No OneTemporary
Actions

Size

55 KB

Referenced Files

None

Subscribers

None

View Options

	diff --git a/include/crm/cluster.h b/include/crm/cluster.h
	index 1e69961802..7ae0a82f95 100644
	--- a/include/crm/cluster.h
	+++ b/include/crm/cluster.h
	@@ -1,234 +1,233 @@
	/*
	* Copyright 2004-2024 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.
	*/

	#ifndef PCMK__CRM_CLUSTER__H
	# define PCMK__CRM_CLUSTER__H

	# include <stdint.h> // uint32_t, uint64_t
	# include <glib.h> // gboolean, GHashTable
	# include <libxml/tree.h> // xmlNode
	# include <crm/common/xml.h>
	# include <crm/common/util.h>

	#ifdef __cplusplus
	extern "C" {
	#endif

	# if SUPPORT_COROSYNC
	# include <corosync/cpg.h>
	# endif

	extern gboolean crm_have_quorum;
	extern GHashTable *crm_peer_cache;
	extern GHashTable *crm_remote_peer_cache;
	extern unsigned long long crm_peer_seq;

	#define CRM_NODE_LOST "lost"
	#define CRM_NODE_MEMBER "member"

	enum crm_join_phase {
	/* @COMPAT: crm_join_nack_quiet can be replaced by crm_node_t:user_data
	* at a compatibility break.
	*/
	//! Not allowed to join, but don't send a nack message
	crm_join_nack_quiet = -2,

	crm_join_nack = -1,
	crm_join_none = 0,
	crm_join_welcomed = 1,
	crm_join_integrated = 2,
	crm_join_finalized = 3,
	crm_join_confirmed = 4,
	};

	enum crm_node_flags {
	/* node is not a cluster node and should not be considered for cluster membership */
	crm_remote_node = 0x0001,

	/* node's cache entry is dirty */
	crm_node_dirty = 0x0010,
	};

	typedef struct crm_peer_node_s {
	char *uname; // Node name as known to cluster

	/* @COMPAT This is less than ideal since the value is not a valid XML ID
	* (for Corosync, it's the string equivalent of the node's numeric node ID,
	* but XML IDs can't start with a number) and the three elements should have
	* different IDs.
	*
	* Ideally, we would use something like node-NODEID, node_state-NODEID, and
	* transient_attributes-NODEID as the element IDs. Unfortunately changing it
	* would be impractical due to backward compatibility; older nodes in a
	* rolling upgrade will always write and expect the value in the old format.
	*
	* This is also named poorly, since the value is not a UUID, but at least
	* that can be changed at an API compatibility break.
	*/
	/*! Value of the PCMK_XA_ID XML attribute to use with the node's
	* PCMK_XE_NODE, PCMK_XE_NODE_STATE, and PCMK_XE_TRANSIENT_ATTRIBUTES
	* XML elements in the CIB
	*/
	char *uuid;

	char *state; // @TODO change to enum
	uint64_t flags; // Bitmask of crm_node_flags
	uint64_t last_seen; // Only needed by cluster nodes
	uint32_t processes; // @TODO most not needed, merge into flags

	/* @TODO When we can break public API compatibility, we can make the rest of
	* these members separate structs and use void *cluster_data and
	* void *user_data here instead, to abstract the cluster layer further.
	*/

	// Currently only needed by corosync stack
	uint32_t id; // Node ID
	time_t when_lost; // When CPG membership was last lost

	// Only used by controller
	enum crm_join_phase join;
	char *expected;

	time_t peer_lost;
	char *conn_host;

	time_t when_member; // Since when node has been a cluster member
	time_t when_online; // Since when peer has been online in CPG
	} crm_node_t;

	void crm_peer_init(void);
	void crm_peer_destroy(void);

	typedef struct crm_cluster_s {
	char *uuid;
	char *uname;
	uint32_t nodeid;

	void (*destroy) (gpointer);

	# if SUPPORT_COROSYNC
	/* @TODO When we can break public API compatibility, make these members a
	* separate struct and use void *cluster_data here instead, to abstract the
	* cluster layer further.
	*/
	struct cpg_name group;
	cpg_callbacks_t cpg;
	cpg_handle_t cpg_handle;
	# endif

	} crm_cluster_t;

	int pcmk_cluster_connect(crm_cluster_t *cluster);
	int pcmk_cluster_disconnect(crm_cluster_t *cluster);

	crm_cluster_t *pcmk_cluster_new(void);
	void pcmk_cluster_free(crm_cluster_t *cluster);

	enum crm_ais_msg_class {
	crm_class_cluster = 0,
	};

	enum crm_ais_msg_types {
	crm_msg_none = 0,
	crm_msg_ais = 1,
	crm_msg_lrmd = 2,
	crm_msg_cib = 3,
	crm_msg_crmd = 4,
	crm_msg_attrd = 5,
	crm_msg_stonithd = 6,
	crm_msg_te = 7,
	crm_msg_pe = 8,
	crm_msg_stonith_ng = 9,
	};

	gboolean send_cluster_message(const crm_node_t *node,
	enum crm_ais_msg_types service,
	const xmlNode *data, gboolean ordered);

	guint crm_active_peers(void);
	-gboolean crm_is_peer_active(const crm_node_t * node);
	guint reap_crm_member(uint32_t id, const char *name);

	# if SUPPORT_COROSYNC
	uint32_t get_local_nodeid(cpg_handle_t handle);

	void pcmk_cpg_membership(cpg_handle_t handle,
	const struct cpg_name *groupName,
	const struct cpg_address *member_list, size_t member_list_entries,
	const struct cpg_address *left_list, size_t left_list_entries,
	const struct cpg_address *joined_list, size_t joined_list_entries);
	gboolean crm_is_corosync_peer_active(const crm_node_t * node);
	gboolean send_cluster_text(enum crm_ais_msg_class msg_class, const char *data,
	gboolean local, const crm_node_t *node,
	enum crm_ais_msg_types dest);
	char pcmk_message_common_cs(cpg_handle_t handle, uint32_t nodeid, uint32_t pid, void msg,
	uint32_t kind, const char *from);
	# endif

	const char crm_peer_uuid(crm_node_t node);
	const char crm_peer_uname(const char uuid);

	enum crm_status_type {
	crm_status_uname,
	crm_status_nstate,
	crm_status_processes,
	};

	enum crm_ais_msg_types text2msg_type(const char *text);
	void crm_set_status_callback(void (dispatch) (enum crm_status_type, crm_node_t , const void *));
	void crm_set_autoreap(gboolean autoreap);

	enum cluster_type_e {
	pcmk_cluster_unknown = 0x0001,
	pcmk_cluster_invalid = 0x0002,
	// 0x0004 was heartbeat
	// 0x0010 was corosync 1 with plugin
	pcmk_cluster_corosync = 0x0020,
	// 0x0040 was corosync 1 with CMAN
	};

	enum cluster_type_e get_cluster_type(void);
	const char *name_for_cluster_type(enum cluster_type_e type);

	gboolean is_corosync_cluster(void);

	const char *get_local_node_name(void);
	char *get_node_name(uint32_t nodeid);

	/*!
	* \brief Get log-friendly string equivalent of a join phase
	*
	* \param[in] phase Join phase
	*
	* \return Log-friendly string equivalent of \p phase
	*/
	static inline const char *
	crm_join_phase_str(enum crm_join_phase phase)
	{
	switch (phase) {
	case crm_join_nack_quiet: return "nack_quiet";
	case crm_join_nack: return "nack";
	case crm_join_none: return "none";
	case crm_join_welcomed: return "welcomed";
	case crm_join_integrated: return "integrated";
	case crm_join_finalized: return "finalized";
	case crm_join_confirmed: return "confirmed";
	default: return "invalid";
	}
	}

	#if !defined(PCMK_ALLOW_DEPRECATED) \|\| (PCMK_ALLOW_DEPRECATED == 1)
	#include <crm/cluster/compat.h>
	#endif

	#ifdef __cplusplus
	}
	#endif

	#endif
	diff --git a/include/crm/cluster/compat.h b/include/crm/cluster/compat.h
	index 66d8acc2e4..79ca505288 100644
	--- a/include/crm/cluster/compat.h
	+++ b/include/crm/cluster/compat.h
	@@ -1,84 +1,89 @@
	/*
	* Copyright 2004-2024 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.
	*/

	#ifndef PCMK__CRM_CLUSTER_COMPAT__H
	# define PCMK__CRM_CLUSTER_COMPAT__H

	+#include <glib.h> // gboolean, guint
	#include <libxml/tree.h> // xmlNode
	+
	#include <crm/cluster.h> // crm_node_t

	#ifdef __cplusplus
	extern "C" {
	#endif

	/**
	* \file
	* \brief Deprecated Pacemaker cluster API
	* \ingroup cluster
	* \deprecated Do not include this header directly. The cluster APIs in this
	* header, and the header itself, will be removed in a future
	* release.
	*/

	// \deprecated Do not use
	enum crm_get_peer_flags {
	CRM_GET_PEER_CLUSTER = 0x0001,
	CRM_GET_PEER_REMOTE = 0x0002,
	CRM_GET_PEER_ANY = CRM_GET_PEER_CLUSTER\|CRM_GET_PEER_REMOTE,
	};

	// \deprecated Do not use Pacemaker for cluster node cacheing
	crm_node_t crm_get_peer(unsigned int id, const char uname);

	// \deprecated Do not use Pacemaker for cluster node cacheing
	crm_node_t crm_get_peer_full(unsigned int id, const char uname, int flags);

	// \deprecated Use stonith_api_kick() from libstonithd instead
	int crm_terminate_member(int nodeid, const char uname, void unused);

	// \deprecated Use stonith_api_kick() from libstonithd instead
	int crm_terminate_member_no_mainloop(int nodeid, const char *uname,
	int *connection);

	// \deprecated Use crm_xml_add(xml, attr, crm_peer_uuid(node)) instead
	void set_uuid(xmlNode xml, const char attr, crm_node_t *node);

	#if SUPPORT_COROSYNC

	// \deprecated Do not use
	gboolean cluster_connect_cpg(crm_cluster_t *cluster);

	// \deprecated Do not use
	void cluster_disconnect_cpg(crm_cluster_t *cluster);

	#endif // SUPPORT_COROSYNC

	// \deprecated Use \c pcmk_cluster_connect() instead
	gboolean crm_cluster_connect(crm_cluster_t *cluster);

	// \deprecated Use \c pcmk_cluster_disconnect() instead
	void crm_cluster_disconnect(crm_cluster_t *cluster);

	// \deprecated Do not use
	int crm_remote_peer_cache_size(void);

	// \deprecated Do not use
	void crm_remote_peer_cache_refresh(xmlNode *cib);

	// \deprecated Do not use
	crm_node_t crm_remote_peer_get(const char node_name);

	// \deprecated Do not use
	void crm_remote_peer_cache_remove(const char *node_name);

	+// \deprecated Do not use
	+gboolean crm_is_peer_active(const crm_node_t *node);
	+
	#ifdef __cplusplus
	}
	#endif

	#endif // PCMK_CLUSTER_COMPAT__H
	diff --git a/lib/cluster/membership.c b/lib/cluster/membership.c
	index 679244e5b0..63fa0c10f4 100644
	--- a/lib/cluster/membership.c
	+++ b/lib/cluster/membership.c
	@@ -1,1441 +1,1441 @@
	/*
	* Copyright 2004-2024 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 <crm_internal.h>

	#ifndef _GNU_SOURCE
	# define _GNU_SOURCE
	#endif

	#include <sys/param.h>
	#include <sys/types.h>
	#include <stdio.h>
	#include <unistd.h>
	#include <string.h>
	#include <glib.h>
	#include <crm/common/ipc.h>
	#include <crm/common/xml_internal.h>
	#include <crm/cluster/internal.h>
	#include <crm/common/xml.h>
	#include <crm/stonith-ng.h>
	#include "crmcluster_private.h"

	/* The peer cache remembers cluster nodes that have been seen.
	* This is managed mostly automatically by libcluster, based on
	* cluster membership events.
	*
	* Because cluster nodes can have conflicting names or UUIDs,
	* the hash table key is a uniquely generated ID.
	*/
	GHashTable *crm_peer_cache = NULL;

	/*
	* The remote peer cache tracks pacemaker_remote nodes. While the
	* value has the same type as the peer cache's, it is tracked separately for
	* three reasons: pacemaker_remote nodes can't have conflicting names or UUIDs,
	* so the name (which is also the UUID) is used as the hash table key; there
	* is no equivalent of membership events, so management is not automatic; and
	* most users of the peer cache need to exclude pacemaker_remote nodes.
	*
	* That said, using a single cache would be more logical and less error-prone,
	* so it would be a good idea to merge them one day.
	*
	* libcluster provides two avenues for populating the cache:
	* pcmk__cluster_lookup_remote_node() and pcmk__cluster_forget_remote_node()
	* directly manage it, while refresh_remote_nodes() populates it via the CIB.
	*/
	GHashTable *crm_remote_peer_cache = NULL;

	/*
	* The known node cache tracks cluster and remote nodes that have been seen in
	* the CIB. It is useful mainly when a caller needs to know about a node that
	* may no longer be in the membership, but doesn't want to add the node to the
	* main peer cache tables.
	*/
	static GHashTable *known_node_cache = NULL;

	unsigned long long crm_peer_seq = 0;
	gboolean crm_have_quorum = FALSE;
	static gboolean crm_autoreap = TRUE;

	// Flag setting and clearing for crm_node_t:flags

	#define set_peer_flags(peer, flags_to_set) do { \
	(peer)->flags = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, \
	"Peer", (peer)->uname, \
	(peer)->flags, (flags_to_set), \
	#flags_to_set); \
	} while (0)

	#define clear_peer_flags(peer, flags_to_clear) do { \
	(peer)->flags = pcmk__clear_flags_as(__func__, __LINE__, \
	LOG_TRACE, \
	"Peer", (peer)->uname, \
	(peer)->flags, (flags_to_clear), \
	#flags_to_clear); \
	} while (0)

	static void update_peer_uname(crm_node_t node, const char uname);
	static crm_node_t find_known_node(const char id, const char *uname);

	/*!
	* \internal
	* \brief Get the number of Pacemaker Remote nodes that have been seen
	*
	* \return Number of cached Pacemaker Remote nodes
	*/
	unsigned int
	pcmk__cluster_num_remote_nodes(void)
	{
	if (crm_remote_peer_cache == NULL) {
	return 0U;
	}
	return g_hash_table_size(crm_remote_peer_cache);
	}

	/*!
	* \internal
	* \brief Get a remote node cache entry, creating it if necessary
	*
	* \param[in] node_name Name of remote node
	*
	* \return Cache entry for node on success, or \c NULL (and set \c errno)
	* otherwise
	*
	* \note When creating a new entry, this will leave the node state undetermined.
	* The caller should also call \c pcmk__update_peer_state() if the state
	* is known.
	* \note Because this can add and remove cache entries, callers should not
	* assume any previously obtained cache entry pointers remain valid.
	*/
	crm_node_t *
	pcmk__cluster_lookup_remote_node(const char *node_name)
	{
	crm_node_t *node;
	char *node_name_copy = NULL;

	if (node_name == NULL) {
	errno = EINVAL;
	return NULL;
	}

	/* It's theoretically possible that the node was added to the cluster peer
	* cache before it was known to be a Pacemaker Remote node. Remove that
	* entry unless it has a node ID, which means the name actually is
	* associated with a cluster node. (@TODO return an error in that case?)
	*/
	node = pcmk__search_node_caches(0, node_name, pcmk__node_search_cluster);
	if ((node != NULL) && (node->uuid == NULL)) {
	/* node_name could be a pointer into the cache entry being removed, so
	* reassign it to a copy before the original gets freed
	*/
	node_name_copy = strdup(node_name);
	if (node_name_copy == NULL) {
	errno = ENOMEM;
	return NULL;
	}
	node_name = node_name_copy;
	reap_crm_member(0, node_name);
	}

	/* Return existing cache entry if one exists */
	node = g_hash_table_lookup(crm_remote_peer_cache, node_name);
	if (node) {
	free(node_name_copy);
	return node;
	}

	/* Allocate a new entry */
	node = calloc(1, sizeof(crm_node_t));
	if (node == NULL) {
	free(node_name_copy);
	return NULL;
	}

	/* Populate the essential information */
	set_peer_flags(node, crm_remote_node);
	node->uuid = strdup(node_name);
	if (node->uuid == NULL) {
	free(node);
	errno = ENOMEM;
	free(node_name_copy);
	return NULL;
	}

	/* Add the new entry to the cache */
	g_hash_table_replace(crm_remote_peer_cache, node->uuid, node);
	crm_trace("added %s to remote cache", node_name);

	/* Update the entry's uname, ensuring peer status callbacks are called */
	update_peer_uname(node, node_name);
	free(node_name_copy);
	return node;
	}

	/*!
	* \internal
	* \brief Remove a node from the Pacemaker Remote node cache
	*
	* \param[in] node_name Name of node to remove from cache
	*
	* \note The caller must be careful not to use \p node_name after calling this
	* function if it might be a pointer into the cache entry being removed.
	*/
	void
	pcmk__cluster_forget_remote_node(const char *node_name)
	{
	/* Do a lookup first, because node_name could be a pointer within the entry
	* being removed -- we can't log it after removing it.
	*/
	if (g_hash_table_lookup(crm_remote_peer_cache, node_name) != NULL) {
	crm_trace("Removing %s from Pacemaker Remote node cache", node_name);
	g_hash_table_remove(crm_remote_peer_cache, node_name);
	}
	}

	/*!
	* \internal
	* \brief Return node status based on a CIB status entry
	*
	* \param[in] node_state XML of node state
	*
	* \return \c CRM_NODE_LOST if \c PCMK__XA_IN_CCM is false in
	* \c PCMK__XE_NODE_STATE, \c CRM_NODE_MEMBER otherwise
	* \note Unlike most boolean XML attributes, this one defaults to true, for
	* backward compatibility with older controllers that don't set it.
	*/
	static const char *
	remote_state_from_cib(const xmlNode *node_state)
	{
	bool status = false;

	if ((pcmk__xe_get_bool_attr(node_state, PCMK__XA_IN_CCM,
	&status) == pcmk_rc_ok) && !status) {
	return CRM_NODE_LOST;
	} else {
	return CRM_NODE_MEMBER;
	}
	}

	/* user data for looping through remote node xpath searches */
	struct refresh_data {
	const char field; / XML attribute to check for node name */
	gboolean has_state; /* whether to update node state based on XML */
	};

	/*!
	* \internal
	* \brief Process one pacemaker_remote node xpath search result
	*
	* \param[in] result XML search result
	* \param[in] user_data what to look for in the XML
	*/
	static void
	remote_cache_refresh_helper(xmlNode result, void user_data)
	{
	const struct refresh_data *data = user_data;
	const char *remote = crm_element_value(result, data->field);
	const char *state = NULL;
	crm_node_t *node;

	CRM_CHECK(remote != NULL, return);

	/* Determine node's state, if the result has it */
	if (data->has_state) {
	state = remote_state_from_cib(result);
	}

	/* Check whether cache already has entry for node */
	node = g_hash_table_lookup(crm_remote_peer_cache, remote);

	if (node == NULL) {
	/* Node is not in cache, so add a new entry for it */
	node = pcmk__cluster_lookup_remote_node(remote);
	CRM_ASSERT(node);
	if (state) {
	pcmk__update_peer_state(__func__, node, state, 0);
	}

	} else if (pcmk_is_set(node->flags, crm_node_dirty)) {
	/* Node is in cache and hasn't been updated already, so mark it clean */
	clear_peer_flags(node, crm_node_dirty);
	if (state) {
	pcmk__update_peer_state(__func__, node, state, 0);
	}
	}
	}

	static void
	mark_dirty(gpointer key, gpointer value, gpointer user_data)
	{
	set_peer_flags((crm_node_t *) value, crm_node_dirty);
	}

	static gboolean
	is_dirty(gpointer key, gpointer value, gpointer user_data)
	{
	return pcmk_is_set(((crm_node_t*)value)->flags, crm_node_dirty);
	}

	/*!
	* \internal
	* \brief Repopulate the remote node cache based on CIB XML
	*
	* \param[in] cib CIB XML to parse
	*/
	static void
	refresh_remote_nodes(xmlNode *cib)
	{
	struct refresh_data data;

	crm_peer_init();

	/* First, we mark all existing cache entries as dirty,
	* so that later we can remove any that weren't in the CIB.
	* We don't empty the cache, because we need to detect changes in state.
	*/
	g_hash_table_foreach(crm_remote_peer_cache, mark_dirty, NULL);

	/* Look for guest nodes and remote nodes in the status section */
	data.field = PCMK_XA_ID;
	data.has_state = TRUE;
	crm_foreach_xpath_result(cib, PCMK__XP_REMOTE_NODE_STATUS,
	remote_cache_refresh_helper, &data);

	/* Look for guest nodes and remote nodes in the configuration section,
	* because they may have just been added and not have a status entry yet.
	* In that case, the cached node state will be left NULL, so that the
	* peer status callback isn't called until we're sure the node started
	* successfully.
	*/
	data.field = PCMK_XA_VALUE;
	data.has_state = FALSE;
	crm_foreach_xpath_result(cib, PCMK__XP_GUEST_NODE_CONFIG,
	remote_cache_refresh_helper, &data);
	data.field = PCMK_XA_ID;
	data.has_state = FALSE;
	crm_foreach_xpath_result(cib, PCMK__XP_REMOTE_NODE_CONFIG,
	remote_cache_refresh_helper, &data);

	/* Remove all old cache entries that weren't seen in the CIB */
	g_hash_table_foreach_remove(crm_remote_peer_cache, is_dirty, NULL);
	}

	/*!
	* \internal
	* \brief Check whether a node is an active cluster node
	*
	* Remote nodes are never considered active. This guarantees that they can never
	* become DC.
	*
	* \param[in] node Node to check
	*
	* \return \c true if the node is an active cluster node, or \c false otherwise
	*/
	bool
	pcmk__cluster_is_node_active(const crm_node_t *node)
	{
	const enum cluster_type_e type = get_cluster_type();

	if ((node == NULL) \|\| pcmk_is_set(node->flags, crm_remote_node)) {
	return false;
	}

	switch (type) {
	case pcmk_cluster_corosync:
	#if SUPPORT_COROSYNC
	return crm_is_corosync_peer_active(node);
	#else
	break;
	#endif // SUPPORT_COROSYNC
	default:
	break;
	}

	crm_err("Unhandled cluster type: %s", name_for_cluster_type(type));
	return false;
	}

	-gboolean
	-crm_is_peer_active(const crm_node_t * node)
	-{
	- return pcmk__cluster_is_node_active(node);
	-}
	-
	static gboolean
	crm_reap_dead_member(gpointer key, gpointer value, gpointer user_data)
	{
	crm_node_t *node = value;
	crm_node_t *search = user_data;

	if (search == NULL) {
	return FALSE;

	} else if (search->id && node->id != search->id) {
	return FALSE;

	} else if (search->id == 0 && !pcmk__str_eq(node->uname, search->uname, pcmk__str_casei)) {
	return FALSE;

	} else if (!pcmk__cluster_is_node_active(value)) {
	crm_info("Removing node with name %s and " PCMK_XA_ID
	" %u from membership cache",
	(node->uname? node->uname : "unknown"), node->id);
	return TRUE;
	}
	return FALSE;
	}

	/*!
	* \brief Remove all peer cache entries matching a node ID and/or uname
	*
	* \param[in] id ID of node to remove (or 0 to ignore)
	* \param[in] name Uname of node to remove (or NULL to ignore)
	*
	* \return Number of cache entries removed
	*
	* \note The caller must be careful not to use \p name after calling this
	* function if it might be a pointer into the cache entry being removed.
	*/
	guint
	reap_crm_member(uint32_t id, const char *name)
	{
	int matches = 0;
	crm_node_t search = { 0, };

	if (crm_peer_cache == NULL) {
	crm_trace("Membership cache not initialized, ignoring purge request");
	return 0;
	}

	search.id = id;
	search.uname = pcmk__str_copy(name);
	matches = g_hash_table_foreach_remove(crm_peer_cache, crm_reap_dead_member, &search);
	if(matches) {
	crm_notice("Purged %d peer%s with " PCMK_XA_ID
	"=%u%s%s from the membership cache",
	matches, pcmk__plural_s(matches), search.id,
	(search.uname? " and/or uname=" : ""),
	(search.uname? search.uname : ""));

	} else {
	crm_info("No peers with " PCMK_XA_ID
	"=%u%s%s to purge from the membership cache",
	search.id, (search.uname? " and/or uname=" : ""),
	(search.uname? search.uname : ""));
	}

	free(search.uname);
	return matches;
	}

	static void
	count_peer(gpointer key, gpointer value, gpointer user_data)
	{
	guint *count = user_data;
	crm_node_t *node = value;

	if (pcmk__cluster_is_node_active(node)) {
	count = count + 1;
	}
	}

	guint
	crm_active_peers(void)
	{
	guint count = 0;

	if (crm_peer_cache) {
	g_hash_table_foreach(crm_peer_cache, count_peer, &count);
	}
	return count;
	}

	static void
	destroy_crm_node(gpointer data)
	{
	crm_node_t *node = data;

	crm_trace("Destroying entry for node %u: %s", node->id, node->uname);

	free(node->uname);
	free(node->state);
	free(node->uuid);
	free(node->expected);
	free(node->conn_host);
	free(node);
	}

	void
	crm_peer_init(void)
	{
	if (crm_peer_cache == NULL) {
	crm_peer_cache = pcmk__strikey_table(free, destroy_crm_node);
	}

	if (crm_remote_peer_cache == NULL) {
	crm_remote_peer_cache = pcmk__strikey_table(NULL, destroy_crm_node);
	}

	if (known_node_cache == NULL) {
	known_node_cache = pcmk__strikey_table(free, destroy_crm_node);
	}
	}

	void
	crm_peer_destroy(void)
	{
	if (crm_peer_cache != NULL) {
	crm_trace("Destroying peer cache with %d members", g_hash_table_size(crm_peer_cache));
	g_hash_table_destroy(crm_peer_cache);
	crm_peer_cache = NULL;
	}

	if (crm_remote_peer_cache != NULL) {
	crm_trace("Destroying remote peer cache with %d members",
	pcmk__cluster_num_remote_nodes());
	g_hash_table_destroy(crm_remote_peer_cache);
	crm_remote_peer_cache = NULL;
	}

	if (known_node_cache != NULL) {
	crm_trace("Destroying known node cache with %d members",
	g_hash_table_size(known_node_cache));
	g_hash_table_destroy(known_node_cache);
	known_node_cache = NULL;
	}

	}

	static void (peer_status_callback)(enum crm_status_type, crm_node_t ,
	const void *) = NULL;

	/*!
	* \brief Set a client function that will be called after peer status changes
	*
	* \param[in] dispatch Pointer to function to use as callback
	*
	* \note Previously, client callbacks were responsible for peer cache
	* management. This is no longer the case, and client callbacks should do
	* only client-specific handling. Callbacks MUST NOT add or remove entries
	* in the peer caches.
	*/
	void
	crm_set_status_callback(void (dispatch) (enum crm_status_type, crm_node_t , const void *))
	{
	peer_status_callback = dispatch;
	}

	/*!
	* \brief Tell the library whether to automatically reap lost nodes
	*
	* If TRUE (the default), calling crm_update_peer_proc() will also update the
	* peer state to CRM_NODE_MEMBER or CRM_NODE_LOST, and pcmk__update_peer_state()
	* will reap peers whose state changes to anything other than CRM_NODE_MEMBER.
	* Callers should leave this enabled unless they plan to manage the cache
	* separately on their own.
	*
	* \param[in] autoreap TRUE to enable automatic reaping, FALSE to disable
	*/
	void
	crm_set_autoreap(gboolean autoreap)
	{
	crm_autoreap = autoreap;
	}

	static void
	dump_peer_hash(int level, const char *caller)
	{
	GHashTableIter iter;
	const char *id = NULL;
	crm_node_t *node = NULL;

	g_hash_table_iter_init(&iter, crm_peer_cache);
	while (g_hash_table_iter_next(&iter, (gpointer ) &id, (gpointer ) &node)) {
	do_crm_log(level, "%s: Node %u/%s = %p - %s", caller, node->id, node->uname, node, id);
	}
	}

	static gboolean
	hash_find_by_data(gpointer key, gpointer value, gpointer user_data)
	{
	return value == user_data;
	}

	/*!
	* \internal
	* \brief Search caches for a node (cluster or Pacemaker Remote)
	*
	* \param[in] id If not 0, cluster node ID to search for
	* \param[in] uname If not NULL, node name to search for
	* \param[in] flags Group of enum pcmk__node_search_flags
	*
	* \return Node cache entry if found, otherwise NULL
	*/
	crm_node_t *
	pcmk__search_node_caches(unsigned int id, const char *uname, uint32_t flags)
	{
	crm_node_t *node = NULL;

	CRM_ASSERT(id > 0 \|\| uname != NULL);

	crm_peer_init();

	if ((uname != NULL) && pcmk_is_set(flags, pcmk__node_search_remote)) {
	node = g_hash_table_lookup(crm_remote_peer_cache, uname);
	}

	if ((node == NULL) && pcmk_is_set(flags, pcmk__node_search_cluster)) {
	node = pcmk__search_cluster_node_cache(id, uname, NULL);
	}

	if ((node == NULL) && pcmk_is_set(flags, pcmk__node_search_known)) {
	char *id_str = (id == 0)? NULL : crm_strdup_printf("%u", id);

	node = find_known_node(id_str, uname);
	free(id_str);
	}

	return node;
	}

	/*!
	* \internal
	* \brief Purge a node from cache (both cluster and Pacemaker Remote)
	*
	* \param[in] node_name If not NULL, purge only nodes with this name
	* \param[in] node_id If not 0, purge cluster nodes only if they have this ID
	*
	* \note If \p node_name is NULL and \p node_id is 0, no nodes will be purged.
	* If \p node_name is not NULL and \p node_id is not 0, Pacemaker Remote
	* nodes that match \p node_name will be purged, and cluster nodes that
	* match both \p node_name and \p node_id will be purged.
	* \note The caller must be careful not to use \p node_name after calling this
	* function if it might be a pointer into a cache entry being removed.
	*/
	void
	pcmk__purge_node_from_cache(const char *node_name, uint32_t node_id)
	{
	char *node_name_copy = NULL;

	if ((node_name == NULL) && (node_id == 0U)) {
	return;
	}

	// Purge from Pacemaker Remote node cache
	if ((node_name != NULL)
	&& (g_hash_table_lookup(crm_remote_peer_cache, node_name) != NULL)) {
	/* node_name could be a pointer into the cache entry being purged,
	* so reassign it to a copy before the original gets freed
	*/
	node_name_copy = pcmk__str_copy(node_name);
	node_name = node_name_copy;

	crm_trace("Purging %s from Pacemaker Remote node cache", node_name);
	g_hash_table_remove(crm_remote_peer_cache, node_name);
	}

	reap_crm_member(node_id, node_name);
	free(node_name_copy);
	}

	/*!
	* \internal
	* \brief Search cluster node cache
	*
	* \param[in] id If not 0, cluster node ID to search for
	* \param[in] uname If not NULL, node name to search for
	* \param[in] uuid If not NULL while id is 0, node UUID instead of cluster
	* node ID to search for
	*
	* \return Cluster node cache entry if found, otherwise NULL
	*/
	crm_node_t *
	pcmk__search_cluster_node_cache(unsigned int id, const char *uname,
	const char *uuid)
	{
	GHashTableIter iter;
	crm_node_t *node = NULL;
	crm_node_t *by_id = NULL;
	crm_node_t *by_name = NULL;

	CRM_ASSERT(id > 0 \|\| uname != NULL);

	crm_peer_init();

	if (uname != NULL) {
	g_hash_table_iter_init(&iter, crm_peer_cache);
	while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) {
	if(node->uname && strcasecmp(node->uname, uname) == 0) {
	crm_trace("Name match: %s = %p", node->uname, node);
	by_name = node;
	break;
	}
	}
	}

	if (id > 0) {
	g_hash_table_iter_init(&iter, crm_peer_cache);
	while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) {
	if(node->id == id) {
	crm_trace("ID match: %u = %p", node->id, node);
	by_id = node;
	break;
	}
	}

	} else if (uuid != NULL) {
	g_hash_table_iter_init(&iter, crm_peer_cache);
	while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) {
	if (pcmk__str_eq(node->uuid, uuid, pcmk__str_casei)) {
	crm_trace("UUID match: %s = %p", node->uuid, node);
	by_id = node;
	break;
	}
	}
	}

	node = by_id; /* Good default */
	if(by_id == by_name) {
	/* Nothing to do if they match (both NULL counts) */
	crm_trace("Consistent: %p for %u/%s", by_id, id, uname);

	} else if(by_id == NULL && by_name) {
	crm_trace("Only one: %p for %u/%s", by_name, id, uname);

	if(id && by_name->id) {
	dump_peer_hash(LOG_WARNING, __func__);
	crm_crit("Node %u and %u share the same name '%s'",
	id, by_name->id, uname);
	node = NULL; /* Create a new one */

	} else {
	node = by_name;
	}

	} else if(by_name == NULL && by_id) {
	crm_trace("Only one: %p for %u/%s", by_id, id, uname);

	if(uname && by_id->uname) {
	dump_peer_hash(LOG_WARNING, __func__);
	crm_crit("Node '%s' and '%s' share the same cluster nodeid %u: assuming '%s' is correct",
	uname, by_id->uname, id, uname);
	}

	} else if(uname && by_id->uname) {
	if(pcmk__str_eq(uname, by_id->uname, pcmk__str_casei)) {
	crm_notice("Node '%s' has changed its ID from %u to %u", by_id->uname, by_name->id, by_id->id);
	g_hash_table_foreach_remove(crm_peer_cache, hash_find_by_data, by_name);

	} else {
	crm_warn("Node '%s' and '%s' share the same cluster nodeid: %u %s", by_id->uname, by_name->uname, id, uname);
	dump_peer_hash(LOG_INFO, __func__);
	crm_abort(__FILE__, __func__, __LINE__, "member weirdness", TRUE,
	TRUE);
	}

	} else if(id && by_name->id) {
	crm_warn("Node %u and %u share the same name: '%s'", by_id->id, by_name->id, uname);

	} else {
	/* Simple merge */

	/* Only corosync-based clusters use node IDs. The functions that call
	* pcmk__update_peer_state() and crm_update_peer_proc() only know
	* nodeid, so 'by_id' is authoritative when merging.
	*/
	dump_peer_hash(LOG_DEBUG, __func__);

	crm_info("Merging %p into %p", by_name, by_id);
	g_hash_table_foreach_remove(crm_peer_cache, hash_find_by_data, by_name);
	}

	return node;
	}

	#if SUPPORT_COROSYNC
	static guint
	remove_conflicting_peer(crm_node_t *node)
	{
	int matches = 0;
	GHashTableIter iter;
	crm_node_t *existing_node = NULL;

	if (node->id == 0 \|\| node->uname == NULL) {
	return 0;
	}

	if (!pcmk__corosync_has_nodelist()) {
	return 0;
	}

	g_hash_table_iter_init(&iter, crm_peer_cache);
	while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &existing_node)) {
	if (existing_node->id > 0
	&& existing_node->id != node->id
	&& existing_node->uname != NULL
	&& strcasecmp(existing_node->uname, node->uname) == 0) {

	if (pcmk__cluster_is_node_active(existing_node)) {
	continue;
	}

	crm_warn("Removing cached offline node %u/%s which has conflicting uname with %u",
	existing_node->id, existing_node->uname, node->id);

	g_hash_table_iter_remove(&iter);
	matches++;
	}
	}

	return matches;
	}
	#endif

	/*!
	* \brief Get a cluster node cache entry
	*
	* \param[in] id If not 0, cluster node ID to search for
	* \param[in] uname If not NULL, node name to search for
	* \param[in] uuid If not NULL while id is 0, node UUID instead of cluster
	* node ID to search for
	* \param[in] flags Group of enum pcmk__node_search_flags
	*
	* \return (Possibly newly created) cluster node cache entry
	*/
	/* coverity[-alloc] Memory is referenced in one or both hashtables */
	crm_node_t *
	pcmk__get_node(unsigned int id, const char uname, const char uuid,
	uint32_t flags)
	{
	crm_node_t *node = NULL;
	char *uname_lookup = NULL;

	CRM_ASSERT(id > 0 \|\| uname != NULL);

	crm_peer_init();

	// Check the Pacemaker Remote node cache first
	if (pcmk_is_set(flags, pcmk__node_search_remote)) {
	node = g_hash_table_lookup(crm_remote_peer_cache, uname);
	if (node != NULL) {
	return node;
	}
	}

	if (!pcmk_is_set(flags, pcmk__node_search_cluster)) {
	return NULL;
	}

	node = pcmk__search_cluster_node_cache(id, uname, uuid);

	/* if uname wasn't provided, and find_peer did not turn up a uname based on id.
	* we need to do a lookup of the node name using the id in the cluster membership. */
	if ((node == NULL \|\| node->uname == NULL) && (uname == NULL)) {
	uname_lookup = get_node_name(id);
	}

	if (uname_lookup) {
	uname = uname_lookup;
	crm_trace("Inferred a name of '%s' for node %u", uname, id);

	/* try to turn up the node one more time now that we know the uname. */
	if (node == NULL) {
	node = pcmk__search_cluster_node_cache(id, uname, uuid);
	}
	}

	if (node == NULL) {
	char *uniqueid = crm_generate_uuid();

	node = pcmk__assert_alloc(1, sizeof(crm_node_t));

	crm_info("Created entry %s/%p for node %s/%u (%d total)",
	uniqueid, node, uname, id, 1 + g_hash_table_size(crm_peer_cache));
	g_hash_table_replace(crm_peer_cache, uniqueid, node);
	}

	if(id > 0 && uname && (node->id == 0 \|\| node->uname == NULL)) {
	crm_info("Node %u is now known as %s", id, uname);
	}

	if(id > 0 && node->id == 0) {
	node->id = id;
	}

	if (uname && (node->uname == NULL)) {
	update_peer_uname(node, uname);
	}

	if(node->uuid == NULL) {
	if (uuid == NULL) {
	uuid = crm_peer_uuid(node);
	}

	if (uuid) {
	crm_info("Node %u has uuid %s", id, uuid);

	} else {
	crm_info("Cannot obtain a UUID for node %u/%s", id, node->uname);
	}
	}

	free(uname_lookup);

	return node;
	}

	/*!
	* \internal
	* \brief Update a node's uname
	*
	* \param[in,out] node Node object to update
	* \param[in] uname New name to set
	*
	* \note This function should not be called within a peer cache iteration,
	* because in some cases it can remove conflicting cache entries,
	* which would invalidate the iterator.
	*/
	static void
	update_peer_uname(crm_node_t node, const char uname)
	{
	CRM_CHECK(uname != NULL,
	crm_err("Bug: can't update node name without name"); return);
	CRM_CHECK(node != NULL,
	crm_err("Bug: can't update node name to %s without node", uname);
	return);

	if (pcmk__str_eq(uname, node->uname, pcmk__str_casei)) {
	crm_debug("Node uname '%s' did not change", uname);
	return;
	}

	for (const char c = uname; c; ++c) {
	if ((c >= 'A') && (c <= 'Z')) {
	crm_warn("Node names with capitals are discouraged, consider changing '%s'",
	uname);
	break;
	}
	}

	pcmk__str_update(&node->uname, uname);

	if (peer_status_callback != NULL) {
	peer_status_callback(crm_status_uname, node, NULL);
	}

	#if SUPPORT_COROSYNC
	if (is_corosync_cluster() && !pcmk_is_set(node->flags, crm_remote_node)) {
	remove_conflicting_peer(node);
	}
	#endif
	}

	/*!
	* \internal
	* \brief Get log-friendly string equivalent of a process flag
	*
	* \param[in] proc Process flag
	*
	* \return Log-friendly string equivalent of \p proc
	*/
	static inline const char *
	proc2text(enum crm_proc_flag proc)
	{
	const char *text = "unknown";

	switch (proc) {
	case crm_proc_none:
	text = "none";
	break;
	case crm_proc_based:
	text = "pacemaker-based";
	break;
	case crm_proc_controld:
	text = "pacemaker-controld";
	break;
	case crm_proc_schedulerd:
	text = "pacemaker-schedulerd";
	break;
	case crm_proc_execd:
	text = "pacemaker-execd";
	break;
	case crm_proc_attrd:
	text = "pacemaker-attrd";
	break;
	case crm_proc_fenced:
	text = "pacemaker-fenced";
	break;
	case crm_proc_cpg:
	text = "corosync-cpg";
	break;
	}
	return text;
	}

	/*!
	* \internal
	* \brief Update a node's process information (and potentially state)
	*
	* \param[in] source Caller's function name (for log messages)
	* \param[in,out] node Node object to update
	* \param[in] flag Bitmask of new process information
	* \param[in] status node status (online, offline, etc.)
	*
	* \return NULL if any node was reaped from peer caches, value of node otherwise
	*
	* \note If this function returns NULL, the supplied node object was likely
	* freed and should not be used again. This function should not be
	* called within a cache iteration if reaping is possible, otherwise
	* reaping could invalidate the iterator.
	*/
	crm_node_t *
	crm_update_peer_proc(const char source, crm_node_t node, uint32_t flag, const char *status)
	{
	uint32_t last = 0;
	gboolean changed = FALSE;

	CRM_CHECK(node != NULL, crm_err("%s: Could not set %s to %s for NULL",
	source, proc2text(flag), status);
	return NULL);

	/* Pacemaker doesn't spawn processes on remote nodes */
	if (pcmk_is_set(node->flags, crm_remote_node)) {
	return node;
	}

	last = node->processes;
	if (status == NULL) {
	node->processes = flag;
	if (node->processes != last) {
	changed = TRUE;
	}

	} else if (pcmk__str_eq(status, PCMK_VALUE_ONLINE, pcmk__str_casei)) {
	if ((node->processes & flag) != flag) {
	node->processes = pcmk__set_flags_as(__func__, __LINE__,
	LOG_TRACE, "Peer process",
	node->uname, node->processes,
	flag, "processes");
	changed = TRUE;
	}

	} else if (node->processes & flag) {
	node->processes = pcmk__clear_flags_as(__func__, __LINE__,
	LOG_TRACE, "Peer process",
	node->uname, node->processes,
	flag, "processes");
	changed = TRUE;
	}

	if (changed) {
	if (status == NULL && flag <= crm_proc_none) {
	crm_info("%s: Node %s[%u] - all processes are now offline", source, node->uname,
	node->id);
	} else {
	crm_info("%s: Node %s[%u] - %s is now %s", source, node->uname, node->id,
	proc2text(flag), status);
	}

	if (pcmk_is_set(node->processes, crm_get_cluster_proc())) {
	node->when_online = time(NULL);

	} else {
	node->when_online = 0;
	}

	/* Call the client callback first, then update the peer state,
	* in case the node will be reaped
	*/
	if (peer_status_callback != NULL) {
	peer_status_callback(crm_status_processes, node, &last);
	}

	/* The client callback shouldn't touch the peer caches,
	* but as a safety net, bail if the peer cache was destroyed.
	*/
	if (crm_peer_cache == NULL) {
	return NULL;
	}

	if (crm_autoreap) {
	const char *peer_state = NULL;

	if (pcmk_is_set(node->processes, crm_get_cluster_proc())) {
	peer_state = CRM_NODE_MEMBER;
	} else {
	peer_state = CRM_NODE_LOST;
	}
	node = pcmk__update_peer_state(__func__, node, peer_state, 0);
	}
	} else {
	crm_trace("%s: Node %s[%u] - %s is unchanged (%s)", source, node->uname, node->id,
	proc2text(flag), status);
	}
	return node;
	}

	/*!
	* \internal
	* \brief Update a cluster node cache entry's expected join state
	*
	* \param[in] source Caller's function name (for logging)
	* \param[in,out] node Node to update
	* \param[in] expected Node's new join state
	*/
	void
	pcmk__update_peer_expected(const char source, crm_node_t node,
	const char *expected)
	{
	char *last = NULL;
	gboolean changed = FALSE;

	CRM_CHECK(node != NULL, crm_err("%s: Could not set 'expected' to %s", source, expected);
	return);

	/* Remote nodes don't participate in joins */
	if (pcmk_is_set(node->flags, crm_remote_node)) {
	return;
	}

	last = node->expected;
	if (expected != NULL && !pcmk__str_eq(node->expected, expected, pcmk__str_casei)) {
	node->expected = strdup(expected);
	changed = TRUE;
	}

	if (changed) {
	crm_info("%s: Node %s[%u] - expected state is now %s (was %s)", source, node->uname, node->id,
	expected, last);
	free(last);
	} else {
	crm_trace("%s: Node %s[%u] - expected state is unchanged (%s)", source, node->uname,
	node->id, expected);
	}
	}

	/*!
	* \internal
	* \brief Update a node's state and membership information
	*
	* \param[in] source Caller's function name (for log messages)
	* \param[in,out] node Node object to update
	* \param[in] state Node's new state
	* \param[in] membership Node's new membership ID
	* \param[in,out] iter If not NULL, pointer to node's peer cache iterator
	*
	* \return NULL if any node was reaped, value of node otherwise
	*
	* \note If this function returns NULL, the supplied node object was likely
	* freed and should not be used again. This function may be called from
	* within a peer cache iteration if the iterator is supplied.
	*/
	static crm_node_t *
	update_peer_state_iter(const char source, crm_node_t node, const char *state,
	uint64_t membership, GHashTableIter *iter)
	{
	gboolean is_member;

	CRM_CHECK(node != NULL,
	crm_err("Could not set state for unknown host to %s"
	CRM_XS " source=%s", state, source);
	return NULL);

	is_member = pcmk__str_eq(state, CRM_NODE_MEMBER, pcmk__str_casei);
	if (is_member) {
	node->when_lost = 0;
	if (membership) {
	node->last_seen = membership;
	}
	}

	if (state && !pcmk__str_eq(node->state, state, pcmk__str_casei)) {
	char *last = node->state;

	if (is_member) {
	node->when_member = time(NULL);

	} else {
	node->when_member = 0;
	}

	node->state = strdup(state);
	crm_notice("Node %s state is now %s " CRM_XS
	" nodeid=%u previous=%s source=%s", node->uname, state,
	node->id, (last? last : "unknown"), source);
	if (peer_status_callback != NULL) {
	peer_status_callback(crm_status_nstate, node, last);
	}
	free(last);

	if (crm_autoreap && !is_member
	&& !pcmk_is_set(node->flags, crm_remote_node)) {
	/* We only autoreap from the peer cache, not the remote peer cache,
	* because the latter should be managed only by
	* refresh_remote_nodes().
	*/
	if(iter) {
	crm_notice("Purged 1 peer with " PCMK_XA_ID
	"=%u and/or uname=%s from the membership cache",
	node->id, node->uname);
	g_hash_table_iter_remove(iter);

	} else {
	reap_crm_member(node->id, node->uname);
	}
	node = NULL;
	}

	} else {
	crm_trace("Node %s state is unchanged (%s) " CRM_XS
	" nodeid=%u source=%s", node->uname, state, node->id, source);
	}
	return node;
	}

	/*!
	* \brief Update a node's state and membership information
	*
	* \param[in] source Caller's function name (for log messages)
	* \param[in,out] node Node object to update
	* \param[in] state Node's new state
	* \param[in] membership Node's new membership ID
	*
	* \return NULL if any node was reaped, value of node otherwise
	*
	* \note If this function returns NULL, the supplied node object was likely
	* freed and should not be used again. This function should not be
	* called within a cache iteration if reaping is possible,
	* otherwise reaping could invalidate the iterator.
	*/
	crm_node_t *
	pcmk__update_peer_state(const char source, crm_node_t node,
	const char *state, uint64_t membership)
	{
	return update_peer_state_iter(source, node, state, membership, NULL);
	}

	/*!
	* \internal
	* \brief Reap all nodes from cache whose membership information does not match
	*
	* \param[in] membership Membership ID of nodes to keep
	*/
	void
	pcmk__reap_unseen_nodes(uint64_t membership)
	{
	GHashTableIter iter;
	crm_node_t *node = NULL;

	crm_trace("Reaping unseen nodes...");
	g_hash_table_iter_init(&iter, crm_peer_cache);
	while (g_hash_table_iter_next(&iter, NULL, (gpointer *)&node)) {
	if (node->last_seen != membership) {
	if (node->state) {
	/*
	* Calling update_peer_state_iter() allows us to
	* remove the node from crm_peer_cache without
	* invalidating our iterator
	*/
	update_peer_state_iter(__func__, node, CRM_NODE_LOST,
	membership, &iter);

	} else {
	crm_info("State of node %s[%u] is still unknown",
	node->uname, node->id);
	}
	}
	}
	}

	static crm_node_t *
	find_known_node(const char id, const char uname)
	{
	GHashTableIter iter;
	crm_node_t *node = NULL;
	crm_node_t *by_id = NULL;
	crm_node_t *by_name = NULL;

	if (uname) {
	g_hash_table_iter_init(&iter, known_node_cache);
	while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) {
	if (node->uname && strcasecmp(node->uname, uname) == 0) {
	crm_trace("Name match: %s = %p", node->uname, node);
	by_name = node;
	break;
	}
	}
	}

	if (id) {
	g_hash_table_iter_init(&iter, known_node_cache);
	while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) {
	if(strcasecmp(node->uuid, id) == 0) {
	crm_trace("ID match: %s= %p", id, node);
	by_id = node;
	break;
	}
	}
	}

	node = by_id; /* Good default */
	if (by_id == by_name) {
	/* Nothing to do if they match (both NULL counts) */
	crm_trace("Consistent: %p for %s/%s", by_id, id, uname);

	} else if (by_id == NULL && by_name) {
	crm_trace("Only one: %p for %s/%s", by_name, id, uname);

	if (id) {
	node = NULL;

	} else {
	node = by_name;
	}

	} else if (by_name == NULL && by_id) {
	crm_trace("Only one: %p for %s/%s", by_id, id, uname);

	if (uname) {
	node = NULL;
	}

	} else if (uname && by_id->uname
	&& pcmk__str_eq(uname, by_id->uname, pcmk__str_casei)) {
	/* Multiple nodes have the same uname in the CIB.
	* Return by_id. */

	} else if (id && by_name->uuid
	&& pcmk__str_eq(id, by_name->uuid, pcmk__str_casei)) {
	/* Multiple nodes have the same id in the CIB.
	* Return by_name. */
	node = by_name;

	} else {
	node = NULL;
	}

	if (node == NULL) {
	crm_debug("Couldn't find node%s%s%s%s",
	id? " " : "",
	id? id : "",
	uname? " with name " : "",
	uname? uname : "");
	}

	return node;
	}

	static void
	known_node_cache_refresh_helper(xmlNode xml_node, void user_data)
	{
	const char *id = crm_element_value(xml_node, PCMK_XA_ID);
	const char *uname = crm_element_value(xml_node, PCMK_XA_UNAME);
	crm_node_t * node = NULL;

	CRM_CHECK(id != NULL && uname !=NULL, return);
	node = find_known_node(id, uname);

	if (node == NULL) {
	char *uniqueid = crm_generate_uuid();

	node = pcmk__assert_alloc(1, sizeof(crm_node_t));

	node->uname = pcmk__str_copy(uname);
	node->uuid = pcmk__str_copy(id);

	g_hash_table_replace(known_node_cache, uniqueid, node);

	} else if (pcmk_is_set(node->flags, crm_node_dirty)) {
	pcmk__str_update(&node->uname, uname);

	/* Node is in cache and hasn't been updated already, so mark it clean */
	clear_peer_flags(node, crm_node_dirty);
	}

	}

	static void
	refresh_known_node_cache(xmlNode *cib)
	{
	crm_peer_init();

	g_hash_table_foreach(known_node_cache, mark_dirty, NULL);

	crm_foreach_xpath_result(cib, PCMK__XP_MEMBER_NODE_CONFIG,
	known_node_cache_refresh_helper, NULL);

	/* Remove all old cache entries that weren't seen in the CIB */
	g_hash_table_foreach_remove(known_node_cache, is_dirty, NULL);
	}

	void
	pcmk__refresh_node_caches_from_cib(xmlNode *cib)
	{
	refresh_remote_nodes(cib);
	refresh_known_node_cache(cib);
	}

	// Deprecated functions kept only for backward API compatibility
	// LCOV_EXCL_START

	#include <crm/cluster/compat.h>

	int
	crm_terminate_member(int nodeid, const char uname, void unused)
	{
	return stonith_api_kick(nodeid, uname, 120, TRUE);
	}

	int
	crm_terminate_member_no_mainloop(int nodeid, const char uname, int connection)
	{
	return stonith_api_kick(nodeid, uname, 120, TRUE);
	}

	crm_node_t *
	crm_get_peer(unsigned int id, const char *uname)
	{
	return pcmk__get_node(id, uname, NULL, pcmk__node_search_cluster);
	}

	crm_node_t *
	crm_get_peer_full(unsigned int id, const char *uname, int flags)
	{
	return pcmk__get_node(id, uname, NULL, flags);
	}

	int
	crm_remote_peer_cache_size(void)
	{
	unsigned int count = pcmk__cluster_num_remote_nodes();

	return QB_MIN(count, INT_MAX);
	}

	void
	crm_remote_peer_cache_refresh(xmlNode *cib)
	{
	refresh_remote_nodes(cib);
	}

	crm_node_t *
	crm_remote_peer_get(const char *node_name)
	{
	return pcmk__cluster_lookup_remote_node(node_name);
	}

	void
	crm_remote_peer_cache_remove(const char *node_name)
	{
	pcmk__cluster_forget_remote_node(node_name);
	}

	+gboolean
	+crm_is_peer_active(const crm_node_t * node)
	+{
	+ return pcmk__cluster_is_node_active(node);
	+}
	+
	// LCOV_EXCL_STOP
	// End deprecated API

File Metadata

Mime Type: text/x-diff
Expires: Sat, Nov 23, 6:08 AM (11 h, 1 m)
Storage Engine: blob
Storage Format: Raw Data
Storage Handle: 1011345
Default Alt Text: (55 KB)

No OneTemporaryActions

View Options

File Metadata

Event Timeline

No OneTemporary
Actions