com>
Date: Fri, 6 Dec 2019 11:57:59 -0600
Subject: [PATCH 07/18] Low: scheduler: respect shutdown locks when placing
active resources
Use new pe_resource_t members to indicate that a resource is locked to a
particular node.
For active resources (i.e. in the transition where the node is scheduled for
shutdown), these are connected by checking each lockable resource for whether
it is running on a single clean node that is shutting down.
When applying constraints, place -INFINITY location constraints for locked
resources on all nodes other than the lock node.
(Inactive resources -- i.e. in later transitions after the node is shut down --
are not yet locked.)
---
include/crm/pengine/pe_types.h | 2 +
lib/pacemaker/pcmk_sched_allocate.c | 87 +++++++++++++++++++++++++++++++++++++
2 files changed, 89 insertions(+)
diff --git a/include/crm/pengine/pe_types.h b/include/crm/pengine/pe_types.h
index 8a735a3..123d8ef 100644
--- a/include/crm/pengine/pe_types.h
+++ b/include/crm/pengine/pe_types.h
@@ -354,6 +354,8 @@ struct pe_resource_s {
GListPtr fillers;
pe_node_t *pending_node; // Node on which pending_task is happening
+ pe_node_t *lock_node; // Resource is shutdown-locked to this node
+ time_t lock_time; // When shutdown lock started
#if ENABLE_VERSIONED_ATTRS
xmlNode *versioned_parameters;
diff --git a/lib/pacemaker/pcmk_sched_allocate.c b/lib/pacemaker/pcmk_sched_allocate.c
index fc2f4cf..0314f1b 100644
--- a/lib/pacemaker/pcmk_sched_allocate.c
+++ b/lib/pacemaker/pcmk_sched_allocate.c
@@ -977,6 +977,87 @@ rsc_discover_filter(resource_t *rsc, node_t *node)
}
}
+static time_t
+shutdown_time(pe_node_t *node, pe_working_set_t *data_set)
+{
+ const char *shutdown = pe_node_attribute_raw(node, XML_CIB_ATTR_SHUTDOWN);
+ time_t result = 0;
+
+ if (shutdown) {
+ errno = 0;
+ result = (time_t) crm_int_helper(shutdown, NULL);
+ if (errno != 0) {
+ result = 0;
+ }
+ }
+ return result? result : get_effective_time(data_set);
+}
+
+static void
+apply_shutdown_lock(pe_resource_t *rsc, pe_working_set_t *data_set)
+{
+ const char *class;
+
+ // Only primitives and (uncloned) groups may be locked
+ if (rsc->variant == pe_group) {
+ for (GList *item = rsc->children; item != NULL;
+ item = item->next) {
+ apply_shutdown_lock((pe_resource_t *) item->data, data_set);
+ }
+ } else if (rsc->variant != pe_native) {
+ return;
+ }
+
+ // Fence devices and remote connections can't be locked
+ class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
+ if ((class == NULL) || !strcmp(class, PCMK_RESOURCE_CLASS_STONITH)
+ || pe__resource_is_remote_conn(rsc, data_set)) {
+ return;
+ }
+
+ // Only a resource active on exactly one node can be locked
+ if (pcmk__list_of_1(rsc->running_on)) {
+ pe_node_t *node = rsc->running_on->data;
+
+ if (node->details->shutdown) {
+ if (node->details->unclean) {
+ pe_rsc_debug(rsc, "Not locking %s to unclean %s for shutdown",
+ rsc->id, node->details->uname);
+ } else {
+ rsc->lock_node = node;
+ rsc->lock_time = shutdown_time(node, data_set);
+ }
+ }
+ }
+
+ if (rsc->lock_node == NULL) {
+ // No lock needed
+ return;
+ }
+
+ if (data_set->shutdown_lock > 0) {
+ time_t lock_expiration = rsc->lock_time + data_set->shutdown_lock;
+
+ pe_rsc_info(rsc, "Locking %s to %s due to shutdown (expires @%lld)",
+ rsc->id, rsc->lock_node->details->uname,
+ (long long) lock_expiration);
+ pe__update_recheck_time(++lock_expiration, data_set);
+ } else {
+ pe_rsc_info(rsc, "Locking %s to %s due to shutdown",
+ rsc->id, rsc->lock_node->details->uname);
+ }
+
+ // If resource is locked to one node, ban it from all other nodes
+ for (GList *item = data_set->nodes; item != NULL; item = item->next) {
+ pe_node_t *node = item->data;
+
+ if (strcmp(node->details->uname, rsc->lock_node->details->uname)) {
+ resource_location(rsc, node, -CRM_SCORE_INFINITY,
+ XML_CONFIG_ATTR_SHUTDOWN_LOCK, data_set);
+ }
+ }
+}
+
/*
* Count how many valid nodes we have (so we know the maximum number of
* colors we can resolve).
@@ -988,6 +1069,12 @@ stage2(pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
+ if (is_set(data_set->flags, pe_flag_shutdown_lock)) {
+ for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) {
+ apply_shutdown_lock((pe_resource_t *) gIter->data, data_set);
+ }
+ }
+
for (gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) {
node_t *node = (node_t *) gIter->data;
--
1.8.3.1