diff --git a/include/crm/pengine/internal.h b/include/crm/pengine/internal.h
index c464295aac..f1ce3aed0d 100644
--- a/include/crm/pengine/internal.h
+++ b/include/crm/pengine/internal.h
@@ -1,598 +1,598 @@
/*
* 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.
*/
#ifndef PE_INTERNAL__H
# define PE_INTERNAL__H
# include <string.h>
# include <crm/pengine/status.h>
# include <crm/pengine/remote_internal.h>
# include <crm/common/internal.h>
# include <crm/common/output_internal.h>
# define pe_rsc_info(rsc, fmt, args...) crm_log_tag(LOG_INFO, rsc ? rsc->id : "<NULL>", fmt, ##args)
# define pe_rsc_debug(rsc, fmt, args...) crm_log_tag(LOG_DEBUG, rsc ? rsc->id : "<NULL>", fmt, ##args)
# define pe_rsc_trace(rsc, fmt, args...) crm_log_tag(LOG_TRACE, rsc ? rsc->id : "<NULL>", fmt, ##args)
# define pe_err(fmt...) do { \
was_processing_error = TRUE; \
pcmk__config_err(fmt); \
} while (0)
# define pe_warn(fmt...) do { \
was_processing_warning = TRUE; \
pcmk__config_warn(fmt); \
} while (0)
# define pe_proc_err(fmt...) { was_processing_error = TRUE; crm_err(fmt); }
# define pe_proc_warn(fmt...) { was_processing_warning = TRUE; crm_warn(fmt); }
#define pe__set_working_set_flags(working_set, flags_to_set) do { \
(working_set)->flags = pcmk__set_flags_as(__func__, __LINE__, \
LOG_TRACE, "Working set", crm_system_name, \
(working_set)->flags, (flags_to_set), #flags_to_set); \
} while (0)
#define pe__clear_working_set_flags(working_set, flags_to_clear) do { \
(working_set)->flags = pcmk__clear_flags_as(__func__, __LINE__, \
LOG_TRACE, "Working set", crm_system_name, \
(working_set)->flags, (flags_to_clear), #flags_to_clear); \
} while (0)
#define pe__set_resource_flags(resource, flags_to_set) do { \
(resource)->flags = pcmk__set_flags_as(__func__, __LINE__, \
LOG_TRACE, "Resource", (resource)->id, (resource)->flags, \
(flags_to_set), #flags_to_set); \
} while (0)
#define pe__clear_resource_flags(resource, flags_to_clear) do { \
(resource)->flags = pcmk__clear_flags_as(__func__, __LINE__, \
LOG_TRACE, "Resource", (resource)->id, (resource)->flags, \
(flags_to_clear), #flags_to_clear); \
} while (0)
#define pe__set_action_flags(action, flags_to_set) do { \
(action)->flags = pcmk__set_flags_as(__func__, __LINE__, \
LOG_TRACE, \
"Action", (action)->uuid, \
(action)->flags, \
(flags_to_set), \
#flags_to_set); \
} while (0)
#define pe__clear_action_flags(action, flags_to_clear) do { \
(action)->flags = pcmk__clear_flags_as(__func__, __LINE__, \
LOG_TRACE, \
"Action", (action)->uuid, \
(action)->flags, \
(flags_to_clear), \
#flags_to_clear); \
} while (0)
#define pe__set_raw_action_flags(action_flags, action_name, flags_to_set) do { \
action_flags = pcmk__set_flags_as(__func__, __LINE__, \
LOG_TRACE, "Action", action_name, \
(action_flags), \
(flags_to_set), #flags_to_set); \
} while (0)
#define pe__clear_raw_action_flags(action_flags, action_name, flags_to_clear) do { \
action_flags = pcmk__clear_flags_as(__func__, __LINE__, \
LOG_TRACE, \
"Action", action_name, \
(action_flags), \
(flags_to_clear), \
#flags_to_clear); \
} while (0)
#define pe__set_action_flags_as(function, line, action, flags_to_set) do { \
(action)->flags = pcmk__set_flags_as((function), (line), \
LOG_TRACE, \
"Action", (action)->uuid, \
(action)->flags, \
(flags_to_set), \
#flags_to_set); \
} while (0)
#define pe__clear_action_flags_as(function, line, action, flags_to_clear) do { \
(action)->flags = pcmk__clear_flags_as((function), (line), \
LOG_TRACE, \
"Action", (action)->uuid, \
(action)->flags, \
(flags_to_clear), \
#flags_to_clear); \
} while (0)
#define pe__set_order_flags(order_flags, flags_to_set) do { \
order_flags = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, \
"Ordering", "constraint", \
order_flags, (flags_to_set), \
#flags_to_set); \
} while (0)
#define pe__clear_order_flags(order_flags, flags_to_clear) do { \
order_flags = pcmk__clear_flags_as(__func__, __LINE__, LOG_TRACE, \
"Ordering", "constraint", \
order_flags, (flags_to_clear), \
#flags_to_clear); \
} while (0)
#define pe__set_graph_flags(graph_flags, gr_action, flags_to_set) do { \
graph_flags = pcmk__set_flags_as(__func__, __LINE__, \
LOG_TRACE, "Graph", \
(gr_action)->uuid, graph_flags, \
(flags_to_set), #flags_to_set); \
} while (0)
#define pe__clear_graph_flags(graph_flags, gr_action, flags_to_clear) do { \
graph_flags = pcmk__clear_flags_as(__func__, __LINE__, \
LOG_TRACE, "Graph", \
(gr_action)->uuid, graph_flags, \
(flags_to_clear), #flags_to_clear); \
} while (0)
// Some warnings we don't want to print every transition
enum pe_warn_once_e {
pe_wo_blind = (1 << 0),
pe_wo_restart_type = (1 << 1),
pe_wo_role_after = (1 << 2),
pe_wo_poweroff = (1 << 3),
pe_wo_require_all = (1 << 4),
pe_wo_order_score = (1 << 5),
pe_wo_neg_threshold = (1 << 6),
pe_wo_remove_after = (1 << 7),
};
extern uint32_t pe_wo;
#define pe_warn_once(pe_wo_bit, fmt...) do { \
if (!pcmk_is_set(pe_wo, pe_wo_bit)) { \
if (pe_wo_bit == pe_wo_blind) { \
crm_warn(fmt); \
} else { \
pe_warn(fmt); \
} \
pe_wo = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, \
"Warn-once", "logging", pe_wo, \
(pe_wo_bit), #pe_wo_bit); \
} \
} while (0);
typedef struct pe__location_constraint_s {
char *id; // Constraint XML ID
pe_resource_t *rsc_lh; // Resource being located
enum rsc_role_e role_filter; // Role to locate
enum pe_discover_e discover_mode; // Resource discovery
GList *node_list_rh; // List of pe_node_t*
} pe__location_t;
typedef struct pe__order_constraint_s {
int id;
enum pe_ordering type;
void *lh_opaque;
pe_resource_t *lh_rsc;
pe_action_t *lh_action;
char *lh_action_task;
void *rh_opaque;
pe_resource_t *rh_rsc;
pe_action_t *rh_action;
char *rh_action_task;
} pe__ordering_t;
typedef struct notify_data_s {
GSList *keys; // Environment variable name/value pairs
const char *action;
pe_action_t *pre;
pe_action_t *post;
pe_action_t *pre_done;
pe_action_t *post_done;
GList *active; /* notify_entry_t* */
GList *inactive; /* notify_entry_t* */
GList *start; /* notify_entry_t* */
GList *stop; /* notify_entry_t* */
GList *demote; /* notify_entry_t* */
GList *promote; /* notify_entry_t* */
GList *promoted; /* notify_entry_t* */
GList *unpromoted; /* notify_entry_t* */
GHashTable *allowed_nodes;
} notify_data_t;
bool pe_can_fence(pe_working_set_t *data_set, pe_node_t *node);
int pe__add_scores(int score1, int score2);
void add_hash_param(GHashTable * hash, const char *name, const char *value);
char *native_parameter(pe_resource_t * rsc, pe_node_t * node, gboolean create, const char *name,
pe_working_set_t * data_set);
pe_node_t *native_location(const pe_resource_t *rsc, GList **list, int current);
void pe_metadata(void);
void verify_pe_options(GHashTable * options);
void common_update_score(pe_resource_t * rsc, const char *id, int score);
-void native_add_running(pe_resource_t * rsc, pe_node_t * node, pe_working_set_t * data_set);
+void native_add_running(pe_resource_t * rsc, pe_node_t * node, pe_working_set_t * data_set, gboolean failed);
gboolean native_unpack(pe_resource_t * rsc, pe_working_set_t * data_set);
gboolean group_unpack(pe_resource_t * rsc, pe_working_set_t * data_set);
gboolean clone_unpack(pe_resource_t * rsc, pe_working_set_t * data_set);
gboolean pe__unpack_bundle(pe_resource_t *rsc, pe_working_set_t *data_set);
pe_resource_t *native_find_rsc(pe_resource_t *rsc, const char *id, const pe_node_t *node,
int flags);
gboolean native_active(pe_resource_t * rsc, gboolean all);
gboolean group_active(pe_resource_t * rsc, gboolean all);
gboolean clone_active(pe_resource_t * rsc, gboolean all);
gboolean pe__bundle_active(pe_resource_t *rsc, gboolean all);
void native_print(pe_resource_t * rsc, const char *pre_text, long options, void *print_data);
void group_print(pe_resource_t * rsc, const char *pre_text, long options, void *print_data);
void clone_print(pe_resource_t * rsc, const char *pre_text, long options, void *print_data);
void pe__print_bundle(pe_resource_t *rsc, const char *pre_text, long options,
void *print_data);
gchar * pcmk__native_output_string(pe_resource_t *rsc, const char *name, pe_node_t *node,
long options, const char *target_role, bool show_nodes);
int pe__name_and_nvpairs_xml(pcmk__output_t *out, bool is_list, const char *tag_name
, size_t pairs_count, ...);
char *pe__node_display_name(pe_node_t *node, bool print_detail);
static inline const char *
pe__rsc_bool_str(pe_resource_t *rsc, uint64_t rsc_flag)
{
return pcmk__btoa(pcmk_is_set(rsc->flags, rsc_flag));
}
int pe__clone_xml(pcmk__output_t *out, va_list args);
int pe__clone_html(pcmk__output_t *out, va_list args);
int pe__clone_text(pcmk__output_t *out, va_list args);
int pe__group_xml(pcmk__output_t *out, va_list args);
int pe__group_html(pcmk__output_t *out, va_list args);
int pe__group_text(pcmk__output_t *out, va_list args);
int pe__bundle_xml(pcmk__output_t *out, va_list args);
int pe__bundle_html(pcmk__output_t *out, va_list args);
int pe__bundle_text(pcmk__output_t *out, va_list args);
int pe__node_html(pcmk__output_t *out, va_list args);
int pe__node_text(pcmk__output_t *out, va_list args);
int pe__node_xml(pcmk__output_t *out, va_list args);
int pe__resource_xml(pcmk__output_t *out, va_list args);
int pe__resource_html(pcmk__output_t *out, va_list args);
int pe__resource_text(pcmk__output_t *out, va_list args);
/* Exported for crm_mon to reference */
int pe__ban_text(pcmk__output_t *out, va_list args);
int pe__cluster_counts_text(pcmk__output_t *out, va_list args);
int pe__cluster_dc_text(pcmk__output_t *out, va_list args);
int pe__cluster_maint_mode_text(pcmk__output_t *out, va_list args);
int pe__cluster_options_text(pcmk__output_t *out, va_list args);
int pe__cluster_stack_text(pcmk__output_t *out, va_list args);
int pe__cluster_summary(pcmk__output_t *out, va_list args);
int pe__cluster_times_text(pcmk__output_t *out, va_list args);
int pe__failed_action_text(pcmk__output_t *out, va_list args);
int pe__node_attribute_text(pcmk__output_t *out, va_list args);
int pe__node_list_text(pcmk__output_t *out, va_list args);
int pe__op_history_text(pcmk__output_t *out, va_list args);
int pe__resource_history_text(pcmk__output_t *out, va_list args);
int pe__ticket_text(pcmk__output_t *out, va_list args);
void native_free(pe_resource_t * rsc);
void group_free(pe_resource_t * rsc);
void clone_free(pe_resource_t * rsc);
void pe__free_bundle(pe_resource_t *rsc);
enum rsc_role_e native_resource_state(const pe_resource_t * rsc, gboolean current);
enum rsc_role_e group_resource_state(const pe_resource_t * rsc, gboolean current);
enum rsc_role_e clone_resource_state(const pe_resource_t * rsc, gboolean current);
enum rsc_role_e pe__bundle_resource_state(const pe_resource_t *rsc,
gboolean current);
void pe__count_common(pe_resource_t *rsc);
void pe__count_bundle(pe_resource_t *rsc);
gboolean common_unpack(xmlNode * xml_obj, pe_resource_t ** rsc, pe_resource_t * parent,
pe_working_set_t * data_set);
void common_free(pe_resource_t * rsc);
pe_node_t *pe__copy_node(const pe_node_t *this_node);
extern time_t get_effective_time(pe_working_set_t * data_set);
/* Failure handling utilities (from failcounts.c) */
// bit flags for fail count handling options
enum pe_fc_flags_e {
pe_fc_default = (1 << 0),
pe_fc_effective = (1 << 1), // don't count expired failures
pe_fc_fillers = (1 << 2), // if container, include filler failures in count
};
int pe_get_failcount(pe_node_t *node, pe_resource_t *rsc, time_t *last_failure,
uint32_t flags, xmlNode *xml_op,
pe_working_set_t *data_set);
pe_action_t *pe__clear_failcount(pe_resource_t *rsc, pe_node_t *node,
const char *reason,
pe_working_set_t *data_set);
/* Functions for finding/counting a resource's active nodes */
pe_node_t *pe__find_active_on(const pe_resource_t *rsc,
unsigned int *count_all,
unsigned int *count_clean);
pe_node_t *pe__find_active_requires(const pe_resource_t *rsc,
unsigned int *count);
static inline pe_node_t *
pe__current_node(const pe_resource_t *rsc)
{
return pe__find_active_on(rsc, NULL, NULL);
}
/* Binary like operators for lists of nodes */
extern void node_list_exclude(GHashTable * list, GList *list2, gboolean merge_scores);
GHashTable *pe__node_list2table(GList *list);
static inline gpointer
pe_hash_table_lookup(GHashTable * hash, gconstpointer key)
{
if (hash) {
return g_hash_table_lookup(hash, key);
}
return NULL;
}
extern pe_action_t *get_pseudo_op(const char *name, pe_working_set_t * data_set);
extern gboolean order_actions(pe_action_t * lh_action, pe_action_t * rh_action, enum pe_ordering order);
/* Printing functions for debug */
extern void print_str_str(gpointer key, gpointer value, gpointer user_data);
extern void pe__output_node(pe_node_t * node, gboolean details, pcmk__output_t *out);
void pe__show_node_weights_as(const char *file, const char *function,
int line, bool to_log, pe_resource_t *rsc,
const char *comment, GHashTable *nodes,
pe_working_set_t *data_set);
#define pe__show_node_weights(level, rsc, text, nodes, data_set) \
pe__show_node_weights_as(__FILE__, __func__, __LINE__, \
(level), (rsc), (text), (nodes), (data_set))
/* Sorting functions */
extern gint sort_rsc_priority(gconstpointer a, gconstpointer b);
extern gint sort_rsc_index(gconstpointer a, gconstpointer b);
extern xmlNode *find_rsc_op_entry(pe_resource_t * rsc, const char *key);
extern pe_action_t *custom_action(pe_resource_t * rsc, char *key, const char *task, pe_node_t * on_node,
gboolean optional, gboolean foo, pe_working_set_t * data_set);
# define delete_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_DELETE, 0)
# define delete_action(rsc, node, optional) custom_action( \
rsc, delete_key(rsc), CRMD_ACTION_DELETE, node, \
optional, TRUE, data_set);
# define stopped_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_STOPPED, 0)
# define stopped_action(rsc, node, optional) custom_action( \
rsc, stopped_key(rsc), CRMD_ACTION_STOPPED, node, \
optional, TRUE, data_set);
# define stop_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_STOP, 0)
# define stop_action(rsc, node, optional) custom_action( \
rsc, stop_key(rsc), CRMD_ACTION_STOP, node, \
optional, TRUE, data_set);
# define reload_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_RELOAD_AGENT, 0)
# define start_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_START, 0)
# define start_action(rsc, node, optional) custom_action( \
rsc, start_key(rsc), CRMD_ACTION_START, node, \
optional, TRUE, data_set)
# define started_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_STARTED, 0)
# define started_action(rsc, node, optional) custom_action( \
rsc, started_key(rsc), CRMD_ACTION_STARTED, node, \
optional, TRUE, data_set)
# define promote_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_PROMOTE, 0)
# define promote_action(rsc, node, optional) custom_action( \
rsc, promote_key(rsc), CRMD_ACTION_PROMOTE, node, \
optional, TRUE, data_set)
# define promoted_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_PROMOTED, 0)
# define promoted_action(rsc, node, optional) custom_action( \
rsc, promoted_key(rsc), CRMD_ACTION_PROMOTED, node, \
optional, TRUE, data_set)
# define demote_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_DEMOTE, 0)
# define demote_action(rsc, node, optional) custom_action( \
rsc, demote_key(rsc), CRMD_ACTION_DEMOTE, node, \
optional, TRUE, data_set)
# define demoted_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_DEMOTED, 0)
# define demoted_action(rsc, node, optional) custom_action( \
rsc, demoted_key(rsc), CRMD_ACTION_DEMOTED, node, \
optional, TRUE, data_set)
extern int pe_get_configured_timeout(pe_resource_t *rsc, const char *action,
pe_working_set_t *data_set);
extern pe_action_t *find_first_action(GList *input, const char *uuid, const char *task,
pe_node_t * on_node);
extern enum action_tasks get_complex_task(pe_resource_t * rsc, const char *name,
gboolean allow_non_atomic);
extern GList *find_actions(GList *input, const char *key, const pe_node_t *on_node);
GList *find_actions_exact(GList *input, const char *key,
const pe_node_t *on_node);
extern GList *find_recurring_actions(GList *input, pe_node_t * not_on_node);
GList *pe__resource_actions(const pe_resource_t *rsc, const pe_node_t *node,
const char *task, bool require_node);
extern void pe_free_action(pe_action_t * action);
extern void resource_location(pe_resource_t * rsc, pe_node_t * node, int score, const char *tag,
pe_working_set_t * data_set);
extern gint sort_op_by_callid(gconstpointer a, gconstpointer b);
extern gboolean get_target_role(pe_resource_t * rsc, enum rsc_role_e *role);
extern pe_resource_t *find_clone_instance(pe_resource_t * rsc, const char *sub_id,
pe_working_set_t * data_set);
extern void destroy_ticket(gpointer data);
extern pe_ticket_t *ticket_new(const char *ticket_id, pe_working_set_t * data_set);
// Resources for manipulating resource names
const char *pe_base_name_end(const char *id);
char *clone_strip(const char *last_rsc_id);
char *clone_zero(const char *last_rsc_id);
static inline bool
pe_base_name_eq(pe_resource_t *rsc, const char *id)
{
if (id && rsc && rsc->id) {
// Number of characters in rsc->id before any clone suffix
size_t base_len = pe_base_name_end(rsc->id) - rsc->id + 1;
return (strlen(id) == base_len) && !strncmp(id, rsc->id, base_len);
}
return FALSE;
}
int pe__target_rc_from_xml(xmlNode *xml_op);
gint sort_node_uname(gconstpointer a, gconstpointer b);
bool is_set_recursive(pe_resource_t * rsc, long long flag, bool any);
enum rsc_digest_cmp_val {
/*! Digests are the same */
RSC_DIGEST_MATCH = 0,
/*! Params that require a restart changed */
RSC_DIGEST_RESTART,
/*! Some parameter changed. */
RSC_DIGEST_ALL,
/*! rsc op didn't have a digest associated with it, so
* it is unknown if parameters changed or not. */
RSC_DIGEST_UNKNOWN,
};
typedef struct op_digest_cache_s {
enum rsc_digest_cmp_val rc;
xmlNode *params_all;
xmlNode *params_secure;
xmlNode *params_restart;
char *digest_all_calc;
char *digest_secure_calc;
char *digest_restart_calc;
} op_digest_cache_t;
op_digest_cache_t *pe__calculate_digests(pe_resource_t *rsc, const char *task,
guint *interval_ms, pe_node_t *node,
xmlNode *xml_op, GHashTable *overrides,
bool calc_secure,
pe_working_set_t *data_set);
void pe__free_digests(gpointer ptr);
op_digest_cache_t *rsc_action_digest_cmp(pe_resource_t * rsc, xmlNode * xml_op, pe_node_t * node,
pe_working_set_t * data_set);
pe_action_t *pe_fence_op(pe_node_t * node, const char *op, bool optional, const char *reason, bool priority_delay, pe_working_set_t * data_set);
void trigger_unfencing(
pe_resource_t * rsc, pe_node_t *node, const char *reason, pe_action_t *dependency, pe_working_set_t * data_set);
void pe_action_set_reason(pe_action_t *action, const char *reason, bool overwrite);
void pe_action_set_flag_reason(const char *function, long line, pe_action_t *action, pe_action_t *reason, const char *text, enum pe_action_flags flags, bool overwrite);
#define pe_action_required(action, reason, text) \
pe_action_set_flag_reason(__func__, __LINE__, action, reason, text, \
pe_action_optional, FALSE)
#define pe_action_implies(action, reason, flag) \
pe_action_set_flag_reason(__func__, __LINE__, action, reason, NULL, \
flag, FALSE)
void pe__set_resource_flags_recursive(pe_resource_t *rsc, uint64_t flags);
void pe__clear_resource_flags_recursive(pe_resource_t *rsc, uint64_t flags);
void pe__clear_resource_flags_on_all(pe_working_set_t *data_set, uint64_t flag);
gboolean add_tag_ref(GHashTable * tags, const char * tag_name, const char * obj_ref);
void print_rscs_brief(GList *rsc_list, const char * pre_text, long options,
void * print_data, gboolean print_all);
int pe__rscs_brief_output(pcmk__output_t *out, GList *rsc_list, long options, gboolean print_all);
void pe_fence_node(pe_working_set_t * data_set, pe_node_t * node, const char *reason, bool priority_delay);
pe_node_t *pe_create_node(const char *id, const char *uname, const char *type,
const char *score, pe_working_set_t * data_set);
void common_print(pe_resource_t * rsc, const char *pre_text, const char *name, pe_node_t *node, long options, void *print_data);
int pe__common_output_text(pcmk__output_t *out, pe_resource_t * rsc, const char *name, pe_node_t *node, long options);
int pe__common_output_html(pcmk__output_t *out, pe_resource_t * rsc, const char *name, pe_node_t *node, long options);
pe_resource_t *pe__find_bundle_replica(const pe_resource_t *bundle,
const pe_node_t *node);
bool pe__bundle_needs_remote_name(pe_resource_t *rsc,
pe_working_set_t *data_set);
const char *pe__add_bundle_remote_name(pe_resource_t *rsc,
pe_working_set_t *data_set,
xmlNode *xml, const char *field);
const char *pe_node_attribute_calculated(const pe_node_t *node,
const char *name,
const pe_resource_t *rsc);
const char *pe_node_attribute_raw(pe_node_t *node, const char *name);
bool pe__is_universal_clone(pe_resource_t *rsc,
pe_working_set_t *data_set);
void pe__add_param_check(xmlNode *rsc_op, pe_resource_t *rsc, pe_node_t *node,
enum pe_check_parameters, pe_working_set_t *data_set);
void pe__foreach_param_check(pe_working_set_t *data_set,
void (*cb)(pe_resource_t*, pe_node_t*, xmlNode*,
enum pe_check_parameters,
pe_working_set_t*));
void pe__free_param_checks(pe_working_set_t *data_set);
bool pe__shutdown_requested(pe_node_t *node);
void pe__update_recheck_time(time_t recheck, pe_working_set_t *data_set);
/*!
* \internal
* \brief Register xml formatting message functions.
*/
void pe__register_messages(pcmk__output_t *out);
void pe__unpack_dataset_nvpairs(xmlNode *xml_obj, const char *set_name,
pe_rule_eval_data_t *rule_data, GHashTable *hash,
const char *always_first, gboolean overwrite,
pe_working_set_t *data_set);
bool pe__resource_is_disabled(pe_resource_t *rsc);
pe_action_t *pe__clear_resource_history(pe_resource_t *rsc, pe_node_t *node,
pe_working_set_t *data_set);
GList *pe__rscs_with_tag(pe_working_set_t *data_set, const char *tag_name);
GList *pe__unames_with_tag(pe_working_set_t *data_set, const char *tag_name);
bool pe__rsc_has_tag(pe_working_set_t *data_set, const char *rsc, const char *tag);
bool pe__uname_has_tag(pe_working_set_t *data_set, const char *node, const char *tag);
bool pe__rsc_running_on_any_node_in_list(pe_resource_t *rsc, GList *node_list);
GList *pe__filter_rsc_list(GList *rscs, GList *filter);
bool pcmk__rsc_filtered_by_node(pe_resource_t *rsc, GList *only_node);
gboolean pe__bundle_is_filtered(pe_resource_t *rsc, GList *only_rsc, gboolean check_parent);
gboolean pe__clone_is_filtered(pe_resource_t *rsc, GList *only_rsc, gboolean check_parent);
gboolean pe__group_is_filtered(pe_resource_t *rsc, GList *only_rsc, gboolean check_parent);
gboolean pe__native_is_filtered(pe_resource_t *rsc, GList *only_rsc, gboolean check_parent);
#endif
diff --git a/lib/pengine/native.c b/lib/pengine/native.c
index 20c2964694..bb19f83f0f 100644
--- a/lib/pengine/native.c
+++ b/lib/pengine/native.c
@@ -1,1355 +1,1355 @@
/*
* 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 <crm_internal.h>
#include <crm/pengine/rules.h>
#include <crm/pengine/status.h>
#include <crm/pengine/complex.h>
#include <crm/pengine/internal.h>
#include <crm/msg_xml.h>
#include <pe_status_private.h>
#define VARIANT_NATIVE 1
#include "./variant.h"
#ifdef PCMK__COMPAT_2_0
#define PROVIDER_SEP "::"
#else
#define PROVIDER_SEP ":"
#endif
/*!
* \internal
* \brief Check whether a resource is active on multiple nodes
*/
static bool
is_multiply_active(pe_resource_t *rsc)
{
unsigned int count = 0;
if (rsc->variant == pe_native) {
pe__find_active_requires(rsc, &count);
}
return count > 1;
}
static void
-native_priority_to_node(pe_resource_t * rsc, pe_node_t * node)
+native_priority_to_node(pe_resource_t * rsc, pe_node_t * node, gboolean failed)
{
int priority = 0;
- if (rsc->priority == 0) {
+ if ((rsc->priority == 0) || (failed == TRUE)) {
return;
}
if (rsc->role == RSC_ROLE_PROMOTED) {
// Promoted instance takes base priority + 1
priority = rsc->priority + 1;
} else {
priority = rsc->priority;
}
node->details->priority += priority;
pe_rsc_trace(rsc, "Node '%s' now has priority %d with %s'%s' (priority: %d%s)",
node->details->uname, node->details->priority,
(rsc->role == RSC_ROLE_PROMOTED)? "promoted " : "",
rsc->id, rsc->priority,
(rsc->role == RSC_ROLE_PROMOTED)? " + 1" : "");
/* Priority of a resource running on a guest node is added to the cluster
* node as well. */
if (node->details->remote_rsc
&& node->details->remote_rsc->container) {
GList *gIter = node->details->remote_rsc->container->running_on;
for (; gIter != NULL; gIter = gIter->next) {
pe_node_t *a_node = gIter->data;
a_node->details->priority += priority;
pe_rsc_trace(rsc, "Node '%s' now has priority %d with %s'%s' (priority: %d%s) "
"from guest node '%s'",
a_node->details->uname, a_node->details->priority,
(rsc->role == RSC_ROLE_PROMOTED)? "promoted " : "",
rsc->id, rsc->priority,
(rsc->role == RSC_ROLE_PROMOTED)? " + 1" : "",
node->details->uname);
}
}
}
void
-native_add_running(pe_resource_t * rsc, pe_node_t * node, pe_working_set_t * data_set)
+native_add_running(pe_resource_t * rsc, pe_node_t * node, pe_working_set_t * data_set, gboolean failed)
{
GList *gIter = rsc->running_on;
CRM_CHECK(node != NULL, return);
for (; gIter != NULL; gIter = gIter->next) {
pe_node_t *a_node = (pe_node_t *) gIter->data;
CRM_CHECK(a_node != NULL, return);
if (pcmk__str_eq(a_node->details->id, node->details->id, pcmk__str_casei)) {
return;
}
}
pe_rsc_trace(rsc, "Adding %s to %s %s", rsc->id, node->details->uname,
pcmk_is_set(rsc->flags, pe_rsc_managed)? "" : "(unmanaged)");
rsc->running_on = g_list_append(rsc->running_on, node);
if (rsc->variant == pe_native) {
node->details->running_rsc = g_list_append(node->details->running_rsc, rsc);
- native_priority_to_node(rsc, node);
+ native_priority_to_node(rsc, node, failed);
}
if (rsc->variant == pe_native && node->details->maintenance) {
pe__clear_resource_flags(rsc, pe_rsc_managed);
}
if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) {
pe_resource_t *p = rsc->parent;
pe_rsc_info(rsc, "resource %s isn't managed", rsc->id);
resource_location(rsc, node, INFINITY, "not_managed_default", data_set);
while(p && node->details->online) {
/* add without the additional location constraint */
p->running_on = g_list_append(p->running_on, node);
p = p->parent;
}
return;
}
if (is_multiply_active(rsc)) {
switch (rsc->recovery_type) {
case recovery_stop_only:
{
GHashTableIter gIter;
pe_node_t *local_node = NULL;
/* make sure it doesn't come up again */
if (rsc->allowed_nodes != NULL) {
g_hash_table_destroy(rsc->allowed_nodes);
}
rsc->allowed_nodes = pe__node_list2table(data_set->nodes);
g_hash_table_iter_init(&gIter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&gIter, NULL, (void **)&local_node)) {
local_node->weight = -INFINITY;
}
}
break;
case recovery_stop_start:
break;
case recovery_block:
pe__clear_resource_flags(rsc, pe_rsc_managed);
pe__set_resource_flags(rsc, pe_rsc_block);
/* If the resource belongs to a group or bundle configured with
* multiple-active=block, block the entire entity.
*/
if (rsc->parent
&& (rsc->parent->variant == pe_group || rsc->parent->variant == pe_container)
&& rsc->parent->recovery_type == recovery_block) {
GList *gIter = rsc->parent->children;
for (; gIter != NULL; gIter = gIter->next) {
pe_resource_t *child = (pe_resource_t *) gIter->data;
pe__clear_resource_flags(child, pe_rsc_managed);
pe__set_resource_flags(child, pe_rsc_block);
}
}
break;
}
crm_debug("%s is active on multiple nodes including %s: %s",
rsc->id, node->details->uname,
recovery2text(rsc->recovery_type));
} else {
pe_rsc_trace(rsc, "Resource %s is active on: %s", rsc->id, node->details->uname);
}
if (rsc->parent != NULL) {
- native_add_running(rsc->parent, node, data_set);
+ native_add_running(rsc->parent, node, data_set, FALSE);
}
}
static void
recursive_clear_unique(pe_resource_t *rsc)
{
pe__clear_resource_flags(rsc, pe_rsc_unique);
add_hash_param(rsc->meta, XML_RSC_ATTR_UNIQUE, XML_BOOLEAN_FALSE);
for (GList *child = rsc->children; child != NULL; child = child->next) {
recursive_clear_unique((pe_resource_t *) child->data);
}
}
gboolean
native_unpack(pe_resource_t * rsc, pe_working_set_t * data_set)
{
pe_resource_t *parent = uber_parent(rsc);
native_variant_data_t *native_data = NULL;
const char *standard = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
uint32_t ra_caps = pcmk_get_ra_caps(standard);
pe_rsc_trace(rsc, "Processing resource %s...", rsc->id);
native_data = calloc(1, sizeof(native_variant_data_t));
rsc->variant_opaque = native_data;
// Only some agent standards support unique and promotable clones
if (!pcmk_is_set(ra_caps, pcmk_ra_cap_unique)
&& pcmk_is_set(rsc->flags, pe_rsc_unique) && pe_rsc_is_clone(parent)) {
/* @COMPAT We should probably reject this situation as an error (as we
* do for promotable below) rather than warn and convert, but that would
* be a backward-incompatible change that we should probably do with a
* transform at a schema major version bump.
*/
pe__force_anon(standard, parent, rsc->id, data_set);
/* Clear globally-unique on the parent and all its descendents unpacked
* so far (clearing the parent should make any future children unpacking
* correct). We have to clear this resource explicitly because it isn't
* hooked into the parent's children yet.
*/
recursive_clear_unique(parent);
recursive_clear_unique(rsc);
}
if (!pcmk_is_set(ra_caps, pcmk_ra_cap_promotable)
&& pcmk_is_set(parent->flags, pe_rsc_promotable)) {
pe_err("Resource %s is of type %s and therefore "
"cannot be used as a promotable clone resource",
rsc->id, standard);
return FALSE;
}
return TRUE;
}
static bool
rsc_is_on_node(pe_resource_t *rsc, const pe_node_t *node, int flags)
{
pe_rsc_trace(rsc, "Checking whether %s is on %s",
rsc->id, node->details->uname);
if (pcmk_is_set(flags, pe_find_current) && rsc->running_on) {
for (GList *iter = rsc->running_on; iter; iter = iter->next) {
pe_node_t *loc = (pe_node_t *) iter->data;
if (loc->details == node->details) {
return TRUE;
}
}
} else if (pcmk_is_set(flags, pe_find_inactive)
&& (rsc->running_on == NULL)) {
return TRUE;
} else if (!pcmk_is_set(flags, pe_find_current) && rsc->allocated_to
&& (rsc->allocated_to->details == node->details)) {
return TRUE;
}
return FALSE;
}
pe_resource_t *
native_find_rsc(pe_resource_t * rsc, const char *id, const pe_node_t *on_node,
int flags)
{
bool match = FALSE;
pe_resource_t *result = NULL;
CRM_CHECK(id && rsc && rsc->id, return NULL);
if (flags & pe_find_clone) {
const char *rid = ID(rsc->xml);
if (!pe_rsc_is_clone(uber_parent(rsc))) {
match = FALSE;
} else if (!strcmp(id, rsc->id) || pcmk__str_eq(id, rid, pcmk__str_casei)) {
match = TRUE;
}
} else if (!strcmp(id, rsc->id)) {
match = TRUE;
} else if (pcmk_is_set(flags, pe_find_renamed)
&& rsc->clone_name && strcmp(rsc->clone_name, id) == 0) {
match = TRUE;
} else if (pcmk_is_set(flags, pe_find_any)
|| (pcmk_is_set(flags, pe_find_anon)
&& !pcmk_is_set(rsc->flags, pe_rsc_unique))) {
match = pe_base_name_eq(rsc, id);
}
if (match && on_node) {
bool match_node = rsc_is_on_node(rsc, on_node, flags);
if (match_node == FALSE) {
match = FALSE;
}
}
if (match) {
return rsc;
}
for (GList *gIter = rsc->children; gIter != NULL; gIter = gIter->next) {
pe_resource_t *child = (pe_resource_t *) gIter->data;
result = rsc->fns->find_rsc(child, id, on_node, flags);
if (result) {
return result;
}
}
return NULL;
}
// create is ignored
char *
native_parameter(pe_resource_t * rsc, pe_node_t * node, gboolean create, const char *name,
pe_working_set_t * data_set)
{
char *value_copy = NULL;
const char *value = NULL;
GHashTable *params = NULL;
CRM_CHECK(rsc != NULL, return NULL);
CRM_CHECK(name != NULL && strlen(name) != 0, return NULL);
pe_rsc_trace(rsc, "Looking up %s in %s", name, rsc->id);
params = pe_rsc_params(rsc, node, data_set);
value = g_hash_table_lookup(params, name);
if (value == NULL) {
/* try meta attributes instead */
value = g_hash_table_lookup(rsc->meta, name);
}
if (value != NULL) {
value_copy = strdup(value);
}
return value_copy;
}
gboolean
native_active(pe_resource_t * rsc, gboolean all)
{
for (GList *gIter = rsc->running_on; gIter != NULL; gIter = gIter->next) {
pe_node_t *a_node = (pe_node_t *) gIter->data;
if (a_node->details->unclean) {
pe_rsc_trace(rsc, "Resource %s: node %s is unclean",
rsc->id, a_node->details->uname);
return TRUE;
} else if (a_node->details->online == FALSE) {
pe_rsc_trace(rsc, "Resource %s: node %s is offline",
rsc->id, a_node->details->uname);
} else {
pe_rsc_trace(rsc, "Resource %s active on %s",
rsc->id, a_node->details->uname);
return TRUE;
}
}
return FALSE;
}
struct print_data_s {
long options;
void *print_data;
};
static const char *
native_pending_state(pe_resource_t * rsc)
{
const char *pending_state = NULL;
if (pcmk__str_eq(rsc->pending_task, CRMD_ACTION_START, pcmk__str_casei)) {
pending_state = "Starting";
} else if (pcmk__str_eq(rsc->pending_task, CRMD_ACTION_STOP, pcmk__str_casei)) {
pending_state = "Stopping";
} else if (pcmk__str_eq(rsc->pending_task, CRMD_ACTION_MIGRATE, pcmk__str_casei)) {
pending_state = "Migrating";
} else if (pcmk__str_eq(rsc->pending_task, CRMD_ACTION_MIGRATED, pcmk__str_casei)) {
/* Work might be done in here. */
pending_state = "Migrating";
} else if (pcmk__str_eq(rsc->pending_task, CRMD_ACTION_PROMOTE, pcmk__str_casei)) {
pending_state = "Promoting";
} else if (pcmk__str_eq(rsc->pending_task, CRMD_ACTION_DEMOTE, pcmk__str_casei)) {
pending_state = "Demoting";
}
return pending_state;
}
static const char *
native_pending_task(pe_resource_t * rsc)
{
const char *pending_task = NULL;
if (pcmk__str_eq(rsc->pending_task, CRMD_ACTION_STATUS, pcmk__str_casei)) {
pending_task = "Monitoring";
/* Pending probes are not printed, even if pending
* operations are requested. If someone ever requests that
* behavior, uncomment this and the corresponding part of
* unpack.c:unpack_rsc_op().
*/
/*
} else if (pcmk__str_eq(rsc->pending_task, "probe", pcmk__str_casei)) {
pending_task = "Checking";
*/
}
return pending_task;
}
static enum rsc_role_e
native_displayable_role(pe_resource_t *rsc)
{
enum rsc_role_e role = rsc->role;
if ((role == RSC_ROLE_STARTED)
&& pcmk_is_set(uber_parent(rsc)->flags, pe_rsc_promotable)) {
role = RSC_ROLE_UNPROMOTED;
}
return role;
}
static const char *
native_displayable_state(pe_resource_t *rsc, long options)
{
const char *rsc_state = NULL;
if (options & pe_print_pending) {
rsc_state = native_pending_state(rsc);
}
if (rsc_state == NULL) {
rsc_state = role2text(native_displayable_role(rsc));
}
return rsc_state;
}
static void
native_print_xml(pe_resource_t * rsc, const char *pre_text, long options, void *print_data)
{
const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
const char *prov = crm_element_value(rsc->xml, XML_AGENT_ATTR_PROVIDER);
const char *rsc_state = native_displayable_state(rsc, options);
const char *target_role = NULL;
/* resource information. */
status_print("%s<resource ", pre_text);
status_print("id=\"%s\" ", rsc_printable_id(rsc));
status_print("resource_agent=\"%s%s%s:%s\" ", class,
((prov == NULL)? "" : PROVIDER_SEP),
((prov == NULL)? "" : prov),
crm_element_value(rsc->xml, XML_ATTR_TYPE));
status_print("role=\"%s\" ", rsc_state);
if (rsc->meta) {
target_role = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE);
}
if (target_role) {
status_print("target_role=\"%s\" ", target_role);
}
status_print("active=\"%s\" ", pcmk__btoa(rsc->fns->active(rsc, TRUE)));
status_print("orphaned=\"%s\" ", pe__rsc_bool_str(rsc, pe_rsc_orphan));
status_print("blocked=\"%s\" ", pe__rsc_bool_str(rsc, pe_rsc_block));
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("failure_ignored=\"%s\" ",
pe__rsc_bool_str(rsc, pe_rsc_failure_ignored));
status_print("nodes_running_on=\"%d\" ", g_list_length(rsc->running_on));
if (options & pe_print_pending) {
const char *pending_task = native_pending_task(rsc);
if (pending_task) {
status_print("pending=\"%s\" ", pending_task);
}
}
/* print out the nodes this resource is running on */
if (options & pe_print_rsconly) {
status_print("/>\n");
/* do nothing */
} else if (rsc->running_on != NULL) {
GList *gIter = rsc->running_on;
status_print(">\n");
for (; gIter != NULL; gIter = gIter->next) {
pe_node_t *node = (pe_node_t *) gIter->data;
status_print("%s <node name=\"%s\" id=\"%s\" cached=\"%s\"/>\n", pre_text,
node->details->uname, node->details->id,
pcmk__btoa(node->details->online == FALSE));
}
status_print("%s</resource>\n", pre_text);
} else {
status_print("/>\n");
}
}
// Append a flag to resource description string's flags list
static bool
add_output_flag(GString *s, const char *flag_desc, bool have_flags)
{
g_string_append(s, (have_flags? ", " : " ("));
g_string_append(s, flag_desc);
return true;
}
// Append a node name to resource description string's node list
static bool
add_output_node(GString *s, const char *node, bool have_nodes)
{
g_string_append(s, (have_nodes? " " : " [ "));
g_string_append(s, node);
return true;
}
/*!
* \internal
* \brief Create a string description of a resource
*
* \param[in] rsc Resource to describe
* \param[in] name Desired identifier for the resource
* \param[in] node If not NULL, node that resource is "on"
* \param[in] options Bitmask of pe_print_*
* \param[in] target_role Resource's target role
* \param[in] show_nodes Whether to display nodes when multiply active
*
* \return Newly allocated string description of resource
* \note Caller must free the result with g_free().
*/
gchar *
pcmk__native_output_string(pe_resource_t *rsc, const char *name, pe_node_t *node,
long options, const char *target_role, bool show_nodes)
{
const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
const char *provider = NULL;
const char *kind = crm_element_value(rsc->xml, XML_ATTR_TYPE);
gchar *retval = NULL;
GString *outstr = NULL;
bool have_flags = false;
if (rsc->variant != pe_native) {
return NULL;
}
CRM_CHECK(name != NULL, name = "unknown");
CRM_CHECK(kind != NULL, kind = "unknown");
CRM_CHECK(class != NULL, class = "unknown");
if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider)) {
provider = crm_element_value(rsc->xml, XML_AGENT_ATTR_PROVIDER);
}
if ((node == NULL) && (rsc->lock_node != NULL)) {
node = rsc->lock_node;
}
if (pcmk_is_set(options, pe_print_rsconly)
|| pcmk__list_of_multiple(rsc->running_on)) {
node = NULL;
}
// We need a string of at least this size
outstr = g_string_sized_new(strlen(name) + strlen(class) + strlen(kind)
+ (provider? (strlen(provider) + 2) : 0)
+ (node? strlen(node->details->uname) + 1 : 0)
+ 11);
// Resource name and agent
g_string_printf(outstr, "%s\t(%s%s%s:%s):\t", name, class,
((provider == NULL)? "" : PROVIDER_SEP),
((provider == NULL)? "" : provider), kind);
// State on node
if (pcmk_is_set(rsc->flags, pe_rsc_orphan)) {
g_string_append(outstr, " ORPHANED");
}
if (pcmk_is_set(rsc->flags, pe_rsc_failed)) {
enum rsc_role_e role = native_displayable_role(rsc);
if (role > RSC_ROLE_UNPROMOTED) {
g_string_append_printf(outstr, " FAILED %s", role2text(role));
} else {
g_string_append(outstr, " FAILED");
}
} else {
g_string_append_printf(outstr, " %s", native_displayable_state(rsc, options));
}
if (node) {
g_string_append_printf(outstr, " %s", node->details->uname);
}
// Flags, as: (<flag> [...])
if (node && !(node->details->online) && node->details->unclean) {
have_flags = add_output_flag(outstr, "UNCLEAN", have_flags);
}
if (node && (node == rsc->lock_node)) {
have_flags = add_output_flag(outstr, "LOCKED", have_flags);
}
if (pcmk_is_set(options, pe_print_pending)) {
const char *pending_task = native_pending_task(rsc);
if (pending_task) {
have_flags = add_output_flag(outstr, pending_task, have_flags);
}
}
if (target_role) {
enum rsc_role_e target_role_e = text2role(target_role);
/* Only show target role if it limits our abilities (i.e. ignore
* Started, as it is the default anyways, and doesn't prevent the
* resource from becoming promoted).
*/
if (target_role_e == RSC_ROLE_STOPPED) {
have_flags = add_output_flag(outstr, "disabled", have_flags);
} else if (pcmk_is_set(uber_parent(rsc)->flags, pe_rsc_promotable)
&& target_role_e == RSC_ROLE_UNPROMOTED) {
have_flags = add_output_flag(outstr, "target-role:", have_flags);
g_string_append(outstr, target_role);
}
}
if (pcmk_is_set(rsc->flags, pe_rsc_block)) {
have_flags = add_output_flag(outstr, "blocked", have_flags);
} else if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) {
have_flags = add_output_flag(outstr, "unmanaged", have_flags);
}
if (pcmk_is_set(rsc->flags, pe_rsc_failure_ignored)) {
have_flags = add_output_flag(outstr, "failure ignored", have_flags);
}
if (have_flags) {
g_string_append(outstr, ")");
}
// User-supplied description
if (pcmk_is_set(options, pe_print_rsconly)
|| pcmk__list_of_multiple(rsc->running_on)) {
const char *desc = crm_element_value(rsc->xml, XML_ATTR_DESC);
if (desc) {
g_string_append_printf(outstr, " %s", desc);
}
}
if (show_nodes && !pcmk_is_set(options, pe_print_rsconly)
&& pcmk__list_of_multiple(rsc->running_on)) {
bool have_nodes = false;
for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) {
pe_node_t *n = (pe_node_t *) iter->data;
have_nodes = add_output_node(outstr, n->details->uname, have_nodes);
}
if (have_nodes) {
g_string_append(outstr, " ]");
}
}
retval = outstr->str;
g_string_free(outstr, FALSE);
return retval;
}
int
pe__common_output_html(pcmk__output_t *out, pe_resource_t * rsc,
const char *name, pe_node_t *node, long options)
{
const char *kind = crm_element_value(rsc->xml, XML_ATTR_TYPE);
const char *target_role = NULL;
xmlNodePtr list_node = NULL;
const char *cl = NULL;
CRM_ASSERT(rsc->variant == pe_native);
CRM_ASSERT(kind != NULL);
if (rsc->meta) {
const char *is_internal = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_INTERNAL_RSC);
if (crm_is_true(is_internal)
&& !pcmk_is_set(options, pe_print_implicit)) {
crm_trace("skipping print of internal resource %s", rsc->id);
return pcmk_rc_no_output;
}
target_role = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE);
}
if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) {
cl = "rsc-managed";
} else if (pcmk_is_set(rsc->flags, pe_rsc_failed)) {
cl = "rsc-failed";
} else if (rsc->variant == pe_native && (rsc->running_on == NULL)) {
cl = "rsc-failed";
} else if (pcmk__list_of_multiple(rsc->running_on)) {
cl = "rsc-multiple";
} else if (pcmk_is_set(rsc->flags, pe_rsc_failure_ignored)) {
cl = "rsc-failure-ignored";
} else {
cl = "rsc-ok";
}
{
gchar *s = pcmk__native_output_string(rsc, name, node, options,
target_role, true);
list_node = pcmk__output_create_html_node(out, "li", NULL, NULL, NULL);
pcmk_create_html_node(list_node, "span", NULL, cl, s);
g_free(s);
}
return pcmk_rc_ok;
}
int
pe__common_output_text(pcmk__output_t *out, pe_resource_t * rsc,
const char *name, pe_node_t *node, long options)
{
const char *target_role = NULL;
CRM_ASSERT(rsc->variant == pe_native);
if (rsc->meta) {
const char *is_internal = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_INTERNAL_RSC);
if (crm_is_true(is_internal)
&& !pcmk_is_set(options, pe_print_implicit)) {
crm_trace("skipping print of internal resource %s", rsc->id);
return pcmk_rc_no_output;
}
target_role = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE);
}
{
gchar *s = pcmk__native_output_string(rsc, name, node, options,
target_role, true);
out->list_item(out, NULL, "%s", s);
g_free(s);
}
return pcmk_rc_ok;
}
void
common_print(pe_resource_t * rsc, const char *pre_text, const char *name, pe_node_t *node, long options, void *print_data)
{
const char *target_role = NULL;
CRM_ASSERT(rsc->variant == pe_native);
if (rsc->meta) {
const char *is_internal = g_hash_table_lookup(rsc->meta,
XML_RSC_ATTR_INTERNAL_RSC);
if (crm_is_true(is_internal)
&& !pcmk_is_set(options, pe_print_implicit)) {
crm_trace("skipping print of internal resource %s", rsc->id);
return;
}
target_role = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE);
}
if (options & pe_print_xml) {
native_print_xml(rsc, pre_text, options, print_data);
return;
}
if ((pre_text == NULL) && (options & pe_print_printf)) {
pre_text = " ";
}
if (options & pe_print_html) {
if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) {
status_print("<font color=\"yellow\">");
} else if (pcmk_is_set(rsc->flags, pe_rsc_failed)) {
status_print("<font color=\"red\">");
} else if (rsc->running_on == NULL) {
status_print("<font color=\"red\">");
} else if (pcmk__list_of_multiple(rsc->running_on)) {
status_print("<font color=\"orange\">");
} else if (pcmk_is_set(rsc->flags, pe_rsc_failure_ignored)) {
status_print("<font color=\"yellow\">");
} else {
status_print("<font color=\"green\">");
}
}
{
gchar *resource_s = pcmk__native_output_string(rsc, name, node, options,
target_role, false);
status_print("%s%s", (pre_text? pre_text : ""), resource_s);
g_free(resource_s);
}
if (pcmk_is_set(options, pe_print_html)) {
status_print(" </font> ");
}
if (!pcmk_is_set(options, pe_print_rsconly)
&& pcmk__list_of_multiple(rsc->running_on)) {
GList *gIter = rsc->running_on;
int counter = 0;
if (options & pe_print_html) {
status_print("<ul>\n");
} else if ((options & pe_print_printf)
|| (options & pe_print_ncurses)) {
status_print("[");
}
for (; gIter != NULL; gIter = gIter->next) {
pe_node_t *n = (pe_node_t *) gIter->data;
counter++;
if (options & pe_print_html) {
status_print("<li>\n%s", n->details->uname);
} else if ((options & pe_print_printf)
|| (options & pe_print_ncurses)) {
status_print(" %s", n->details->uname);
} else if ((options & pe_print_log)) {
status_print("\t%d : %s", counter, n->details->uname);
} else {
status_print("%s", n->details->uname);
}
if (options & pe_print_html) {
status_print("</li>\n");
}
}
if (options & pe_print_html) {
status_print("</ul>\n");
} else if ((options & pe_print_printf)
|| (options & pe_print_ncurses)) {
status_print(" ]");
}
}
if (options & pe_print_html) {
status_print("<br/>\n");
} else if (options & pe_print_suppres_nl) {
/* nothing */
} else if ((options & pe_print_printf) || (options & pe_print_ncurses)) {
status_print("\n");
}
}
void
native_print(pe_resource_t * rsc, const char *pre_text, long options, void *print_data)
{
pe_node_t *node = NULL;
CRM_ASSERT(rsc->variant == pe_native);
if (options & pe_print_xml) {
native_print_xml(rsc, pre_text, options, print_data);
return;
}
node = pe__current_node(rsc);
if (node == NULL) {
// This is set only if a non-probe action is pending on this node
node = rsc->pending_node;
}
common_print(rsc, pre_text, rsc_printable_id(rsc), node, options, print_data);
}
PCMK__OUTPUT_ARGS("primitive", "unsigned int", "pe_resource_t *", "GList *", "GList *")
int
pe__resource_xml(pcmk__output_t *out, va_list args)
{
unsigned int options = va_arg(args, unsigned int);
pe_resource_t *rsc = va_arg(args, pe_resource_t *);
GList *only_node G_GNUC_UNUSED = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
const char *prov = crm_element_value(rsc->xml, XML_AGENT_ATTR_PROVIDER);
const char *rsc_state = native_displayable_state(rsc, options);
long is_print_pending = options & pe_print_pending;
char ra_name[LINE_MAX];
char *nodes_running_on = NULL;
char *priority = NULL;
int rc = pcmk_rc_no_output;
const char *target_role = NULL;
if (rsc->meta != NULL) {
target_role = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE);
}
CRM_ASSERT(rsc->variant == pe_native);
if (rsc->fns->is_filtered(rsc, only_rsc, TRUE)) {
return pcmk_rc_no_output;
}
/* resource information. */
snprintf(ra_name, LINE_MAX, "%s%s%s:%s", class,
((prov == NULL)? "" : PROVIDER_SEP), ((prov == NULL)? "" : prov),
crm_element_value(rsc->xml, XML_ATTR_TYPE));
nodes_running_on = pcmk__itoa(g_list_length(rsc->running_on));
priority = pcmk__ftoa(rsc->priority);
rc = pe__name_and_nvpairs_xml(out, true, "resource", 12,
"id", rsc_printable_id(rsc),
"resource_agent", ra_name,
"role", rsc_state,
"target_role", target_role,
"active", pcmk__btoa(rsc->fns->active(rsc, TRUE)),
"orphaned", pe__rsc_bool_str(rsc, pe_rsc_orphan),
"blocked", pe__rsc_bool_str(rsc, pe_rsc_block),
"managed", pe__rsc_bool_str(rsc, pe_rsc_managed),
"failed", pe__rsc_bool_str(rsc, pe_rsc_failed),
"failure_ignored", pe__rsc_bool_str(rsc, pe_rsc_failure_ignored),
"nodes_running_on", nodes_running_on,
"pending", (is_print_pending? native_pending_task(rsc) : NULL));
free(priority);
free(nodes_running_on);
CRM_ASSERT(rc == pcmk_rc_ok);
if (rsc->running_on != NULL) {
GList *gIter = rsc->running_on;
for (; gIter != NULL; gIter = gIter->next) {
pe_node_t *node = (pe_node_t *) gIter->data;
rc = pe__name_and_nvpairs_xml(out, false, "node", 3,
"name", node->details->uname,
"id", node->details->id,
"cached", pcmk__btoa(node->details->online));
CRM_ASSERT(rc == pcmk_rc_ok);
}
}
pcmk__output_xml_pop_parent(out);
return rc;
}
PCMK__OUTPUT_ARGS("primitive", "unsigned int", "pe_resource_t *", "GList *", "GList *")
int
pe__resource_html(pcmk__output_t *out, va_list args)
{
unsigned int options = va_arg(args, unsigned int);
pe_resource_t *rsc = va_arg(args, pe_resource_t *);
GList *only_node G_GNUC_UNUSED = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
pe_node_t *node = pe__current_node(rsc);
if (rsc->fns->is_filtered(rsc, only_rsc, TRUE)) {
return pcmk_rc_no_output;
}
CRM_ASSERT(rsc->variant == pe_native);
if (node == NULL) {
// This is set only if a non-probe action is pending on this node
node = rsc->pending_node;
}
return pe__common_output_html(out, rsc, rsc_printable_id(rsc), node, options);
}
PCMK__OUTPUT_ARGS("primitive", "unsigned int", "pe_resource_t *", "GList *", "GList *")
int
pe__resource_text(pcmk__output_t *out, va_list args)
{
unsigned int options = va_arg(args, unsigned int);
pe_resource_t *rsc = va_arg(args, pe_resource_t *);
GList *only_node G_GNUC_UNUSED = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
pe_node_t *node = pe__current_node(rsc);
CRM_ASSERT(rsc->variant == pe_native);
if (rsc->fns->is_filtered(rsc, only_rsc, TRUE)) {
return pcmk_rc_no_output;
}
if (node == NULL) {
// This is set only if a non-probe action is pending on this node
node = rsc->pending_node;
}
return pe__common_output_text(out, rsc, rsc_printable_id(rsc), node, options);
}
void
native_free(pe_resource_t * rsc)
{
pe_rsc_trace(rsc, "Freeing resource action list (not the data)");
common_free(rsc);
}
enum rsc_role_e
native_resource_state(const pe_resource_t * rsc, gboolean current)
{
enum rsc_role_e role = rsc->next_role;
if (current) {
role = rsc->role;
}
pe_rsc_trace(rsc, "%s state: %s", rsc->id, role2text(role));
return role;
}
/*!
* \internal
* \brief List nodes where a resource (or any of its children) is
*
* \param[in] rsc Resource to check
* \param[out] list List to add result to
* \param[in] current 0 = where allocated, 1 = where running,
* 2 = where running or pending
*
* \return If list contains only one node, that node, or NULL otherwise
*/
pe_node_t *
native_location(const pe_resource_t *rsc, GList **list, int current)
{
pe_node_t *one = NULL;
GList *result = NULL;
if (rsc->children) {
GList *gIter = rsc->children;
for (; gIter != NULL; gIter = gIter->next) {
pe_resource_t *child = (pe_resource_t *) gIter->data;
child->fns->location(child, &result, current);
}
} else if (current) {
if (rsc->running_on) {
result = g_list_copy(rsc->running_on);
}
if ((current == 2) && rsc->pending_node
&& !pe_find_node_id(result, rsc->pending_node->details->id)) {
result = g_list_append(result, rsc->pending_node);
}
} else if (current == FALSE && rsc->allocated_to) {
result = g_list_append(NULL, rsc->allocated_to);
}
if (result && (result->next == NULL)) {
one = result->data;
}
if (list) {
GList *gIter = result;
for (; gIter != NULL; gIter = gIter->next) {
pe_node_t *node = (pe_node_t *) gIter->data;
if (*list == NULL || pe_find_node_id(*list, node->details->id) == NULL) {
*list = g_list_append(*list, node);
}
}
}
g_list_free(result);
return one;
}
static void
get_rscs_brief(GList *rsc_list, GHashTable * rsc_table, GHashTable * active_table)
{
GList *gIter = rsc_list;
for (; gIter != NULL; gIter = gIter->next) {
pe_resource_t *rsc = (pe_resource_t *) gIter->data;
const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
const char *kind = crm_element_value(rsc->xml, XML_ATTR_TYPE);
int offset = 0;
char buffer[LINE_MAX];
int *rsc_counter = NULL;
int *active_counter = NULL;
if (rsc->variant != pe_native) {
continue;
}
offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", class);
if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider)) {
const char *prov = crm_element_value(rsc->xml, XML_AGENT_ATTR_PROVIDER);
if (prov != NULL) {
offset += snprintf(buffer + offset, LINE_MAX - offset,
PROVIDER_SEP "%s", prov);
}
}
offset += snprintf(buffer + offset, LINE_MAX - offset, ":%s", kind);
CRM_LOG_ASSERT(offset > 0);
if (rsc_table) {
rsc_counter = g_hash_table_lookup(rsc_table, buffer);
if (rsc_counter == NULL) {
rsc_counter = calloc(1, sizeof(int));
*rsc_counter = 0;
g_hash_table_insert(rsc_table, strdup(buffer), rsc_counter);
}
(*rsc_counter)++;
}
if (active_table) {
GList *gIter2 = rsc->running_on;
for (; gIter2 != NULL; gIter2 = gIter2->next) {
pe_node_t *node = (pe_node_t *) gIter2->data;
GHashTable *node_table = NULL;
if (node->details->unclean == FALSE && node->details->online == FALSE) {
continue;
}
node_table = g_hash_table_lookup(active_table, node->details->uname);
if (node_table == NULL) {
node_table = pcmk__strkey_table(free, free);
g_hash_table_insert(active_table, strdup(node->details->uname), node_table);
}
active_counter = g_hash_table_lookup(node_table, buffer);
if (active_counter == NULL) {
active_counter = calloc(1, sizeof(int));
*active_counter = 0;
g_hash_table_insert(node_table, strdup(buffer), active_counter);
}
(*active_counter)++;
}
}
}
}
static void
destroy_node_table(gpointer data)
{
GHashTable *node_table = data;
if (node_table) {
g_hash_table_destroy(node_table);
}
}
void
print_rscs_brief(GList *rsc_list, const char *pre_text, long options,
void *print_data, gboolean print_all)
{
GHashTable *rsc_table = pcmk__strkey_table(free, free);
GHashTable *active_table = pcmk__strkey_table(free, destroy_node_table);
GHashTableIter hash_iter;
char *type = NULL;
int *rsc_counter = NULL;
get_rscs_brief(rsc_list, rsc_table, active_table);
g_hash_table_iter_init(&hash_iter, rsc_table);
while (g_hash_table_iter_next(&hash_iter, (gpointer *)&type, (gpointer *)&rsc_counter)) {
GHashTableIter hash_iter2;
char *node_name = NULL;
GHashTable *node_table = NULL;
int active_counter_all = 0;
g_hash_table_iter_init(&hash_iter2, active_table);
while (g_hash_table_iter_next(&hash_iter2, (gpointer *)&node_name, (gpointer *)&node_table)) {
int *active_counter = g_hash_table_lookup(node_table, type);
if (active_counter == NULL || *active_counter == 0) {
continue;
} else {
active_counter_all += *active_counter;
}
if (options & pe_print_rsconly) {
node_name = NULL;
}
if (options & pe_print_html) {
status_print("<li>\n");
}
if (print_all) {
status_print("%s%d/%d\t(%s):\tActive %s\n", pre_text ? pre_text : "",
active_counter ? *active_counter : 0,
rsc_counter ? *rsc_counter : 0, type,
active_counter && (*active_counter > 0) && node_name ? node_name : "");
} else {
status_print("%s%d\t(%s):\tActive %s\n", pre_text ? pre_text : "",
active_counter ? *active_counter : 0, type,
active_counter && (*active_counter > 0) && node_name ? node_name : "");
}
if (options & pe_print_html) {
status_print("</li>\n");
}
}
if (print_all && active_counter_all == 0) {
if (options & pe_print_html) {
status_print("<li>\n");
}
status_print("%s%d/%d\t(%s):\tActive\n", pre_text ? pre_text : "",
active_counter_all,
rsc_counter ? *rsc_counter : 0, type);
if (options & pe_print_html) {
status_print("</li>\n");
}
}
}
if (rsc_table) {
g_hash_table_destroy(rsc_table);
rsc_table = NULL;
}
if (active_table) {
g_hash_table_destroy(active_table);
active_table = NULL;
}
}
int
pe__rscs_brief_output(pcmk__output_t *out, GList *rsc_list, long options, gboolean print_all)
{
GHashTable *rsc_table = pcmk__strkey_table(free, free);
GHashTable *active_table = pcmk__strkey_table(free, destroy_node_table);
GList *sorted_rscs;
int rc = pcmk_rc_no_output;
get_rscs_brief(rsc_list, rsc_table, active_table);
/* Make a list of the rsc_table keys so that it can be sorted. This is to make sure
* output order stays consistent between systems.
*/
sorted_rscs = g_hash_table_get_keys(rsc_table);
sorted_rscs = g_list_sort(sorted_rscs, (GCompareFunc) strcmp);
for (GList *gIter = sorted_rscs; gIter; gIter = gIter->next) {
char *type = (char *) gIter->data;
int *rsc_counter = g_hash_table_lookup(rsc_table, type);
GHashTableIter hash_iter2;
char *node_name = NULL;
GHashTable *node_table = NULL;
int active_counter_all = 0;
g_hash_table_iter_init(&hash_iter2, active_table);
while (g_hash_table_iter_next(&hash_iter2, (gpointer *)&node_name, (gpointer *)&node_table)) {
int *active_counter = g_hash_table_lookup(node_table, type);
if (active_counter == NULL || *active_counter == 0) {
continue;
} else {
active_counter_all += *active_counter;
}
if (options & pe_print_rsconly) {
node_name = NULL;
}
if (print_all) {
out->list_item(out, NULL, "%d/%d\t(%s):\tActive %s",
*active_counter,
rsc_counter ? *rsc_counter : 0, type,
(*active_counter > 0) && node_name ? node_name : "");
} else {
out->list_item(out, NULL, "%d\t(%s):\tActive %s",
*active_counter, type,
(*active_counter > 0) && node_name ? node_name : "");
}
rc = pcmk_rc_ok;
}
if (print_all && active_counter_all == 0) {
out->list_item(out, NULL, "%d/%d\t(%s):\tActive",
active_counter_all,
rsc_counter ? *rsc_counter : 0, type);
rc = pcmk_rc_ok;
}
}
if (rsc_table) {
g_hash_table_destroy(rsc_table);
rsc_table = NULL;
}
if (active_table) {
g_hash_table_destroy(active_table);
active_table = NULL;
}
if (sorted_rscs) {
g_list_free(sorted_rscs);
}
return rc;
}
gboolean
pe__native_is_filtered(pe_resource_t *rsc, GList *only_rsc, gboolean check_parent)
{
if (pcmk__str_in_list(only_rsc, rsc_printable_id(rsc)) ||
pcmk__str_in_list(only_rsc, rsc->id)) {
return FALSE;
} else if (check_parent) {
pe_resource_t *up = uber_parent(rsc);
if (pe_rsc_is_bundled(rsc)) {
return up->parent->fns->is_filtered(up->parent, only_rsc, FALSE);
} else {
return up->fns->is_filtered(up, only_rsc, FALSE);
}
}
return TRUE;
}
diff --git a/lib/pengine/unpack.c b/lib/pengine/unpack.c
index e3d60017a1..dc40c074e3 100644
--- a/lib/pengine/unpack.c
+++ b/lib/pengine/unpack.c
@@ -1,4105 +1,4109 @@
/*
* 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 <crm_internal.h>
#include <stdio.h>
#include <string.h>
#include <glib.h>
#include <time.h>
#include <crm/crm.h>
#include <crm/services.h>
#include <crm/msg_xml.h>
#include <crm/common/xml.h>
#include <crm/common/xml_internal.h>
#include <crm/common/util.h>
#include <crm/pengine/rules.h>
#include <crm/pengine/internal.h>
#include <crm/common/iso8601_internal.h>
#include <pe_status_private.h>
CRM_TRACE_INIT_DATA(pe_status);
/* This uses pcmk__set_flags_as()/pcmk__clear_flags_as() directly rather than
* use pe__set_working_set_flags()/pe__clear_working_set_flags() so that the
* flag is stringified more readably in log messages.
*/
#define set_config_flag(data_set, option, flag) do { \
const char *scf_value = pe_pref((data_set)->config_hash, (option)); \
if (scf_value != NULL) { \
if (crm_is_true(scf_value)) { \
(data_set)->flags = pcmk__set_flags_as(__func__, __LINE__, \
LOG_TRACE, "Working set", \
crm_system_name, (data_set)->flags, \
(flag), #flag); \
} else { \
(data_set)->flags = pcmk__clear_flags_as(__func__, __LINE__,\
LOG_TRACE, "Working set", \
crm_system_name, (data_set)->flags, \
(flag), #flag); \
} \
} \
} while(0)
static void unpack_rsc_op(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op,
xmlNode **last_failure,
enum action_fail_response *failed,
pe_working_set_t *data_set);
static void determine_remote_online_status(pe_working_set_t *data_set,
pe_node_t *this_node);
static void add_node_attrs(xmlNode *attrs, pe_node_t *node, bool overwrite,
pe_working_set_t *data_set);
static void determine_online_status(xmlNode *node_state, pe_node_t *this_node,
pe_working_set_t *data_set);
static void unpack_node_lrm(pe_node_t *node, xmlNode *xml,
pe_working_set_t *data_set);
// Bitmask for warnings we only want to print once
uint32_t pe_wo = 0;
static gboolean
is_dangling_guest_node(pe_node_t *node)
{
/* we are looking for a remote-node that was supposed to be mapped to a
* container resource, but all traces of that container have disappeared
* from both the config and the status section. */
if (pe__is_guest_or_remote_node(node) &&
node->details->remote_rsc &&
node->details->remote_rsc->container == NULL &&
pcmk_is_set(node->details->remote_rsc->flags,
pe_rsc_orphan_container_filler)) {
return TRUE;
}
return FALSE;
}
/*!
* \brief Schedule a fence action for a node
*
* \param[in,out] data_set Current working set of cluster
* \param[in,out] node Node to fence
* \param[in] reason Text description of why fencing is needed
* \param[in] priority_delay Whether to consider `priority-fencing-delay`
*/
void
pe_fence_node(pe_working_set_t * data_set, pe_node_t * node,
const char *reason, bool priority_delay)
{
CRM_CHECK(node, return);
/* A guest node is fenced by marking its container as failed */
if (pe__is_guest_node(node)) {
pe_resource_t *rsc = node->details->remote_rsc->container;
if (!pcmk_is_set(rsc->flags, pe_rsc_failed)) {
if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) {
crm_notice("Not fencing guest node %s "
"(otherwise would because %s): "
"its guest resource %s is unmanaged",
node->details->uname, reason, rsc->id);
} else {
crm_warn("Guest node %s will be fenced "
"(by recovering its guest resource %s): %s",
node->details->uname, rsc->id, reason);
/* We don't mark the node as unclean because that would prevent the
* node from running resources. We want to allow it to run resources
* in this transition if the recovery succeeds.
*/
node->details->remote_requires_reset = TRUE;
pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop);
}
}
} else if (is_dangling_guest_node(node)) {
crm_info("Cleaning up dangling connection for guest node %s: "
"fencing was already done because %s, "
"and guest resource no longer exists",
node->details->uname, reason);
pe__set_resource_flags(node->details->remote_rsc,
pe_rsc_failed|pe_rsc_stop);
} else if (pe__is_remote_node(node)) {
pe_resource_t *rsc = node->details->remote_rsc;
if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pe_rsc_managed)) {
crm_notice("Not fencing remote node %s "
"(otherwise would because %s): connection is unmanaged",
node->details->uname, reason);
} else if(node->details->remote_requires_reset == FALSE) {
node->details->remote_requires_reset = TRUE;
crm_warn("Remote node %s %s: %s",
node->details->uname,
pe_can_fence(data_set, node)? "will be fenced" : "is unclean",
reason);
}
node->details->unclean = TRUE;
// No need to apply `priority-fencing-delay` for remote nodes
pe_fence_op(node, NULL, TRUE, reason, FALSE, data_set);
} else if (node->details->unclean) {
crm_trace("Cluster node %s %s because %s",
node->details->uname,
pe_can_fence(data_set, node)? "would also be fenced" : "also is unclean",
reason);
} else {
crm_warn("Cluster node %s %s: %s",
node->details->uname,
pe_can_fence(data_set, node)? "will be fenced" : "is unclean",
reason);
node->details->unclean = TRUE;
pe_fence_op(node, NULL, TRUE, reason, priority_delay, data_set);
}
}
// @TODO xpaths can't handle templates, rules, or id-refs
// nvpair with provides or requires set to unfencing
#define XPATH_UNFENCING_NVPAIR XML_CIB_TAG_NVPAIR \
"[(@" XML_NVPAIR_ATTR_NAME "='" PCMK_STONITH_PROVIDES "'" \
"or @" XML_NVPAIR_ATTR_NAME "='" XML_RSC_ATTR_REQUIRES "') " \
"and @" XML_NVPAIR_ATTR_VALUE "='unfencing']"
// unfencing in rsc_defaults or any resource
#define XPATH_ENABLE_UNFENCING \
"/" XML_TAG_CIB "/" XML_CIB_TAG_CONFIGURATION "/" XML_CIB_TAG_RESOURCES \
"//" XML_TAG_META_SETS "/" XPATH_UNFENCING_NVPAIR \
"|/" XML_TAG_CIB "/" XML_CIB_TAG_CONFIGURATION "/" XML_CIB_TAG_RSCCONFIG \
"/" XML_TAG_META_SETS "/" XPATH_UNFENCING_NVPAIR
static void
set_if_xpath(uint64_t flag, const char *xpath, pe_working_set_t *data_set)
{
xmlXPathObjectPtr result = NULL;
if (!pcmk_is_set(data_set->flags, flag)) {
result = xpath_search(data_set->input, xpath);
if (result && (numXpathResults(result) > 0)) {
pe__set_working_set_flags(data_set, flag);
}
freeXpathObject(result);
}
}
gboolean
unpack_config(xmlNode * config, pe_working_set_t * data_set)
{
const char *value = NULL;
GHashTable *config_hash = pcmk__strkey_table(free, free);
pe_rule_eval_data_t rule_data = {
.node_hash = NULL,
.role = RSC_ROLE_UNKNOWN,
.now = data_set->now,
.match_data = NULL,
.rsc_data = NULL,
.op_data = NULL
};
data_set->config_hash = config_hash;
pe__unpack_dataset_nvpairs(config, XML_CIB_TAG_PROPSET, &rule_data, config_hash,
CIB_OPTIONS_FIRST, FALSE, data_set);
verify_pe_options(data_set->config_hash);
set_config_flag(data_set, "enable-startup-probes", pe_flag_startup_probes);
if (!pcmk_is_set(data_set->flags, pe_flag_startup_probes)) {
crm_info("Startup probes: disabled (dangerous)");
}
value = pe_pref(data_set->config_hash, XML_ATTR_HAVE_WATCHDOG);
if (value && crm_is_true(value)) {
crm_info("Watchdog-based self-fencing will be performed via SBD if "
"fencing is required and stonith-watchdog-timeout is nonzero");
pe__set_working_set_flags(data_set, pe_flag_have_stonith_resource);
}
/* Set certain flags via xpath here, so they can be used before the relevant
* configuration sections are unpacked.
*/
set_if_xpath(pe_flag_enable_unfencing, XPATH_ENABLE_UNFENCING, data_set);
value = pe_pref(data_set->config_hash, "stonith-timeout");
data_set->stonith_timeout = (int) crm_parse_interval_spec(value);
crm_debug("STONITH timeout: %d", data_set->stonith_timeout);
set_config_flag(data_set, "stonith-enabled", pe_flag_stonith_enabled);
crm_debug("STONITH of failed nodes is %s",
pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)? "enabled" : "disabled");
data_set->stonith_action = pe_pref(data_set->config_hash, "stonith-action");
if (!strcmp(data_set->stonith_action, "poweroff")) {
pe_warn_once(pe_wo_poweroff,
"Support for stonith-action of 'poweroff' is deprecated "
"and will be removed in a future release (use 'off' instead)");
data_set->stonith_action = "off";
}
crm_trace("STONITH will %s nodes", data_set->stonith_action);
set_config_flag(data_set, "concurrent-fencing", pe_flag_concurrent_fencing);
crm_debug("Concurrent fencing is %s",
pcmk_is_set(data_set->flags, pe_flag_concurrent_fencing)? "enabled" : "disabled");
value = pe_pref(data_set->config_hash,
XML_CONFIG_ATTR_PRIORITY_FENCING_DELAY);
if (value) {
data_set->priority_fencing_delay = crm_parse_interval_spec(value) / 1000;
crm_trace("Priority fencing delay is %ds", data_set->priority_fencing_delay);
}
set_config_flag(data_set, "stop-all-resources", pe_flag_stop_everything);
crm_debug("Stop all active resources: %s",
pcmk__btoa(pcmk_is_set(data_set->flags, pe_flag_stop_everything)));
set_config_flag(data_set, "symmetric-cluster", pe_flag_symmetric_cluster);
if (pcmk_is_set(data_set->flags, pe_flag_symmetric_cluster)) {
crm_debug("Cluster is symmetric" " - resources can run anywhere by default");
}
value = pe_pref(data_set->config_hash, "no-quorum-policy");
if (pcmk__str_eq(value, "ignore", pcmk__str_casei)) {
data_set->no_quorum_policy = no_quorum_ignore;
} else if (pcmk__str_eq(value, "freeze", pcmk__str_casei)) {
data_set->no_quorum_policy = no_quorum_freeze;
} else if (pcmk__str_eq(value, "demote", pcmk__str_casei)) {
data_set->no_quorum_policy = no_quorum_demote;
} else if (pcmk__str_eq(value, "suicide", pcmk__str_casei)) {
if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) {
int do_panic = 0;
crm_element_value_int(data_set->input, XML_ATTR_QUORUM_PANIC,
&do_panic);
if (do_panic || pcmk_is_set(data_set->flags, pe_flag_have_quorum)) {
data_set->no_quorum_policy = no_quorum_suicide;
} else {
crm_notice("Resetting no-quorum-policy to 'stop': cluster has never had quorum");
data_set->no_quorum_policy = no_quorum_stop;
}
} else {
pcmk__config_err("Resetting no-quorum-policy to 'stop' because "
"fencing is disabled");
data_set->no_quorum_policy = no_quorum_stop;
}
} else {
data_set->no_quorum_policy = no_quorum_stop;
}
switch (data_set->no_quorum_policy) {
case no_quorum_freeze:
crm_debug("On loss of quorum: Freeze resources");
break;
case no_quorum_stop:
crm_debug("On loss of quorum: Stop ALL resources");
break;
case no_quorum_demote:
crm_debug("On loss of quorum: "
"Demote promotable resources and stop other resources");
break;
case no_quorum_suicide:
crm_notice("On loss of quorum: Fence all remaining nodes");
break;
case no_quorum_ignore:
crm_notice("On loss of quorum: Ignore");
break;
}
set_config_flag(data_set, "stop-orphan-resources", pe_flag_stop_rsc_orphans);
crm_trace("Orphan resources are %s",
pcmk_is_set(data_set->flags, pe_flag_stop_rsc_orphans)? "stopped" : "ignored");
set_config_flag(data_set, "stop-orphan-actions", pe_flag_stop_action_orphans);
crm_trace("Orphan resource actions are %s",
pcmk_is_set(data_set->flags, pe_flag_stop_action_orphans)? "stopped" : "ignored");
value = pe_pref(data_set->config_hash, "remove-after-stop");
if (value != NULL) {
if (crm_is_true(value)) {
pe__set_working_set_flags(data_set, pe_flag_remove_after_stop);
#ifndef PCMK__COMPAT_2_0
pe_warn_once(pe_wo_remove_after,
"Support for the remove-after-stop cluster property is"
" deprecated and will be removed in a future release");
#endif
} else {
pe__clear_working_set_flags(data_set, pe_flag_remove_after_stop);
}
}
set_config_flag(data_set, "maintenance-mode", pe_flag_maintenance_mode);
crm_trace("Maintenance mode: %s",
pcmk__btoa(pcmk_is_set(data_set->flags, pe_flag_maintenance_mode)));
set_config_flag(data_set, "start-failure-is-fatal", pe_flag_start_failure_fatal);
crm_trace("Start failures are %s",
pcmk_is_set(data_set->flags, pe_flag_start_failure_fatal)? "always fatal" : "handled by failcount");
if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) {
set_config_flag(data_set, "startup-fencing", pe_flag_startup_fencing);
}
if (pcmk_is_set(data_set->flags, pe_flag_startup_fencing)) {
crm_trace("Unseen nodes will be fenced");
} else {
pe_warn_once(pe_wo_blind, "Blind faith: not fencing unseen nodes");
}
pcmk__score_red = char2score(pe_pref(data_set->config_hash, "node-health-red"));
pcmk__score_green = char2score(pe_pref(data_set->config_hash, "node-health-green"));
pcmk__score_yellow = char2score(pe_pref(data_set->config_hash, "node-health-yellow"));
crm_debug("Node scores: 'red' = %s, 'yellow' = %s, 'green' = %s",
pe_pref(data_set->config_hash, "node-health-red"),
pe_pref(data_set->config_hash, "node-health-yellow"),
pe_pref(data_set->config_hash, "node-health-green"));
data_set->placement_strategy = pe_pref(data_set->config_hash, "placement-strategy");
crm_trace("Placement strategy: %s", data_set->placement_strategy);
set_config_flag(data_set, "shutdown-lock", pe_flag_shutdown_lock);
crm_trace("Resources will%s be locked to cleanly shut down nodes",
(pcmk_is_set(data_set->flags, pe_flag_shutdown_lock)? "" : " not"));
if (pcmk_is_set(data_set->flags, pe_flag_shutdown_lock)) {
value = pe_pref(data_set->config_hash,
XML_CONFIG_ATTR_SHUTDOWN_LOCK_LIMIT);
data_set->shutdown_lock = crm_parse_interval_spec(value) / 1000;
crm_trace("Shutdown locks expire after %us", data_set->shutdown_lock);
}
return TRUE;
}
pe_node_t *
pe_create_node(const char *id, const char *uname, const char *type,
const char *score, pe_working_set_t * data_set)
{
pe_node_t *new_node = NULL;
if (pe_find_node(data_set->nodes, uname) != NULL) {
pcmk__config_warn("More than one node entry has name '%s'", uname);
}
new_node = calloc(1, sizeof(pe_node_t));
if (new_node == NULL) {
return NULL;
}
new_node->weight = char2score(score);
new_node->fixed = FALSE;
new_node->details = calloc(1, sizeof(struct pe_node_shared_s));
if (new_node->details == NULL) {
free(new_node);
return NULL;
}
crm_trace("Creating node for entry %s/%s", uname, id);
new_node->details->id = id;
new_node->details->uname = uname;
new_node->details->online = FALSE;
new_node->details->shutdown = FALSE;
new_node->details->rsc_discovery_enabled = TRUE;
new_node->details->running_rsc = NULL;
new_node->details->type = node_ping;
if (pcmk__str_eq(type, "remote", pcmk__str_casei)) {
new_node->details->type = node_remote;
pe__set_working_set_flags(data_set, pe_flag_have_remote_nodes);
} else if (pcmk__str_eq(type, "member", pcmk__str_null_matches | pcmk__str_casei)) {
new_node->details->type = node_member;
}
new_node->details->attrs = pcmk__strkey_table(free, free);
if (pe__is_guest_or_remote_node(new_node)) {
g_hash_table_insert(new_node->details->attrs, strdup(CRM_ATTR_KIND),
strdup("remote"));
} else {
g_hash_table_insert(new_node->details->attrs, strdup(CRM_ATTR_KIND),
strdup("cluster"));
}
new_node->details->utilization = pcmk__strkey_table(free, free);
new_node->details->digest_cache = pcmk__strkey_table(free,
pe__free_digests);
data_set->nodes = g_list_insert_sorted(data_set->nodes, new_node, sort_node_uname);
return new_node;
}
static const char *
expand_remote_rsc_meta(xmlNode *xml_obj, xmlNode *parent, pe_working_set_t *data)
{
xmlNode *attr_set = NULL;
xmlNode *attr = NULL;
const char *container_id = ID(xml_obj);
const char *remote_name = NULL;
const char *remote_server = NULL;
const char *remote_port = NULL;
const char *connect_timeout = "60s";
const char *remote_allow_migrate=NULL;
const char *is_managed = NULL;
for (attr_set = pcmk__xe_first_child(xml_obj); attr_set != NULL;
attr_set = pcmk__xe_next(attr_set)) {
if (!pcmk__str_eq((const char *)attr_set->name, XML_TAG_META_SETS,
pcmk__str_casei)) {
continue;
}
for (attr = pcmk__xe_first_child(attr_set); attr != NULL;
attr = pcmk__xe_next(attr)) {
const char *value = crm_element_value(attr, XML_NVPAIR_ATTR_VALUE);
const char *name = crm_element_value(attr, XML_NVPAIR_ATTR_NAME);
if (pcmk__str_eq(name, XML_RSC_ATTR_REMOTE_NODE, pcmk__str_casei)) {
remote_name = value;
} else if (pcmk__str_eq(name, "remote-addr", pcmk__str_casei)) {
remote_server = value;
} else if (pcmk__str_eq(name, "remote-port", pcmk__str_casei)) {
remote_port = value;
} else if (pcmk__str_eq(name, "remote-connect-timeout", pcmk__str_casei)) {
connect_timeout = value;
} else if (pcmk__str_eq(name, "remote-allow-migrate", pcmk__str_casei)) {
remote_allow_migrate=value;
} else if (pcmk__str_eq(name, XML_RSC_ATTR_MANAGED, pcmk__str_casei)) {
is_managed = value;
}
}
}
if (remote_name == NULL) {
return NULL;
}
if (pe_find_resource(data->resources, remote_name) != NULL) {
return NULL;
}
pe_create_remote_xml(parent, remote_name, container_id,
remote_allow_migrate, is_managed,
connect_timeout, remote_server, remote_port);
return remote_name;
}
static void
handle_startup_fencing(pe_working_set_t *data_set, pe_node_t *new_node)
{
if ((new_node->details->type == node_remote) && (new_node->details->remote_rsc == NULL)) {
/* Ignore fencing for remote nodes that don't have a connection resource
* associated with them. This happens when remote node entries get left
* in the nodes section after the connection resource is removed.
*/
return;
}
if (pcmk_is_set(data_set->flags, pe_flag_startup_fencing)) {
// All nodes are unclean until we've seen their status entry
new_node->details->unclean = TRUE;
} else {
// Blind faith ...
new_node->details->unclean = FALSE;
}
/* We need to be able to determine if a node's status section
* exists or not separate from whether the node is unclean. */
new_node->details->unseen = TRUE;
}
gboolean
unpack_nodes(xmlNode * xml_nodes, pe_working_set_t * data_set)
{
xmlNode *xml_obj = NULL;
pe_node_t *new_node = NULL;
const char *id = NULL;
const char *uname = NULL;
const char *type = NULL;
const char *score = NULL;
pe_rule_eval_data_t rule_data = {
.node_hash = NULL,
.role = RSC_ROLE_UNKNOWN,
.now = data_set->now,
.match_data = NULL,
.rsc_data = NULL,
.op_data = NULL
};
for (xml_obj = pcmk__xe_first_child(xml_nodes); xml_obj != NULL;
xml_obj = pcmk__xe_next(xml_obj)) {
if (pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_NODE, pcmk__str_none)) {
new_node = NULL;
id = crm_element_value(xml_obj, XML_ATTR_ID);
uname = crm_element_value(xml_obj, XML_ATTR_UNAME);
type = crm_element_value(xml_obj, XML_ATTR_TYPE);
score = crm_element_value(xml_obj, XML_RULE_ATTR_SCORE);
crm_trace("Processing node %s/%s", uname, id);
if (id == NULL) {
pcmk__config_err("Ignoring <" XML_CIB_TAG_NODE
"> entry in configuration without id");
continue;
}
new_node = pe_create_node(id, uname, type, score, data_set);
if (new_node == NULL) {
return FALSE;
}
/* if(data_set->have_quorum == FALSE */
/* && data_set->no_quorum_policy == no_quorum_stop) { */
/* /\* start shutting resources down *\/ */
/* new_node->weight = -INFINITY; */
/* } */
handle_startup_fencing(data_set, new_node);
add_node_attrs(xml_obj, new_node, FALSE, data_set);
pe__unpack_dataset_nvpairs(xml_obj, XML_TAG_UTILIZATION, &rule_data,
new_node->details->utilization, NULL,
FALSE, data_set);
crm_trace("Done with node %s", crm_element_value(xml_obj, XML_ATTR_UNAME));
}
}
if (data_set->localhost && pe_find_node(data_set->nodes, data_set->localhost) == NULL) {
crm_info("Creating a fake local node");
pe_create_node(data_set->localhost, data_set->localhost, NULL, 0,
data_set);
}
return TRUE;
}
static void
setup_container(pe_resource_t * rsc, pe_working_set_t * data_set)
{
const char *container_id = NULL;
if (rsc->children) {
GList *gIter = rsc->children;
for (; gIter != NULL; gIter = gIter->next) {
pe_resource_t *child_rsc = (pe_resource_t *) gIter->data;
setup_container(child_rsc, data_set);
}
return;
}
container_id = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_CONTAINER);
if (container_id && !pcmk__str_eq(container_id, rsc->id, pcmk__str_casei)) {
pe_resource_t *container = pe_find_resource(data_set->resources, container_id);
if (container) {
rsc->container = container;
pe__set_resource_flags(container, pe_rsc_is_container);
container->fillers = g_list_append(container->fillers, rsc);
pe_rsc_trace(rsc, "Resource %s's container is %s", rsc->id, container_id);
} else {
pe_err("Resource %s: Unknown resource container (%s)", rsc->id, container_id);
}
}
}
gboolean
unpack_remote_nodes(xmlNode * xml_resources, pe_working_set_t * data_set)
{
xmlNode *xml_obj = NULL;
/* Create remote nodes and guest nodes from the resource configuration
* before unpacking resources.
*/
for (xml_obj = pcmk__xe_first_child(xml_resources); xml_obj != NULL;
xml_obj = pcmk__xe_next(xml_obj)) {
const char *new_node_id = NULL;
/* Check for remote nodes, which are defined by ocf:pacemaker:remote
* primitives.
*/
if (xml_contains_remote_node(xml_obj)) {
new_node_id = ID(xml_obj);
/* The "pe_find_node" check is here to make sure we don't iterate over
* an expanded node that has already been added to the node list. */
if (new_node_id && pe_find_node(data_set->nodes, new_node_id) == NULL) {
crm_trace("Found remote node %s defined by resource %s",
new_node_id, ID(xml_obj));
pe_create_node(new_node_id, new_node_id, "remote", NULL,
data_set);
}
continue;
}
/* Check for guest nodes, which are defined by special meta-attributes
* of a primitive of any type (for example, VirtualDomain or Xen).
*/
if (pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_RESOURCE, pcmk__str_none)) {
/* This will add an ocf:pacemaker:remote primitive to the
* configuration for the guest node's connection, to be unpacked
* later.
*/
new_node_id = expand_remote_rsc_meta(xml_obj, xml_resources, data_set);
if (new_node_id && pe_find_node(data_set->nodes, new_node_id) == NULL) {
crm_trace("Found guest node %s in resource %s",
new_node_id, ID(xml_obj));
pe_create_node(new_node_id, new_node_id, "remote", NULL,
data_set);
}
continue;
}
/* Check for guest nodes inside a group. Clones are currently not
* supported as guest nodes.
*/
if (pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_GROUP, pcmk__str_none)) {
xmlNode *xml_obj2 = NULL;
for (xml_obj2 = pcmk__xe_first_child(xml_obj); xml_obj2 != NULL;
xml_obj2 = pcmk__xe_next(xml_obj2)) {
new_node_id = expand_remote_rsc_meta(xml_obj2, xml_resources, data_set);
if (new_node_id && pe_find_node(data_set->nodes, new_node_id) == NULL) {
crm_trace("Found guest node %s in resource %s inside group %s",
new_node_id, ID(xml_obj2), ID(xml_obj));
pe_create_node(new_node_id, new_node_id, "remote", NULL,
data_set);
}
}
}
}
return TRUE;
}
/* Call this after all the nodes and resources have been
* unpacked, but before the status section is read.
*
* A remote node's online status is reflected by the state
* of the remote node's connection resource. We need to link
* the remote node to this connection resource so we can have
* easy access to the connection resource during the scheduler calculations.
*/
static void
link_rsc2remotenode(pe_working_set_t *data_set, pe_resource_t *new_rsc)
{
pe_node_t *remote_node = NULL;
if (new_rsc->is_remote_node == FALSE) {
return;
}
if (pcmk_is_set(data_set->flags, pe_flag_quick_location)) {
/* remote_nodes and remote_resources are not linked in quick location calculations */
return;
}
remote_node = pe_find_node(data_set->nodes, new_rsc->id);
CRM_CHECK(remote_node != NULL, return;);
pe_rsc_trace(new_rsc, "Linking remote connection resource %s to node %s",
new_rsc->id, remote_node->details->uname);
remote_node->details->remote_rsc = new_rsc;
if (new_rsc->container == NULL) {
/* Handle start-up fencing for remote nodes (as opposed to guest nodes)
* the same as is done for cluster nodes.
*/
handle_startup_fencing(data_set, remote_node);
} else {
/* pe_create_node() marks the new node as "remote" or "cluster"; now
* that we know the node is a guest node, update it correctly.
*/
g_hash_table_replace(remote_node->details->attrs, strdup(CRM_ATTR_KIND),
strdup("container"));
}
}
static void
destroy_tag(gpointer data)
{
pe_tag_t *tag = data;
if (tag) {
free(tag->id);
g_list_free_full(tag->refs, free);
free(tag);
}
}
/*!
* \internal
* \brief Parse configuration XML for resource information
*
* \param[in] xml_resources Top of resource configuration XML
* \param[in,out] data_set Where to put resource information
*
* \return TRUE
*
* \note unpack_remote_nodes() MUST be called before this, so that the nodes can
* be used when common_unpack() calls resource_location()
*/
gboolean
unpack_resources(xmlNode * xml_resources, pe_working_set_t * data_set)
{
xmlNode *xml_obj = NULL;
GList *gIter = NULL;
data_set->template_rsc_sets = pcmk__strkey_table(free, destroy_tag);
for (xml_obj = pcmk__xe_first_child(xml_resources); xml_obj != NULL;
xml_obj = pcmk__xe_next(xml_obj)) {
pe_resource_t *new_rsc = NULL;
if (pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_RSC_TEMPLATE, pcmk__str_none)) {
const char *template_id = ID(xml_obj);
if (template_id && g_hash_table_lookup_extended(data_set->template_rsc_sets,
template_id, NULL, NULL) == FALSE) {
/* Record the template's ID for the knowledge of its existence anyway. */
g_hash_table_insert(data_set->template_rsc_sets, strdup(template_id), NULL);
}
continue;
}
crm_trace("Beginning unpack... <%s id=%s... >", crm_element_name(xml_obj), ID(xml_obj));
if (common_unpack(xml_obj, &new_rsc, NULL, data_set) && (new_rsc != NULL)) {
data_set->resources = g_list_append(data_set->resources, new_rsc);
pe_rsc_trace(new_rsc, "Added resource %s", new_rsc->id);
} else {
pcmk__config_err("Ignoring <%s> resource '%s' "
"because configuration is invalid",
crm_element_name(xml_obj), crm_str(ID(xml_obj)));
if (new_rsc != NULL && new_rsc->fns != NULL) {
new_rsc->fns->free(new_rsc);
}
}
}
for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) {
pe_resource_t *rsc = (pe_resource_t *) gIter->data;
setup_container(rsc, data_set);
link_rsc2remotenode(data_set, rsc);
}
data_set->resources = g_list_sort(data_set->resources, sort_rsc_priority);
if (pcmk_is_set(data_set->flags, pe_flag_quick_location)) {
/* Ignore */
} else if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)
&& !pcmk_is_set(data_set->flags, pe_flag_have_stonith_resource)) {
pcmk__config_err("Resource start-up disabled since no STONITH resources have been defined");
pcmk__config_err("Either configure some or disable STONITH with the stonith-enabled option");
pcmk__config_err("NOTE: Clusters with shared data need STONITH to ensure data integrity");
}
return TRUE;
}
gboolean
unpack_tags(xmlNode * xml_tags, pe_working_set_t * data_set)
{
xmlNode *xml_tag = NULL;
data_set->tags = pcmk__strkey_table(free, destroy_tag);
for (xml_tag = pcmk__xe_first_child(xml_tags); xml_tag != NULL;
xml_tag = pcmk__xe_next(xml_tag)) {
xmlNode *xml_obj_ref = NULL;
const char *tag_id = ID(xml_tag);
if (!pcmk__str_eq((const char *)xml_tag->name, XML_CIB_TAG_TAG, pcmk__str_none)) {
continue;
}
if (tag_id == NULL) {
pcmk__config_err("Ignoring <%s> without " XML_ATTR_ID,
crm_element_name(xml_tag));
continue;
}
for (xml_obj_ref = pcmk__xe_first_child(xml_tag); xml_obj_ref != NULL;
xml_obj_ref = pcmk__xe_next(xml_obj_ref)) {
const char *obj_ref = ID(xml_obj_ref);
if (!pcmk__str_eq((const char *)xml_obj_ref->name, XML_CIB_TAG_OBJ_REF, pcmk__str_none)) {
continue;
}
if (obj_ref == NULL) {
pcmk__config_err("Ignoring <%s> for tag '%s' without " XML_ATTR_ID,
crm_element_name(xml_obj_ref), tag_id);
continue;
}
if (add_tag_ref(data_set->tags, tag_id, obj_ref) == FALSE) {
return FALSE;
}
}
}
return TRUE;
}
/* The ticket state section:
* "/cib/status/tickets/ticket_state" */
static gboolean
unpack_ticket_state(xmlNode * xml_ticket, pe_working_set_t * data_set)
{
const char *ticket_id = NULL;
const char *granted = NULL;
const char *last_granted = NULL;
const char *standby = NULL;
xmlAttrPtr xIter = NULL;
pe_ticket_t *ticket = NULL;
ticket_id = ID(xml_ticket);
if (pcmk__str_empty(ticket_id)) {
return FALSE;
}
crm_trace("Processing ticket state for %s", ticket_id);
ticket = g_hash_table_lookup(data_set->tickets, ticket_id);
if (ticket == NULL) {
ticket = ticket_new(ticket_id, data_set);
if (ticket == NULL) {
return FALSE;
}
}
for (xIter = xml_ticket->properties; xIter; xIter = xIter->next) {
const char *prop_name = (const char *)xIter->name;
const char *prop_value = crm_element_value(xml_ticket, prop_name);
if (pcmk__str_eq(prop_name, XML_ATTR_ID, pcmk__str_none)) {
continue;
}
g_hash_table_replace(ticket->state, strdup(prop_name), strdup(prop_value));
}
granted = g_hash_table_lookup(ticket->state, "granted");
if (granted && crm_is_true(granted)) {
ticket->granted = TRUE;
crm_info("We have ticket '%s'", ticket->id);
} else {
ticket->granted = FALSE;
crm_info("We do not have ticket '%s'", ticket->id);
}
last_granted = g_hash_table_lookup(ticket->state, "last-granted");
if (last_granted) {
long long last_granted_ll;
pcmk__scan_ll(last_granted, &last_granted_ll, 0LL);
ticket->last_granted = (time_t) last_granted_ll;
}
standby = g_hash_table_lookup(ticket->state, "standby");
if (standby && crm_is_true(standby)) {
ticket->standby = TRUE;
if (ticket->granted) {
crm_info("Granted ticket '%s' is in standby-mode", ticket->id);
}
} else {
ticket->standby = FALSE;
}
crm_trace("Done with ticket state for %s", ticket_id);
return TRUE;
}
static gboolean
unpack_tickets_state(xmlNode * xml_tickets, pe_working_set_t * data_set)
{
xmlNode *xml_obj = NULL;
for (xml_obj = pcmk__xe_first_child(xml_tickets); xml_obj != NULL;
xml_obj = pcmk__xe_next(xml_obj)) {
if (!pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_TICKET_STATE, pcmk__str_none)) {
continue;
}
unpack_ticket_state(xml_obj, data_set);
}
return TRUE;
}
static void
unpack_handle_remote_attrs(pe_node_t *this_node, xmlNode *state, pe_working_set_t * data_set)
{
const char *resource_discovery_enabled = NULL;
xmlNode *attrs = NULL;
pe_resource_t *rsc = NULL;
if (!pcmk__str_eq((const char *)state->name, XML_CIB_TAG_STATE, pcmk__str_none)) {
return;
}
if ((this_node == NULL) || !pe__is_guest_or_remote_node(this_node)) {
return;
}
crm_trace("Processing remote node id=%s, uname=%s", this_node->details->id, this_node->details->uname);
pcmk__scan_min_int(crm_element_value(state, XML_NODE_IS_MAINTENANCE),
&(this_node->details->remote_maintenance), 0);
rsc = this_node->details->remote_rsc;
if (this_node->details->remote_requires_reset == FALSE) {
this_node->details->unclean = FALSE;
this_node->details->unseen = FALSE;
}
attrs = find_xml_node(state, XML_TAG_TRANSIENT_NODEATTRS, FALSE);
add_node_attrs(attrs, this_node, TRUE, data_set);
if (pe__shutdown_requested(this_node)) {
crm_info("Node %s is shutting down", this_node->details->uname);
this_node->details->shutdown = TRUE;
}
if (crm_is_true(pe_node_attribute_raw(this_node, "standby"))) {
crm_info("Node %s is in standby-mode", this_node->details->uname);
this_node->details->standby = TRUE;
}
if (crm_is_true(pe_node_attribute_raw(this_node, "maintenance")) ||
((rsc != NULL) && !pcmk_is_set(rsc->flags, pe_rsc_managed))) {
crm_info("Node %s is in maintenance-mode", this_node->details->uname);
this_node->details->maintenance = TRUE;
}
resource_discovery_enabled = pe_node_attribute_raw(this_node, XML_NODE_ATTR_RSC_DISCOVERY);
if (resource_discovery_enabled && !crm_is_true(resource_discovery_enabled)) {
if (pe__is_remote_node(this_node)
&& !pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) {
crm_warn("Ignoring %s attribute on remote node %s because stonith is disabled",
XML_NODE_ATTR_RSC_DISCOVERY, this_node->details->uname);
} else {
/* This is either a remote node with fencing enabled, or a guest
* node. We don't care whether fencing is enabled when fencing guest
* nodes, because they are "fenced" by recovering their containing
* resource.
*/
crm_info("Node %s has resource discovery disabled", this_node->details->uname);
this_node->details->rsc_discovery_enabled = FALSE;
}
}
}
/*!
* \internal
* \brief Unpack a cluster node's transient attributes
*
* \param[in] state CIB node state XML
* \param[in] node Cluster node whose attributes are being unpacked
* \param[in] data_set Cluster working set
*/
static void
unpack_transient_attributes(xmlNode *state, pe_node_t *node,
pe_working_set_t *data_set)
{
const char *discovery = NULL;
xmlNode *attrs = find_xml_node(state, XML_TAG_TRANSIENT_NODEATTRS, FALSE);
add_node_attrs(attrs, node, TRUE, data_set);
if (crm_is_true(pe_node_attribute_raw(node, "standby"))) {
crm_info("Node %s is in standby-mode", node->details->uname);
node->details->standby = TRUE;
}
if (crm_is_true(pe_node_attribute_raw(node, "maintenance"))) {
crm_info("Node %s is in maintenance-mode", node->details->uname);
node->details->maintenance = TRUE;
}
discovery = pe_node_attribute_raw(node, XML_NODE_ATTR_RSC_DISCOVERY);
if ((discovery != NULL) && !crm_is_true(discovery)) {
crm_warn("Ignoring %s attribute for node %s because disabling "
"resource discovery is not allowed for cluster nodes",
XML_NODE_ATTR_RSC_DISCOVERY, node->details->uname);
}
}
/*!
* \internal
* \brief Unpack a node state entry (first pass)
*
* Unpack one node state entry from status. This unpacks information from the
* node_state element itself and node attributes inside it, but not the
* resource history inside it. Multiple passes through the status are needed to
* fully unpack everything.
*
* \param[in] state CIB node state XML
* \param[in] data_set Cluster working set
*/
static void
unpack_node_state(xmlNode *state, pe_working_set_t *data_set)
{
const char *id = NULL;
const char *uname = NULL;
pe_node_t *this_node = NULL;
id = crm_element_value(state, XML_ATTR_ID);
if (id == NULL) {
crm_warn("Ignoring malformed " XML_CIB_TAG_STATE " entry without "
XML_ATTR_ID);
return;
}
uname = crm_element_value(state, XML_ATTR_UNAME);
if (uname == NULL) {
crm_warn("Ignoring malformed " XML_CIB_TAG_STATE " entry without "
XML_ATTR_UNAME);
return;
}
this_node = pe_find_node_any(data_set->nodes, id, uname);
if (this_node == NULL) {
pcmk__config_warn("Ignoring recorded node state for '%s' because "
"it is no longer in the configuration", uname);
return;
}
if (pe__is_guest_or_remote_node(this_node)) {
/* We can't determine the online status of Pacemaker Remote nodes until
* after all resource history has been unpacked. In this first pass, we
* do need to mark whether the node has been fenced, as this plays a
* role during unpacking cluster node resource state.
*/
pcmk__scan_min_int(crm_element_value(state, XML_NODE_IS_FENCED),
&(this_node->details->remote_was_fenced), 0);
return;
}
unpack_transient_attributes(state, this_node, data_set);
/* Provisionally mark this cluster node as clean. We have at least seen it
* in the current cluster's lifetime.
*/
this_node->details->unclean = FALSE;
this_node->details->unseen = FALSE;
crm_trace("Determining online status of cluster node %s (id %s)",
this_node->details->uname, id);
determine_online_status(state, this_node, data_set);
if (!pcmk_is_set(data_set->flags, pe_flag_have_quorum)
&& this_node->details->online
&& (data_set->no_quorum_policy == no_quorum_suicide)) {
/* Everything else should flow from this automatically
* (at least until the scheduler becomes able to migrate off
* healthy resources)
*/
pe_fence_node(data_set, this_node, "cluster does not have quorum",
FALSE);
}
}
/*!
* \internal
* \brief Unpack nodes' resource history as much as possible
*
* Unpack as many nodes' resource history as possible in one pass through the
* status. We need to process Pacemaker Remote nodes' connections/containers
* before unpacking their history; the connection/container history will be
* in another node's history, so it might take multiple passes to unpack
* everything.
*
* \param[in] status CIB XML status section
* \param[in] fence If true, treat any not-yet-unpacked nodes as unseen
* \param[in] data_set Cluster working set
*
* \return Standard Pacemaker return code (specifically pcmk_rc_ok if done,
* or EAGAIN if more unpacking remains to be done)
*/
static int
unpack_node_history(xmlNode *status, bool fence, pe_working_set_t *data_set)
{
int rc = pcmk_rc_ok;
// Loop through all node_state entries in CIB status
for (xmlNode *state = first_named_child(status, XML_CIB_TAG_STATE);
state != NULL; state = crm_next_same_xml(state)) {
const char *id = ID(state);
const char *uname = crm_element_value(state, XML_ATTR_UNAME);
pe_node_t *this_node = NULL;
if ((id == NULL) || (uname == NULL)) {
// Warning already logged in first pass through status section
crm_trace("Not unpacking resource history from malformed "
XML_CIB_TAG_STATE " without id and/or uname");
continue;
}
this_node = pe_find_node_any(data_set->nodes, id, uname);
if (this_node == NULL) {
// Warning already logged in first pass through status section
crm_trace("Not unpacking resource history for node %s because "
"no longer in configuration", id);
continue;
}
if (this_node->details->unpacked) {
crm_trace("Not unpacking resource history for node %s because "
"already unpacked", id);
continue;
}
if (fence) {
// We're processing all remaining nodes
} else if (pe__is_guest_node(this_node)) {
/* We can unpack a guest node's history only after we've unpacked
* other resource history to the point that we know that the node's
* connection and containing resource are both up.
*/
pe_resource_t *rsc = this_node->details->remote_rsc;
if ((rsc == NULL) || (rsc->role != RSC_ROLE_STARTED)
|| (rsc->container->role != RSC_ROLE_STARTED)) {
crm_trace("Not unpacking resource history for guest node %s "
"because container and connection are not known to "
"be up", id);
continue;
}
} else if (pe__is_remote_node(this_node)) {
/* We can unpack a remote node's history only after we've unpacked
* other resource history to the point that we know that the node's
* connection is up, with the exception of when shutdown locks are
* in use.
*/
pe_resource_t *rsc = this_node->details->remote_rsc;
if ((rsc == NULL)
|| (!pcmk_is_set(data_set->flags, pe_flag_shutdown_lock)
&& (rsc->role != RSC_ROLE_STARTED))) {
crm_trace("Not unpacking resource history for remote node %s "
"because connection is not known to be up", id);
continue;
}
/* If fencing and shutdown locks are disabled and we're not processing
* unseen nodes, then we don't want to unpack offline nodes until online
* nodes have been unpacked. This allows us to number active clone
* instances first.
*/
} else if (!pcmk_any_flags_set(data_set->flags, pe_flag_stonith_enabled
|pe_flag_shutdown_lock)
&& !this_node->details->online) {
crm_trace("Not unpacking resource history for offline "
"cluster node %s", id);
continue;
}
if (pe__is_guest_or_remote_node(this_node)) {
determine_remote_online_status(data_set, this_node);
unpack_handle_remote_attrs(this_node, state, data_set);
}
crm_trace("Unpacking resource history for %snode %s",
(fence? "unseen " : ""), id);
this_node->details->unpacked = TRUE;
unpack_node_lrm(this_node, state, data_set);
rc = EAGAIN; // Other node histories might depend on this one
}
return rc;
}
/* remove nodes that are down, stopping */
/* create positive rsc_to_node constraints between resources and the nodes they are running on */
/* anything else? */
gboolean
unpack_status(xmlNode * status, pe_working_set_t * data_set)
{
xmlNode *state = NULL;
crm_trace("Beginning unpack");
if (data_set->tickets == NULL) {
data_set->tickets = pcmk__strkey_table(free, destroy_ticket);
}
for (state = pcmk__xe_first_child(status); state != NULL;
state = pcmk__xe_next(state)) {
if (pcmk__str_eq((const char *)state->name, XML_CIB_TAG_TICKETS, pcmk__str_none)) {
unpack_tickets_state((xmlNode *) state, data_set);
} else if (pcmk__str_eq((const char *)state->name, XML_CIB_TAG_STATE, pcmk__str_none)) {
unpack_node_state(state, data_set);
}
}
while (unpack_node_history(status, FALSE, data_set) == EAGAIN) {
crm_trace("Another pass through node resource histories is needed");
}
// Now catch any nodes we didn't see
unpack_node_history(status,
pcmk_is_set(data_set->flags, pe_flag_stonith_enabled),
data_set);
/* Now that we know where resources are, we can schedule stops of containers
* with failed bundle connections
*/
if (data_set->stop_needed != NULL) {
for (GList *item = data_set->stop_needed; item; item = item->next) {
pe_resource_t *container = item->data;
pe_node_t *node = pe__current_node(container);
if (node) {
stop_action(container, node, FALSE);
}
}
g_list_free(data_set->stop_needed);
data_set->stop_needed = NULL;
}
/* Now that we know status of all Pacemaker Remote connections and nodes,
* we can stop connections for node shutdowns, and check the online status
* of remote/guest nodes that didn't have any node history to unpack.
*/
for (GList *gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) {
pe_node_t *this_node = gIter->data;
if (!pe__is_guest_or_remote_node(this_node)) {
continue;
}
if (this_node->details->shutdown
&& (this_node->details->remote_rsc != NULL)) {
pe__set_next_role(this_node->details->remote_rsc, RSC_ROLE_STOPPED,
"remote shutdown");
}
if (!this_node->details->unpacked) {
determine_remote_online_status(data_set, this_node);
}
}
return TRUE;
}
static gboolean
determine_online_status_no_fencing(pe_working_set_t * data_set, xmlNode * node_state,
pe_node_t * this_node)
{
gboolean online = FALSE;
const char *join = crm_element_value(node_state, XML_NODE_JOIN_STATE);
const char *is_peer = crm_element_value(node_state, XML_NODE_IS_PEER);
const char *in_cluster = crm_element_value(node_state, XML_NODE_IN_CLUSTER);
const char *exp_state = crm_element_value(node_state, XML_NODE_EXPECTED);
if (!crm_is_true(in_cluster)) {
crm_trace("Node is down: in_cluster=%s", crm_str(in_cluster));
} else if (pcmk__str_eq(is_peer, ONLINESTATUS, pcmk__str_casei)) {
if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) {
online = TRUE;
} else {
crm_debug("Node is not ready to run resources: %s", join);
}
} else if (this_node->details->expected_up == FALSE) {
crm_trace("Controller is down: in_cluster=%s", crm_str(in_cluster));
crm_trace("\tis_peer=%s, join=%s, expected=%s",
crm_str(is_peer), crm_str(join), crm_str(exp_state));
} else {
/* mark it unclean */
pe_fence_node(data_set, this_node, "peer is unexpectedly down", FALSE);
crm_info("\tin_cluster=%s, is_peer=%s, join=%s, expected=%s",
crm_str(in_cluster), crm_str(is_peer), crm_str(join), crm_str(exp_state));
}
return online;
}
static gboolean
determine_online_status_fencing(pe_working_set_t * data_set, xmlNode * node_state,
pe_node_t * this_node)
{
gboolean online = FALSE;
gboolean do_terminate = FALSE;
bool crmd_online = FALSE;
const char *join = crm_element_value(node_state, XML_NODE_JOIN_STATE);
const char *is_peer = crm_element_value(node_state, XML_NODE_IS_PEER);
const char *in_cluster = crm_element_value(node_state, XML_NODE_IN_CLUSTER);
const char *exp_state = crm_element_value(node_state, XML_NODE_EXPECTED);
const char *terminate = pe_node_attribute_raw(this_node, "terminate");
/*
- XML_NODE_IN_CLUSTER ::= true|false
- XML_NODE_IS_PEER ::= online|offline
- XML_NODE_JOIN_STATE ::= member|down|pending|banned
- XML_NODE_EXPECTED ::= member|down
*/
if (crm_is_true(terminate)) {
do_terminate = TRUE;
} else if (terminate != NULL && strlen(terminate) > 0) {
/* could be a time() value */
char t = terminate[0];
if (t != '0' && isdigit(t)) {
do_terminate = TRUE;
}
}
crm_trace("%s: in_cluster=%s, is_peer=%s, join=%s, expected=%s, term=%d",
this_node->details->uname, crm_str(in_cluster), crm_str(is_peer),
crm_str(join), crm_str(exp_state), do_terminate);
online = crm_is_true(in_cluster);
crmd_online = pcmk__str_eq(is_peer, ONLINESTATUS, pcmk__str_casei);
if (exp_state == NULL) {
exp_state = CRMD_JOINSTATE_DOWN;
}
if (this_node->details->shutdown) {
crm_debug("%s is shutting down", this_node->details->uname);
/* Slightly different criteria since we can't shut down a dead peer */
online = crmd_online;
} else if (in_cluster == NULL) {
pe_fence_node(data_set, this_node, "peer has not been seen by the cluster", FALSE);
} else if (pcmk__str_eq(join, CRMD_JOINSTATE_NACK, pcmk__str_casei)) {
pe_fence_node(data_set, this_node, "peer failed the pacemaker membership criteria", FALSE);
} else if (do_terminate == FALSE && pcmk__str_eq(exp_state, CRMD_JOINSTATE_DOWN, pcmk__str_casei)) {
if (crm_is_true(in_cluster) || crmd_online) {
crm_info("- Node %s is not ready to run resources", this_node->details->uname);
this_node->details->standby = TRUE;
this_node->details->pending = TRUE;
} else {
crm_trace("%s is down or still coming up", this_node->details->uname);
}
} else if (do_terminate && pcmk__str_eq(join, CRMD_JOINSTATE_DOWN, pcmk__str_casei)
&& crm_is_true(in_cluster) == FALSE && !crmd_online) {
crm_info("Node %s was just shot", this_node->details->uname);
online = FALSE;
} else if (crm_is_true(in_cluster) == FALSE) {
// Consider `priority-fencing-delay` for lost nodes
pe_fence_node(data_set, this_node, "peer is no longer part of the cluster", TRUE);
} else if (!crmd_online) {
pe_fence_node(data_set, this_node, "peer process is no longer available", FALSE);
/* Everything is running at this point, now check join state */
} else if (do_terminate) {
pe_fence_node(data_set, this_node, "termination was requested", FALSE);
} else if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) {
crm_info("Node %s is active", this_node->details->uname);
} else if (pcmk__strcase_any_of(join, CRMD_JOINSTATE_PENDING, CRMD_JOINSTATE_DOWN, NULL)) {
crm_info("Node %s is not ready to run resources", this_node->details->uname);
this_node->details->standby = TRUE;
this_node->details->pending = TRUE;
} else {
pe_fence_node(data_set, this_node, "peer was in an unknown state", FALSE);
crm_warn("%s: in-cluster=%s, is-peer=%s, join=%s, expected=%s, term=%d, shutdown=%d",
this_node->details->uname, crm_str(in_cluster), crm_str(is_peer),
crm_str(join), crm_str(exp_state), do_terminate, this_node->details->shutdown);
}
return online;
}
static void
determine_remote_online_status(pe_working_set_t * data_set, pe_node_t * this_node)
{
pe_resource_t *rsc = this_node->details->remote_rsc;
pe_resource_t *container = NULL;
pe_node_t *host = NULL;
/* If there is a node state entry for a (former) Pacemaker Remote node
* but no resource creating that node, the node's connection resource will
* be NULL. Consider it an offline remote node in that case.
*/
if (rsc == NULL) {
this_node->details->online = FALSE;
goto remote_online_done;
}
container = rsc->container;
if (container && pcmk__list_of_1(rsc->running_on)) {
host = rsc->running_on->data;
}
/* If the resource is currently started, mark it online. */
if (rsc->role == RSC_ROLE_STARTED) {
crm_trace("%s node %s presumed ONLINE because connection resource is started",
(container? "Guest" : "Remote"), this_node->details->id);
this_node->details->online = TRUE;
}
/* consider this node shutting down if transitioning start->stop */
if (rsc->role == RSC_ROLE_STARTED && rsc->next_role == RSC_ROLE_STOPPED) {
crm_trace("%s node %s shutting down because connection resource is stopping",
(container? "Guest" : "Remote"), this_node->details->id);
this_node->details->shutdown = TRUE;
}
/* Now check all the failure conditions. */
if(container && pcmk_is_set(container->flags, pe_rsc_failed)) {
crm_trace("Guest node %s UNCLEAN because guest resource failed",
this_node->details->id);
this_node->details->online = FALSE;
this_node->details->remote_requires_reset = TRUE;
} else if (pcmk_is_set(rsc->flags, pe_rsc_failed)) {
crm_trace("%s node %s OFFLINE because connection resource failed",
(container? "Guest" : "Remote"), this_node->details->id);
this_node->details->online = FALSE;
} else if (rsc->role == RSC_ROLE_STOPPED
|| (container && container->role == RSC_ROLE_STOPPED)) {
crm_trace("%s node %s OFFLINE because its resource is stopped",
(container? "Guest" : "Remote"), this_node->details->id);
this_node->details->online = FALSE;
this_node->details->remote_requires_reset = FALSE;
} else if (host && (host->details->online == FALSE)
&& host->details->unclean) {
crm_trace("Guest node %s UNCLEAN because host is unclean",
this_node->details->id);
this_node->details->online = FALSE;
this_node->details->remote_requires_reset = TRUE;
}
remote_online_done:
crm_trace("Remote node %s online=%s",
this_node->details->id, this_node->details->online ? "TRUE" : "FALSE");
}
static void
determine_online_status(xmlNode * node_state, pe_node_t * this_node, pe_working_set_t * data_set)
{
gboolean online = FALSE;
const char *exp_state = crm_element_value(node_state, XML_NODE_EXPECTED);
CRM_CHECK(this_node != NULL, return);
this_node->details->shutdown = FALSE;
this_node->details->expected_up = FALSE;
if (pe__shutdown_requested(this_node)) {
this_node->details->shutdown = TRUE;
} else if (pcmk__str_eq(exp_state, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) {
this_node->details->expected_up = TRUE;
}
if (this_node->details->type == node_ping) {
this_node->details->unclean = FALSE;
online = FALSE; /* As far as resource management is concerned,
* the node is safely offline.
* Anyone caught abusing this logic will be shot
*/
} else if (!pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) {
online = determine_online_status_no_fencing(data_set, node_state, this_node);
} else {
online = determine_online_status_fencing(data_set, node_state, this_node);
}
if (online) {
this_node->details->online = TRUE;
} else {
/* remove node from contention */
this_node->fixed = TRUE;
this_node->weight = -INFINITY;
}
if (online && this_node->details->shutdown) {
/* don't run resources here */
this_node->fixed = TRUE;
this_node->weight = -INFINITY;
}
if (this_node->details->type == node_ping) {
crm_info("Node %s is not a pacemaker node", this_node->details->uname);
} else if (this_node->details->unclean) {
pe_proc_warn("Node %s is unclean", this_node->details->uname);
} else if (this_node->details->online) {
crm_info("Node %s is %s", this_node->details->uname,
this_node->details->shutdown ? "shutting down" :
this_node->details->pending ? "pending" :
this_node->details->standby ? "standby" :
this_node->details->maintenance ? "maintenance" : "online");
} else {
crm_trace("Node %s is offline", this_node->details->uname);
}
}
/*!
* \internal
* \brief Find the end of a resource's name, excluding any clone suffix
*
* \param[in] id Resource ID to check
*
* \return Pointer to last character of resource's base name
*/
const char *
pe_base_name_end(const char *id)
{
if (!pcmk__str_empty(id)) {
const char *end = id + strlen(id) - 1;
for (const char *s = end; s > id; --s) {
switch (*s) {
case '0':
case '1':
case '2':
case '3':
case '4':
case '5':
case '6':
case '7':
case '8':
case '9':
break;
case ':':
return (s == end)? s : (s - 1);
default:
return end;
}
}
return end;
}
return NULL;
}
/*!
* \internal
* \brief Get a resource name excluding any clone suffix
*
* \param[in] last_rsc_id Resource ID to check
*
* \return Pointer to newly allocated string with resource's base name
* \note It is the caller's responsibility to free() the result.
* This asserts on error, so callers can assume result is not NULL.
*/
char *
clone_strip(const char *last_rsc_id)
{
const char *end = pe_base_name_end(last_rsc_id);
char *basename = NULL;
CRM_ASSERT(end);
basename = strndup(last_rsc_id, end - last_rsc_id + 1);
CRM_ASSERT(basename);
return basename;
}
/*!
* \internal
* \brief Get the name of the first instance of a cloned resource
*
* \param[in] last_rsc_id Resource ID to check
*
* \return Pointer to newly allocated string with resource's base name plus :0
* \note It is the caller's responsibility to free() the result.
* This asserts on error, so callers can assume result is not NULL.
*/
char *
clone_zero(const char *last_rsc_id)
{
const char *end = pe_base_name_end(last_rsc_id);
size_t base_name_len = end - last_rsc_id + 1;
char *zero = NULL;
CRM_ASSERT(end);
zero = calloc(base_name_len + 3, sizeof(char));
CRM_ASSERT(zero);
memcpy(zero, last_rsc_id, base_name_len);
zero[base_name_len] = ':';
zero[base_name_len + 1] = '0';
return zero;
}
static pe_resource_t *
create_fake_resource(const char *rsc_id, xmlNode * rsc_entry, pe_working_set_t * data_set)
{
pe_resource_t *rsc = NULL;
xmlNode *xml_rsc = create_xml_node(NULL, XML_CIB_TAG_RESOURCE);
copy_in_properties(xml_rsc, rsc_entry);
crm_xml_add(xml_rsc, XML_ATTR_ID, rsc_id);
crm_log_xml_debug(xml_rsc, "Orphan resource");
if (!common_unpack(xml_rsc, &rsc, NULL, data_set)) {
return NULL;
}
if (xml_contains_remote_node(xml_rsc)) {
pe_node_t *node;
crm_debug("Detected orphaned remote node %s", rsc_id);
node = pe_find_node(data_set->nodes, rsc_id);
if (node == NULL) {
node = pe_create_node(rsc_id, rsc_id, "remote", NULL, data_set);
}
link_rsc2remotenode(data_set, rsc);
if (node) {
crm_trace("Setting node %s as shutting down due to orphaned connection resource", rsc_id);
node->details->shutdown = TRUE;
}
}
if (crm_element_value(rsc_entry, XML_RSC_ATTR_CONTAINER)) {
/* This orphaned rsc needs to be mapped to a container. */
crm_trace("Detected orphaned container filler %s", rsc_id);
pe__set_resource_flags(rsc, pe_rsc_orphan_container_filler);
}
pe__set_resource_flags(rsc, pe_rsc_orphan);
data_set->resources = g_list_append(data_set->resources, rsc);
return rsc;
}
/*!
* \internal
* \brief Create orphan instance for anonymous clone resource history
*/
static pe_resource_t *
create_anonymous_orphan(pe_resource_t *parent, const char *rsc_id,
pe_node_t *node, pe_working_set_t *data_set)
{
pe_resource_t *top = pe__create_clone_child(parent, data_set);
// find_rsc() because we might be a cloned group
pe_resource_t *orphan = top->fns->find_rsc(top, rsc_id, NULL, pe_find_clone);
pe_rsc_debug(parent, "Created orphan %s for %s: %s on %s",
top->id, parent->id, rsc_id, node->details->uname);
return orphan;
}
/*!
* \internal
* \brief Check a node for an instance of an anonymous clone
*
* Return a child instance of the specified anonymous clone, in order of
* preference: (1) the instance running on the specified node, if any;
* (2) an inactive instance (i.e. within the total of clone-max instances);
* (3) a newly created orphan (i.e. clone-max instances are already active).
*
* \param[in] data_set Cluster information
* \param[in] node Node on which to check for instance
* \param[in] parent Clone to check
* \param[in] rsc_id Name of cloned resource in history (without instance)
*/
static pe_resource_t *
find_anonymous_clone(pe_working_set_t * data_set, pe_node_t * node, pe_resource_t * parent,
const char *rsc_id)
{
GList *rIter = NULL;
pe_resource_t *rsc = NULL;
pe_resource_t *inactive_instance = NULL;
gboolean skip_inactive = FALSE;
CRM_ASSERT(parent != NULL);
CRM_ASSERT(pe_rsc_is_clone(parent));
CRM_ASSERT(!pcmk_is_set(parent->flags, pe_rsc_unique));
// Check for active (or partially active, for cloned groups) instance
pe_rsc_trace(parent, "Looking for %s on %s in %s", rsc_id, node->details->uname, parent->id);
for (rIter = parent->children; rsc == NULL && rIter; rIter = rIter->next) {
GList *locations = NULL;
pe_resource_t *child = rIter->data;
/* Check whether this instance is already known to be active or pending
* anywhere, at this stage of unpacking. Because this function is called
* for a resource before the resource's individual operation history
* entries are unpacked, locations will generally not contain the
* desired node.
*
* However, there are three exceptions:
* (1) when child is a cloned group and we have already unpacked the
* history of another member of the group on the same node;
* (2) when we've already unpacked the history of another numbered
* instance on the same node (which can happen if globally-unique
* was flipped from true to false); and
* (3) when we re-run calculations on the same data set as part of a
* simulation.
*/
child->fns->location(child, &locations, 2);
if (locations) {
/* We should never associate the same numbered anonymous clone
* instance with multiple nodes, and clone instances can't migrate,
* so there must be only one location, regardless of history.
*/
CRM_LOG_ASSERT(locations->next == NULL);
if (((pe_node_t *)locations->data)->details == node->details) {
/* This child instance is active on the requested node, so check
* for a corresponding configured resource. We use find_rsc()
* instead of child because child may be a cloned group, and we
* need the particular member corresponding to rsc_id.
*
* If the history entry is orphaned, rsc will be NULL.
*/
rsc = parent->fns->find_rsc(child, rsc_id, NULL, pe_find_clone);
if (rsc) {
/* If there are multiple instance history entries for an
* anonymous clone in a single node's history (which can
* happen if globally-unique is switched from true to
* false), we want to consider the instances beyond the
* first as orphans, even if there are inactive instance
* numbers available.
*/
if (rsc->running_on) {
crm_notice("Active (now-)anonymous clone %s has "
"multiple (orphan) instance histories on %s",
parent->id, node->details->uname);
skip_inactive = TRUE;
rsc = NULL;
} else {
pe_rsc_trace(parent, "Resource %s, active", rsc->id);
}
}
}
g_list_free(locations);
} else {
pe_rsc_trace(parent, "Resource %s, skip inactive", child->id);
if (!skip_inactive && !inactive_instance
&& !pcmk_is_set(child->flags, pe_rsc_block)) {
// Remember one inactive instance in case we don't find active
inactive_instance = parent->fns->find_rsc(child, rsc_id, NULL,
pe_find_clone);
/* ... but don't use it if it was already associated with a
* pending action on another node
*/
if (inactive_instance && inactive_instance->pending_node
&& (inactive_instance->pending_node->details != node->details)) {
inactive_instance = NULL;
}
}
}
}
if ((rsc == NULL) && !skip_inactive && (inactive_instance != NULL)) {
pe_rsc_trace(parent, "Resource %s, empty slot", inactive_instance->id);
rsc = inactive_instance;
}
/* If the resource has "requires" set to "quorum" or "nothing", and we don't
* have a clone instance for every node, we don't want to consume a valid
* instance number for unclean nodes. Such instances may appear to be active
* according to the history, but should be considered inactive, so we can
* start an instance elsewhere. Treat such instances as orphans.
*
* An exception is instances running on guest nodes -- since guest node
* "fencing" is actually just a resource stop, requires shouldn't apply.
*
* @TODO Ideally, we'd use an inactive instance number if it is not needed
* for any clean instances. However, we don't know that at this point.
*/
if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pe_rsc_needs_fencing)
&& (!node->details->online || node->details->unclean)
&& !pe__is_guest_node(node)
&& !pe__is_universal_clone(parent, data_set)) {
rsc = NULL;
}
if (rsc == NULL) {
rsc = create_anonymous_orphan(parent, rsc_id, node, data_set);
pe_rsc_trace(parent, "Resource %s, orphan", rsc->id);
}
return rsc;
}
static pe_resource_t *
unpack_find_resource(pe_working_set_t * data_set, pe_node_t * node, const char *rsc_id,
xmlNode * rsc_entry)
{
pe_resource_t *rsc = NULL;
pe_resource_t *parent = NULL;
crm_trace("looking for %s", rsc_id);
rsc = pe_find_resource(data_set->resources, rsc_id);
if (rsc == NULL) {
/* If we didn't find the resource by its name in the operation history,
* check it again as a clone instance. Even when clone-max=0, we create
* a single :0 orphan to match against here.
*/
char *clone0_id = clone_zero(rsc_id);
pe_resource_t *clone0 = pe_find_resource(data_set->resources, clone0_id);
if (clone0 && !pcmk_is_set(clone0->flags, pe_rsc_unique)) {
rsc = clone0;
parent = uber_parent(clone0);
crm_trace("%s found as %s (%s)", rsc_id, clone0_id, parent->id);
} else {
crm_trace("%s is not known as %s either (orphan)",
rsc_id, clone0_id);
}
free(clone0_id);
} else if (rsc->variant > pe_native) {
crm_trace("Resource history for %s is orphaned because it is no longer primitive",
rsc_id);
return NULL;
} else {
parent = uber_parent(rsc);
}
if (pe_rsc_is_anon_clone(parent)) {
if (pe_rsc_is_bundled(parent)) {
rsc = pe__find_bundle_replica(parent->parent, node);
} else {
char *base = clone_strip(rsc_id);
rsc = find_anonymous_clone(data_set, node, parent, base);
free(base);
CRM_ASSERT(rsc != NULL);
}
}
if (rsc && !pcmk__str_eq(rsc_id, rsc->id, pcmk__str_casei)
&& !pcmk__str_eq(rsc_id, rsc->clone_name, pcmk__str_casei)) {
free(rsc->clone_name);
rsc->clone_name = strdup(rsc_id);
pe_rsc_debug(rsc, "Internally renamed %s on %s to %s%s",
rsc_id, node->details->uname, rsc->id,
(pcmk_is_set(rsc->flags, pe_rsc_orphan)? " (ORPHAN)" : ""));
}
return rsc;
}
static pe_resource_t *
process_orphan_resource(xmlNode * rsc_entry, pe_node_t * node, pe_working_set_t * data_set)
{
pe_resource_t *rsc = NULL;
const char *rsc_id = crm_element_value(rsc_entry, XML_ATTR_ID);
crm_debug("Detected orphan resource %s on %s", rsc_id, node->details->uname);
rsc = create_fake_resource(rsc_id, rsc_entry, data_set);
if (!pcmk_is_set(data_set->flags, pe_flag_stop_rsc_orphans)) {
pe__clear_resource_flags(rsc, pe_rsc_managed);
} else {
CRM_CHECK(rsc != NULL, return NULL);
pe_rsc_trace(rsc, "Added orphan %s", rsc->id);
resource_location(rsc, NULL, -INFINITY, "__orphan_do_not_run__", data_set);
}
return rsc;
}
static void
process_rsc_state(pe_resource_t * rsc, pe_node_t * node,
enum action_fail_response on_fail,
xmlNode * migrate_op, pe_working_set_t * data_set)
{
pe_node_t *tmpnode = NULL;
char *reason = NULL;
+ enum action_fail_response save_on_fail = action_fail_ignore;
CRM_ASSERT(rsc);
pe_rsc_trace(rsc, "Resource %s is %s on %s: on_fail=%s",
rsc->id, role2text(rsc->role), node->details->uname, fail2text(on_fail));
/* process current state */
if (rsc->role != RSC_ROLE_UNKNOWN) {
pe_resource_t *iter = rsc;
while (iter) {
if (g_hash_table_lookup(iter->known_on, node->details->id) == NULL) {
pe_node_t *n = pe__copy_node(node);
pe_rsc_trace(rsc, "%s (aka. %s) known on %s", rsc->id, rsc->clone_name,
n->details->uname);
g_hash_table_insert(iter->known_on, (gpointer) n->details->id, n);
}
if (pcmk_is_set(iter->flags, pe_rsc_unique)) {
break;
}
iter = iter->parent;
}
}
/* If a managed resource is believed to be running, but node is down ... */
if (rsc->role > RSC_ROLE_STOPPED
&& node->details->online == FALSE
&& node->details->maintenance == FALSE
&& pcmk_is_set(rsc->flags, pe_rsc_managed)) {
gboolean should_fence = FALSE;
/* If this is a guest node, fence it (regardless of whether fencing is
* enabled, because guest node fencing is done by recovery of the
* container resource rather than by the fencer). Mark the resource
* we're processing as failed. When the guest comes back up, its
* operation history in the CIB will be cleared, freeing the affected
* resource to run again once we are sure we know its state.
*/
if (pe__is_guest_node(node)) {
pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop);
should_fence = TRUE;
} else if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) {
if (pe__is_remote_node(node) && node->details->remote_rsc
&& !pcmk_is_set(node->details->remote_rsc->flags, pe_rsc_failed)) {
/* Setting unseen means that fencing of the remote node will
* occur only if the connection resource is not going to start
* somewhere. This allows connection resources on a failed
* cluster node to move to another node without requiring the
* remote nodes to be fenced as well.
*/
node->details->unseen = TRUE;
reason = crm_strdup_printf("%s is active there (fencing will be"
" revoked if remote connection can "
"be re-established elsewhere)",
rsc->id);
}
should_fence = TRUE;
}
if (should_fence) {
if (reason == NULL) {
reason = crm_strdup_printf("%s is thought to be active there", rsc->id);
}
pe_fence_node(data_set, node, reason, FALSE);
}
free(reason);
}
+ /* In order to calculate priority_fencing_delay correctly, save the failure information and pass it to native_add_running(). */
+ save_on_fail = on_fail;
+
if (node->details->unclean) {
/* No extra processing needed
* Also allows resources to be started again after a node is shot
*/
on_fail = action_fail_ignore;
}
switch (on_fail) {
case action_fail_ignore:
/* nothing to do */
break;
case action_fail_demote:
pe__set_resource_flags(rsc, pe_rsc_failed);
demote_action(rsc, node, FALSE);
break;
case action_fail_fence:
/* treat it as if it is still running
* but also mark the node as unclean
*/
reason = crm_strdup_printf("%s failed there", rsc->id);
pe_fence_node(data_set, node, reason, FALSE);
free(reason);
break;
case action_fail_standby:
node->details->standby = TRUE;
node->details->standby_onfail = TRUE;
break;
case action_fail_block:
/* is_managed == FALSE will prevent any
* actions being sent for the resource
*/
pe__clear_resource_flags(rsc, pe_rsc_managed);
pe__set_resource_flags(rsc, pe_rsc_block);
break;
case action_fail_migrate:
/* make sure it comes up somewhere else
* or not at all
*/
resource_location(rsc, node, -INFINITY, "__action_migration_auto__", data_set);
break;
case action_fail_stop:
pe__set_next_role(rsc, RSC_ROLE_STOPPED, "on-fail=stop");
break;
case action_fail_recover:
if (rsc->role != RSC_ROLE_STOPPED && rsc->role != RSC_ROLE_UNKNOWN) {
pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop);
stop_action(rsc, node, FALSE);
}
break;
case action_fail_restart_container:
pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop);
if (rsc->container && pe_rsc_is_bundled(rsc)) {
/* A bundle's remote connection can run on a different node than
* the bundle's container. We don't necessarily know where the
* container is running yet, so remember it and add a stop
* action for it later.
*/
data_set->stop_needed = g_list_prepend(data_set->stop_needed,
rsc->container);
} else if (rsc->container) {
stop_action(rsc->container, node, FALSE);
} else if (rsc->role != RSC_ROLE_STOPPED && rsc->role != RSC_ROLE_UNKNOWN) {
stop_action(rsc, node, FALSE);
}
break;
case action_fail_reset_remote:
pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop);
if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) {
tmpnode = NULL;
if (rsc->is_remote_node) {
tmpnode = pe_find_node(data_set->nodes, rsc->id);
}
if (tmpnode &&
pe__is_remote_node(tmpnode) &&
tmpnode->details->remote_was_fenced == 0) {
/* The remote connection resource failed in a way that
* should result in fencing the remote node.
*/
pe_fence_node(data_set, tmpnode,
"remote connection is unrecoverable", FALSE);
}
}
/* require the stop action regardless if fencing is occurring or not. */
if (rsc->role > RSC_ROLE_STOPPED) {
stop_action(rsc, node, FALSE);
}
/* if reconnect delay is in use, prevent the connection from exiting the
* "STOPPED" role until the failure is cleared by the delay timeout. */
if (rsc->remote_reconnect_ms) {
pe__set_next_role(rsc, RSC_ROLE_STOPPED, "remote reset");
}
break;
}
/* ensure a remote-node connection failure forces an unclean remote-node
* to be fenced. By setting unseen = FALSE, the remote-node failure will
* result in a fencing operation regardless if we're going to attempt to
* reconnect to the remote-node in this transition or not. */
if (pcmk_is_set(rsc->flags, pe_rsc_failed) && rsc->is_remote_node) {
tmpnode = pe_find_node(data_set->nodes, rsc->id);
if (tmpnode && tmpnode->details->unclean) {
tmpnode->details->unseen = FALSE;
}
}
if (rsc->role != RSC_ROLE_STOPPED && rsc->role != RSC_ROLE_UNKNOWN) {
if (pcmk_is_set(rsc->flags, pe_rsc_orphan)) {
if (pcmk_is_set(rsc->flags, pe_rsc_managed)) {
pcmk__config_warn("Detected active orphan %s running on %s",
rsc->id, node->details->uname);
} else {
pcmk__config_warn("Resource '%s' must be stopped manually on "
"%s because cluster is configured not to "
"stop active orphans",
rsc->id, node->details->uname);
}
}
- native_add_running(rsc, node, data_set);
+ native_add_running(rsc, node, data_set, (save_on_fail != action_fail_ignore));
switch (on_fail) {
case action_fail_ignore:
break;
case action_fail_demote:
case action_fail_block:
pe__set_resource_flags(rsc, pe_rsc_failed);
break;
default:
pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop);
break;
}
} else if (rsc->clone_name && strchr(rsc->clone_name, ':') != NULL) {
/* Only do this for older status sections that included instance numbers
* Otherwise stopped instances will appear as orphans
*/
pe_rsc_trace(rsc, "Resetting clone_name %s for %s (stopped)", rsc->clone_name, rsc->id);
free(rsc->clone_name);
rsc->clone_name = NULL;
} else {
GList *possible_matches = pe__resource_actions(rsc, node, RSC_STOP,
FALSE);
GList *gIter = possible_matches;
for (; gIter != NULL; gIter = gIter->next) {
pe_action_t *stop = (pe_action_t *) gIter->data;
pe__set_action_flags(stop, pe_action_optional);
}
g_list_free(possible_matches);
}
}
/* create active recurring operations as optional */
static void
process_recurring(pe_node_t * node, pe_resource_t * rsc,
int start_index, int stop_index,
GList *sorted_op_list, pe_working_set_t * data_set)
{
int counter = -1;
const char *task = NULL;
const char *status = NULL;
GList *gIter = sorted_op_list;
CRM_ASSERT(rsc);
pe_rsc_trace(rsc, "%s: Start index %d, stop index = %d", rsc->id, start_index, stop_index);
for (; gIter != NULL; gIter = gIter->next) {
xmlNode *rsc_op = (xmlNode *) gIter->data;
guint interval_ms = 0;
char *key = NULL;
const char *id = ID(rsc_op);
counter++;
if (node->details->online == FALSE) {
pe_rsc_trace(rsc, "Skipping %s/%s: node is offline", rsc->id, node->details->uname);
break;
/* Need to check if there's a monitor for role="Stopped" */
} else if (start_index < stop_index && counter <= stop_index) {
pe_rsc_trace(rsc, "Skipping %s/%s: resource is not active", id, node->details->uname);
continue;
} else if (counter < start_index) {
pe_rsc_trace(rsc, "Skipping %s/%s: old %d", id, node->details->uname, counter);
continue;
}
crm_element_value_ms(rsc_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms);
if (interval_ms == 0) {
pe_rsc_trace(rsc, "Skipping %s/%s: non-recurring", id, node->details->uname);
continue;
}
status = crm_element_value(rsc_op, XML_LRM_ATTR_OPSTATUS);
if (pcmk__str_eq(status, "-1", pcmk__str_casei)) {
pe_rsc_trace(rsc, "Skipping %s/%s: status", id, node->details->uname);
continue;
}
task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK);
/* create the action */
key = pcmk__op_key(rsc->id, task, interval_ms);
pe_rsc_trace(rsc, "Creating %s/%s", key, node->details->uname);
custom_action(rsc, key, task, node, TRUE, TRUE, data_set);
}
}
void
calculate_active_ops(GList *sorted_op_list, int *start_index, int *stop_index)
{
int counter = -1;
int implied_monitor_start = -1;
int implied_clone_start = -1;
const char *task = NULL;
const char *status = NULL;
GList *gIter = sorted_op_list;
*stop_index = -1;
*start_index = -1;
for (; gIter != NULL; gIter = gIter->next) {
xmlNode *rsc_op = (xmlNode *) gIter->data;
counter++;
task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK);
status = crm_element_value(rsc_op, XML_LRM_ATTR_OPSTATUS);
if (pcmk__str_eq(task, CRMD_ACTION_STOP, pcmk__str_casei)
&& pcmk__str_eq(status, "0", pcmk__str_casei)) {
*stop_index = counter;
} else if (pcmk__strcase_any_of(task, CRMD_ACTION_START, CRMD_ACTION_MIGRATED, NULL)) {
*start_index = counter;
} else if ((implied_monitor_start <= *stop_index) && pcmk__str_eq(task, CRMD_ACTION_STATUS, pcmk__str_casei)) {
const char *rc = crm_element_value(rsc_op, XML_LRM_ATTR_RC);
if (pcmk__strcase_any_of(rc, "0", "8", NULL)) {
implied_monitor_start = counter;
}
} else if (pcmk__strcase_any_of(task, CRMD_ACTION_PROMOTE, CRMD_ACTION_DEMOTE, NULL)) {
implied_clone_start = counter;
}
}
if (*start_index == -1) {
if (implied_clone_start != -1) {
*start_index = implied_clone_start;
} else if (implied_monitor_start != -1) {
*start_index = implied_monitor_start;
}
}
}
// If resource history entry has shutdown lock, remember lock node and time
static void
unpack_shutdown_lock(xmlNode *rsc_entry, pe_resource_t *rsc, pe_node_t *node,
pe_working_set_t *data_set)
{
time_t lock_time = 0; // When lock started (i.e. node shutdown time)
if ((crm_element_value_epoch(rsc_entry, XML_CONFIG_ATTR_SHUTDOWN_LOCK,
&lock_time) == pcmk_ok) && (lock_time != 0)) {
if ((data_set->shutdown_lock > 0)
&& (get_effective_time(data_set)
> (lock_time + data_set->shutdown_lock))) {
pe_rsc_info(rsc, "Shutdown lock for %s on %s expired",
rsc->id, node->details->uname);
pe__clear_resource_history(rsc, node, data_set);
} else {
rsc->lock_node = node;
rsc->lock_time = lock_time;
}
}
}
/*!
* \internal
* \brief Unpack one lrm_resource entry from a node's CIB status
*
* \param[in] node Node whose status is being unpacked
* \param[in] rsc_entry lrm_resource XML being unpacked
* \param[in] data_set Cluster working set
*
* \return Resource corresponding to the entry, or NULL if no operation history
*/
static pe_resource_t *
unpack_lrm_resource(pe_node_t *node, xmlNode *lrm_resource,
pe_working_set_t *data_set)
{
GList *gIter = NULL;
int stop_index = -1;
int start_index = -1;
enum rsc_role_e req_role = RSC_ROLE_UNKNOWN;
const char *task = NULL;
const char *rsc_id = ID(lrm_resource);
pe_resource_t *rsc = NULL;
GList *op_list = NULL;
GList *sorted_op_list = NULL;
xmlNode *migrate_op = NULL;
xmlNode *rsc_op = NULL;
xmlNode *last_failure = NULL;
enum action_fail_response on_fail = action_fail_ignore;
enum rsc_role_e saved_role = RSC_ROLE_UNKNOWN;
if (rsc_id == NULL) {
crm_warn("Ignoring malformed " XML_LRM_TAG_RESOURCE
" entry without id");
return NULL;
}
crm_trace("Unpacking " XML_LRM_TAG_RESOURCE " for %s on %s",
rsc_id, node->details->uname);
// Build a list of individual lrm_rsc_op entries, so we can sort them
for (rsc_op = first_named_child(lrm_resource, XML_LRM_TAG_RSC_OP);
rsc_op != NULL; rsc_op = crm_next_same_xml(rsc_op)) {
op_list = g_list_prepend(op_list, rsc_op);
}
if (!pcmk_is_set(data_set->flags, pe_flag_shutdown_lock)) {
if (op_list == NULL) {
// If there are no operations, there is nothing to do
return NULL;
}
}
/* find the resource */
rsc = unpack_find_resource(data_set, node, rsc_id, lrm_resource);
if (rsc == NULL) {
if (op_list == NULL) {
// If there are no operations, there is nothing to do
return NULL;
} else {
rsc = process_orphan_resource(lrm_resource, node, data_set);
}
}
CRM_ASSERT(rsc != NULL);
// Check whether the resource is "shutdown-locked" to this node
if (pcmk_is_set(data_set->flags, pe_flag_shutdown_lock)) {
unpack_shutdown_lock(lrm_resource, rsc, node, data_set);
}
/* process operations */
saved_role = rsc->role;
rsc->role = RSC_ROLE_UNKNOWN;
sorted_op_list = g_list_sort(op_list, sort_op_by_callid);
for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) {
xmlNode *rsc_op = (xmlNode *) gIter->data;
task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK);
if (pcmk__str_eq(task, CRMD_ACTION_MIGRATED, pcmk__str_casei)) {
migrate_op = rsc_op;
}
unpack_rsc_op(rsc, node, rsc_op, &last_failure, &on_fail, data_set);
}
/* create active recurring operations as optional */
calculate_active_ops(sorted_op_list, &start_index, &stop_index);
process_recurring(node, rsc, start_index, stop_index, sorted_op_list, data_set);
/* no need to free the contents */
g_list_free(sorted_op_list);
process_rsc_state(rsc, node, on_fail, migrate_op, data_set);
if (get_target_role(rsc, &req_role)) {
if (rsc->next_role == RSC_ROLE_UNKNOWN || req_role < rsc->next_role) {
pe__set_next_role(rsc, req_role, XML_RSC_ATTR_TARGET_ROLE);
} else if (req_role > rsc->next_role) {
pe_rsc_info(rsc, "%s: Not overwriting calculated next role %s"
" with requested next role %s",
rsc->id, role2text(rsc->next_role), role2text(req_role));
}
}
if (saved_role > rsc->role) {
rsc->role = saved_role;
}
return rsc;
}
static void
handle_orphaned_container_fillers(xmlNode * lrm_rsc_list, pe_working_set_t * data_set)
{
xmlNode *rsc_entry = NULL;
for (rsc_entry = pcmk__xe_first_child(lrm_rsc_list); rsc_entry != NULL;
rsc_entry = pcmk__xe_next(rsc_entry)) {
pe_resource_t *rsc;
pe_resource_t *container;
const char *rsc_id;
const char *container_id;
if (!pcmk__str_eq((const char *)rsc_entry->name, XML_LRM_TAG_RESOURCE, pcmk__str_casei)) {
continue;
}
container_id = crm_element_value(rsc_entry, XML_RSC_ATTR_CONTAINER);
rsc_id = crm_element_value(rsc_entry, XML_ATTR_ID);
if (container_id == NULL || rsc_id == NULL) {
continue;
}
container = pe_find_resource(data_set->resources, container_id);
if (container == NULL) {
continue;
}
rsc = pe_find_resource(data_set->resources, rsc_id);
if (rsc == NULL ||
!pcmk_is_set(rsc->flags, pe_rsc_orphan_container_filler) ||
rsc->container != NULL) {
continue;
}
pe_rsc_trace(rsc, "Mapped container of orphaned resource %s to %s",
rsc->id, container_id);
rsc->container = container;
container->fillers = g_list_append(container->fillers, rsc);
}
}
/*!
* \internal
* \brief Unpack one node's lrm status section
*
* \param[in] node Node whose status is being unpacked
* \param[in] xml CIB node state XML
* \param[in] data_set Cluster working set
*/
static void
unpack_node_lrm(pe_node_t *node, xmlNode *xml, pe_working_set_t *data_set)
{
bool found_orphaned_container_filler = false;
// Drill down to lrm_resources section
xml = find_xml_node(xml, XML_CIB_TAG_LRM, FALSE);
if (xml == NULL) {
return;
}
xml = find_xml_node(xml, XML_LRM_TAG_RESOURCES, FALSE);
if (xml == NULL) {
return;
}
// Unpack each lrm_resource entry
for (xmlNode *rsc_entry = first_named_child(xml, XML_LRM_TAG_RESOURCE);
rsc_entry != NULL; rsc_entry = crm_next_same_xml(rsc_entry)) {
pe_resource_t *rsc = unpack_lrm_resource(node, rsc_entry, data_set);
if ((rsc != NULL)
&& pcmk_is_set(rsc->flags, pe_rsc_orphan_container_filler)) {
found_orphaned_container_filler = true;
}
}
/* Now that all resource state has been unpacked for this node, map any
* orphaned container fillers to their container resource.
*/
if (found_orphaned_container_filler) {
handle_orphaned_container_fillers(xml, data_set);
}
}
static void
set_active(pe_resource_t * rsc)
{
pe_resource_t *top = uber_parent(rsc);
if (top && pcmk_is_set(top->flags, pe_rsc_promotable)) {
rsc->role = RSC_ROLE_UNPROMOTED;
} else {
rsc->role = RSC_ROLE_STARTED;
}
}
static void
set_node_score(gpointer key, gpointer value, gpointer user_data)
{
pe_node_t *node = value;
int *score = user_data;
node->weight = *score;
}
#define STATUS_PATH_MAX 1024
static xmlNode *
find_lrm_op(const char *resource, const char *op, const char *node, const char *source,
bool success_only, pe_working_set_t *data_set)
{
int offset = 0;
char xpath[STATUS_PATH_MAX];
xmlNode *xml = NULL;
offset += snprintf(xpath + offset, STATUS_PATH_MAX - offset, "//node_state[@uname='%s']", node);
offset +=
snprintf(xpath + offset, STATUS_PATH_MAX - offset, "//" XML_LRM_TAG_RESOURCE "[@id='%s']",
resource);
/* Need to check against transition_magic too? */
if (source && pcmk__str_eq(op, CRMD_ACTION_MIGRATE, pcmk__str_casei)) {
offset +=
snprintf(xpath + offset, STATUS_PATH_MAX - offset,
"/" XML_LRM_TAG_RSC_OP "[@operation='%s' and @migrate_target='%s']", op,
source);
} else if (source && pcmk__str_eq(op, CRMD_ACTION_MIGRATED, pcmk__str_casei)) {
offset +=
snprintf(xpath + offset, STATUS_PATH_MAX - offset,
"/" XML_LRM_TAG_RSC_OP "[@operation='%s' and @migrate_source='%s']", op,
source);
} else {
offset +=
snprintf(xpath + offset, STATUS_PATH_MAX - offset,
"/" XML_LRM_TAG_RSC_OP "[@operation='%s']", op);
}
CRM_LOG_ASSERT(offset > 0);
xml = get_xpath_object(xpath, data_set->input, LOG_DEBUG);
if (xml && success_only) {
int rc = PCMK_OCF_UNKNOWN_ERROR;
int status = PCMK_LRM_OP_ERROR;
crm_element_value_int(xml, XML_LRM_ATTR_RC, &rc);
crm_element_value_int(xml, XML_LRM_ATTR_OPSTATUS, &status);
if ((rc != PCMK_OCF_OK) || (status != PCMK_LRM_OP_DONE)) {
return NULL;
}
}
return xml;
}
static int
pe__call_id(xmlNode *op_xml)
{
int id = 0;
if (op_xml) {
crm_element_value_int(op_xml, XML_LRM_ATTR_CALLID, &id);
}
return id;
}
/*!
* \brief Check whether a stop happened on the same node after some event
*
* \param[in] rsc Resource being checked
* \param[in] node Node being checked
* \param[in] xml_op Event that stop is being compared to
* \param[in] data_set Cluster working set
*
* \return TRUE if stop happened after event, FALSE otherwise
*
* \note This is really unnecessary, but kept as a safety mechanism. We
* currently don't save more than one successful event in history, so this
* only matters when processing really old CIB files that we don't
* technically support anymore, or as preparation for logging an extended
* history in the future.
*/
static bool
stop_happened_after(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op,
pe_working_set_t *data_set)
{
xmlNode *stop_op = find_lrm_op(rsc->id, CRMD_ACTION_STOP,
node->details->uname, NULL, TRUE, data_set);
return (stop_op && (pe__call_id(stop_op) > pe__call_id(xml_op)));
}
static void
unpack_migrate_to_success(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op,
pe_working_set_t *data_set)
{
/* A successful migration sequence is:
* migrate_to on source node
* migrate_from on target node
* stop on source node
*
* If a migrate_to is followed by a stop, the entire migration (successful
* or failed) is complete, and we don't care what happened on the target.
*
* If no migrate_from has happened, the migration is considered to be
* "partial". If the migrate_from failed, make sure the resource gets
* stopped on both source and target (if up).
*
* If the migrate_to and migrate_from both succeeded (which also implies the
* resource is no longer running on the source), but there is no stop, the
* migration is considered to be "dangling". Schedule a stop on the source
* in this case.
*/
int from_rc = 0;
int from_status = 0;
pe_node_t *target_node = NULL;
pe_node_t *source_node = NULL;
xmlNode *migrate_from = NULL;
const char *source = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_SOURCE);
const char *target = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_TARGET);
// Sanity check
CRM_CHECK(source && target && !strcmp(source, node->details->uname), return);
if (stop_happened_after(rsc, node, xml_op, data_set)) {
return;
}
// Clones are not allowed to migrate, so role can't be promoted
rsc->role = RSC_ROLE_STARTED;
target_node = pe_find_node(data_set->nodes, target);
source_node = pe_find_node(data_set->nodes, source);
// Check whether there was a migrate_from action on the target
migrate_from = find_lrm_op(rsc->id, CRMD_ACTION_MIGRATED, target,
source, FALSE, data_set);
if (migrate_from) {
crm_element_value_int(migrate_from, XML_LRM_ATTR_RC, &from_rc);
crm_element_value_int(migrate_from, XML_LRM_ATTR_OPSTATUS, &from_status);
pe_rsc_trace(rsc, "%s op on %s exited with status=%d, rc=%d",
ID(migrate_from), target, from_status, from_rc);
}
if (migrate_from && from_rc == PCMK_OCF_OK
&& from_status == PCMK_LRM_OP_DONE) {
/* The migrate_to and migrate_from both succeeded, so mark the migration
* as "dangling". This will be used to schedule a stop action on the
* source without affecting the target.
*/
pe_rsc_trace(rsc, "Detected dangling migration op: %s on %s", ID(xml_op),
source);
rsc->role = RSC_ROLE_STOPPED;
rsc->dangling_migrations = g_list_prepend(rsc->dangling_migrations, node);
} else if (migrate_from && (from_status != PCMK_LRM_OP_PENDING)) { // Failed
if (target_node && target_node->details->online) {
pe_rsc_trace(rsc, "Marking active on %s %p %d", target, target_node,
target_node->details->online);
- native_add_running(rsc, target_node, data_set);
+ native_add_running(rsc, target_node, data_set, TRUE);
}
} else { // Pending, or complete but erased
if (target_node && target_node->details->online) {
pe_rsc_trace(rsc, "Marking active on %s %p %d", target, target_node,
target_node->details->online);
- native_add_running(rsc, target_node, data_set);
+ native_add_running(rsc, target_node, data_set, FALSE);
if (source_node && source_node->details->online) {
/* This is a partial migration: the migrate_to completed
* successfully on the source, but the migrate_from has not
* completed. Remember the source and target; if the newly
* chosen target remains the same when we schedule actions
* later, we may continue with the migration.
*/
rsc->partial_migration_target = target_node;
rsc->partial_migration_source = source_node;
}
} else {
/* Consider it failed here - forces a restart, prevents migration */
pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop);
pe__clear_resource_flags(rsc, pe_rsc_allow_migrate);
}
}
}
// Is there an action_name in node_name's rsc history newer than call_id?
static bool
newer_op(pe_resource_t *rsc, const char *action_name, const char *node_name,
int call_id, pe_working_set_t *data_set)
{
xmlNode *action = find_lrm_op(rsc->id, action_name, node_name, NULL, TRUE,
data_set);
return pe__call_id(action) > call_id;
}
static void
unpack_migrate_to_failure(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op,
pe_working_set_t *data_set)
{
int target_stop_id = 0;
int target_migrate_from_id = 0;
xmlNode *target_stop = NULL;
xmlNode *target_migrate_from = NULL;
const char *source = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_SOURCE);
const char *target = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_TARGET);
// Sanity check
CRM_CHECK(source && target && !strcmp(source, node->details->uname), return);
/* If a migration failed, we have to assume the resource is active. Clones
* are not allowed to migrate, so role can't be promoted.
*/
rsc->role = RSC_ROLE_STARTED;
// Check for stop on the target
target_stop = find_lrm_op(rsc->id, CRMD_ACTION_STOP, target, NULL,
TRUE, data_set);
target_stop_id = pe__call_id(target_stop);
// Check for migrate_from on the target
target_migrate_from = find_lrm_op(rsc->id, CRMD_ACTION_MIGRATED, target,
source, TRUE, data_set);
target_migrate_from_id = pe__call_id(target_migrate_from);
if ((target_stop == NULL) || (target_stop_id < target_migrate_from_id)) {
/* There was no stop on the target, or a stop that happened before a
* migrate_from, so assume the resource is still active on the target
* (if it is up).
*/
pe_node_t *target_node = pe_find_node(data_set->nodes, target);
pe_rsc_trace(rsc, "stop (%d) + migrate_from (%d)",
target_stop_id, target_migrate_from_id);
if (target_node && target_node->details->online) {
- native_add_running(rsc, target_node, data_set);
+ native_add_running(rsc, target_node, data_set, FALSE);
}
} else if (target_migrate_from == NULL) {
/* We know there was a stop on the target, but there may not have been a
* migrate_from (the stop could have happened before migrate_from was
* scheduled or attempted).
*
* That means this could be a "dangling" migration. But first, check
* whether there is a newer successful stop, start, or migrate_from on
* the source node -- it's possible the failed migration was followed by
* a successful stop, full restart, or migration in the reverse
* direction, in which case we don't want to force a stop.
*/
int source_migrate_to_id = pe__call_id(xml_op);
if (newer_op(rsc, CRMD_ACTION_MIGRATED, source, source_migrate_to_id,
data_set)
|| newer_op(rsc, CRMD_ACTION_START, source, source_migrate_to_id,
data_set)
|| newer_op(rsc, CRMD_ACTION_STOP, source, source_migrate_to_id,
data_set)) {
return;
}
// Mark node as having dangling migration so we can force a stop later
rsc->dangling_migrations = g_list_prepend(rsc->dangling_migrations, node);
}
}
static void
unpack_migrate_from_failure(pe_resource_t *rsc, pe_node_t *node,
xmlNode *xml_op, pe_working_set_t *data_set)
{
xmlNode *source_stop = NULL;
xmlNode *source_migrate_to = NULL;
const char *source = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_SOURCE);
const char *target = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_TARGET);
// Sanity check
CRM_CHECK(source && target && !strcmp(target, node->details->uname), return);
/* If a migration failed, we have to assume the resource is active. Clones
* are not allowed to migrate, so role can't be promoted.
*/
rsc->role = RSC_ROLE_STARTED;
// Check for a stop on the source
source_stop = find_lrm_op(rsc->id, CRMD_ACTION_STOP, source, NULL,
TRUE, data_set);
// Check for a migrate_to on the source
source_migrate_to = find_lrm_op(rsc->id, CRMD_ACTION_MIGRATE,
source, target, TRUE, data_set);
if ((source_stop == NULL)
|| (pe__call_id(source_stop) < pe__call_id(source_migrate_to))) {
/* There was no stop on the source, or a stop that happened before
* migrate_to, so assume the resource is still active on the source (if
* it is up).
*/
pe_node_t *source_node = pe_find_node(data_set->nodes, source);
if (source_node && source_node->details->online) {
- native_add_running(rsc, source_node, data_set);
+ native_add_running(rsc, source_node, data_set, TRUE);
}
}
}
static void
record_failed_op(xmlNode *op, const pe_node_t *node,
const pe_resource_t *rsc, pe_working_set_t *data_set)
{
xmlNode *xIter = NULL;
const char *op_key = crm_element_value(op, XML_LRM_ATTR_TASK_KEY);
if (node->details->online == FALSE) {
return;
}
for (xIter = data_set->failed->children; xIter; xIter = xIter->next) {
const char *key = crm_element_value(xIter, XML_LRM_ATTR_TASK_KEY);
const char *uname = crm_element_value(xIter, XML_ATTR_UNAME);
if(pcmk__str_eq(op_key, key, pcmk__str_casei) && pcmk__str_eq(uname, node->details->uname, pcmk__str_casei)) {
crm_trace("Skipping duplicate entry %s on %s", op_key, node->details->uname);
return;
}
}
crm_trace("Adding entry %s on %s", op_key, node->details->uname);
crm_xml_add(op, XML_ATTR_UNAME, node->details->uname);
crm_xml_add(op, XML_LRM_ATTR_RSCID, rsc->id);
add_node_copy(data_set->failed, op);
}
static const char *get_op_key(xmlNode *xml_op)
{
const char *key = crm_element_value(xml_op, XML_LRM_ATTR_TASK_KEY);
if(key == NULL) {
key = ID(xml_op);
}
return key;
}
static const char *
last_change_str(xmlNode *xml_op)
{
time_t when;
const char *when_s = NULL;
if (crm_element_value_epoch(xml_op, XML_RSC_OP_LAST_CHANGE,
&when) == pcmk_ok) {
when_s = pcmk__epoch2str(&when);
if (when_s) {
// Skip day of week to make message shorter
when_s = strchr(when_s, ' ');
if (when_s) {
++when_s;
}
}
}
return ((when_s && *when_s)? when_s : "unknown time");
}
/*!
* \internal
* \brief Compare two on-fail values
*
* \param[in] first One on-fail value to compare
* \param[in] second The other on-fail value to compare
*
* \return A negative number if second is more severe than first, zero if they
* are equal, or a positive number if first is more severe than second.
* \note This is only needed until the action_fail_response values can be
* renumbered at the next API compatibility break.
*/
static int
cmp_on_fail(enum action_fail_response first, enum action_fail_response second)
{
switch (first) {
case action_fail_demote:
switch (second) {
case action_fail_ignore:
return 1;
case action_fail_demote:
return 0;
default:
return -1;
}
break;
case action_fail_reset_remote:
switch (second) {
case action_fail_ignore:
case action_fail_demote:
case action_fail_recover:
return 1;
case action_fail_reset_remote:
return 0;
default:
return -1;
}
break;
case action_fail_restart_container:
switch (second) {
case action_fail_ignore:
case action_fail_demote:
case action_fail_recover:
case action_fail_reset_remote:
return 1;
case action_fail_restart_container:
return 0;
default:
return -1;
}
break;
default:
break;
}
switch (second) {
case action_fail_demote:
return (first == action_fail_ignore)? -1 : 1;
case action_fail_reset_remote:
switch (first) {
case action_fail_ignore:
case action_fail_demote:
case action_fail_recover:
return -1;
default:
return 1;
}
break;
case action_fail_restart_container:
switch (first) {
case action_fail_ignore:
case action_fail_demote:
case action_fail_recover:
case action_fail_reset_remote:
return -1;
default:
return 1;
}
break;
default:
break;
}
return first - second;
}
static void
unpack_rsc_op_failure(pe_resource_t * rsc, pe_node_t * node, int rc, xmlNode * xml_op, xmlNode ** last_failure,
enum action_fail_response * on_fail, pe_working_set_t * data_set)
{
guint interval_ms = 0;
bool is_probe = false;
pe_action_t *action = NULL;
const char *key = get_op_key(xml_op);
const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK);
const char *exit_reason = crm_element_value(xml_op,
XML_LRM_ATTR_EXIT_REASON);
CRM_ASSERT(rsc);
CRM_CHECK(task != NULL, return);
*last_failure = xml_op;
crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms);
if ((interval_ms == 0) && !strcmp(task, CRMD_ACTION_STATUS)) {
is_probe = true;
}
if (exit_reason == NULL) {
exit_reason = "";
}
if (!pcmk_is_set(data_set->flags, pe_flag_symmetric_cluster)
&& (rc == PCMK_OCF_NOT_INSTALLED)) {
crm_trace("Unexpected result (%s%s%s) was recorded for "
"%s of %s on %s at %s " CRM_XS " rc=%d id=%s",
services_ocf_exitcode_str(rc),
(*exit_reason? ": " : ""), exit_reason,
(is_probe? "probe" : task), rsc->id, node->details->uname,
last_change_str(xml_op), rc, ID(xml_op));
} else {
crm_warn("Unexpected result (%s%s%s) was recorded for "
"%s of %s on %s at %s " CRM_XS " rc=%d id=%s",
services_ocf_exitcode_str(rc),
(*exit_reason? ": " : ""), exit_reason,
(is_probe? "probe" : task), rsc->id, node->details->uname,
last_change_str(xml_op), rc, ID(xml_op));
if (is_probe && (rc != PCMK_OCF_OK)
&& (rc != PCMK_OCF_NOT_RUNNING)
&& (rc != PCMK_OCF_RUNNING_PROMOTED)) {
/* A failed (not just unexpected) probe result could mean the user
* didn't know resources will be probed even where they can't run.
*/
crm_notice("If it is not possible for %s to run on %s, see "
"the resource-discovery option for location constraints",
rsc->id, node->details->uname);
}
record_failed_op(xml_op, node, rsc, data_set);
}
action = custom_action(rsc, strdup(key), task, NULL, TRUE, FALSE, data_set);
if (cmp_on_fail(*on_fail, action->on_fail) < 0) {
pe_rsc_trace(rsc, "on-fail %s -> %s for %s (%s)", fail2text(*on_fail),
fail2text(action->on_fail), action->uuid, key);
*on_fail = action->on_fail;
}
if (!strcmp(task, CRMD_ACTION_STOP)) {
resource_location(rsc, node, -INFINITY, "__stop_fail__", data_set);
} else if (!strcmp(task, CRMD_ACTION_MIGRATE)) {
unpack_migrate_to_failure(rsc, node, xml_op, data_set);
} else if (!strcmp(task, CRMD_ACTION_MIGRATED)) {
unpack_migrate_from_failure(rsc, node, xml_op, data_set);
} else if (!strcmp(task, CRMD_ACTION_PROMOTE)) {
rsc->role = RSC_ROLE_PROMOTED;
} else if (!strcmp(task, CRMD_ACTION_DEMOTE)) {
if (action->on_fail == action_fail_block) {
rsc->role = RSC_ROLE_PROMOTED;
pe__set_next_role(rsc, RSC_ROLE_STOPPED,
"demote with on-fail=block");
} else if(rc == PCMK_OCF_NOT_RUNNING) {
rsc->role = RSC_ROLE_STOPPED;
} else {
/* Staying in the promoted role would put the scheduler and
* controller into a loop. Setting the role to unpromoted is not
* dangerous because the resource will be stopped as part of
* recovery, and any promotion will be ordered after that stop.
*/
rsc->role = RSC_ROLE_UNPROMOTED;
}
}
if(is_probe && rc == PCMK_OCF_NOT_INSTALLED) {
/* leave stopped */
pe_rsc_trace(rsc, "Leaving %s stopped", rsc->id);
rsc->role = RSC_ROLE_STOPPED;
} else if (rsc->role < RSC_ROLE_STARTED) {
pe_rsc_trace(rsc, "Setting %s active", rsc->id);
set_active(rsc);
}
pe_rsc_trace(rsc, "Resource %s: role=%s, unclean=%s, on_fail=%s, fail_role=%s",
rsc->id, role2text(rsc->role),
pcmk__btoa(node->details->unclean),
fail2text(action->on_fail), role2text(action->fail_role));
if (action->fail_role != RSC_ROLE_STARTED && rsc->next_role < action->fail_role) {
pe__set_next_role(rsc, action->fail_role, "failure");
}
if (action->fail_role == RSC_ROLE_STOPPED) {
int score = -INFINITY;
pe_resource_t *fail_rsc = rsc;
if (fail_rsc->parent) {
pe_resource_t *parent = uber_parent(fail_rsc);
if (pe_rsc_is_clone(parent)
&& !pcmk_is_set(parent->flags, pe_rsc_unique)) {
/* For clone resources, if a child fails on an operation
* with on-fail = stop, all the resources fail. Do this by preventing
* the parent from coming up again. */
fail_rsc = parent;
}
}
crm_notice("%s will not be started under current conditions",
fail_rsc->id);
/* make sure it doesn't come up again */
if (fail_rsc->allowed_nodes != NULL) {
g_hash_table_destroy(fail_rsc->allowed_nodes);
}
fail_rsc->allowed_nodes = pe__node_list2table(data_set->nodes);
g_hash_table_foreach(fail_rsc->allowed_nodes, set_node_score, &score);
}
pe_free_action(action);
}
/*!
* \internal
* \brief Remap operation status based on action result
*
* Given an action result, determine an appropriate operation status for the
* purposes of responding to the action (the status provided by the executor is
* not directly usable since the executor does not know what was expected).
*
* \param[in,out] rsc Resource that operation history entry is for
* \param[in] rc Actual return code of operation
* \param[in] target_rc Expected return code of operation
* \param[in] node Node where operation was executed
* \param[in] xml_op Operation history entry XML from CIB status
* \param[in,out] on_fail What should be done about the result
* \param[in] data_set Current cluster working set
*
* \return Operation status based on return code and action info
* \note This may update the resource's current and next role.
*/
static int
determine_op_status(
pe_resource_t *rsc, int rc, int target_rc, pe_node_t * node, xmlNode * xml_op, enum action_fail_response * on_fail, pe_working_set_t * data_set)
{
guint interval_ms = 0;
bool is_probe = false;
int result = PCMK_LRM_OP_DONE;
const char *key = get_op_key(xml_op);
const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK);
const char *exit_reason = crm_element_value(xml_op,
XML_LRM_ATTR_EXIT_REASON);
CRM_ASSERT(rsc);
CRM_CHECK(task != NULL, return PCMK_LRM_OP_ERROR);
if (exit_reason == NULL) {
exit_reason = "";
}
crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms);
if ((interval_ms == 0) && !strcmp(task, CRMD_ACTION_STATUS)) {
is_probe = true;
task = "probe";
}
if (target_rc < 0) {
/* Pre-1.0 Pacemaker versions, and Pacemaker 1.1.6 or earlier with
* Heartbeat 2.0.7 or earlier as the cluster layer, did not include the
* target_rc in the transition key, which (along with the similar case
* of a corrupted transition key in the CIB) will be reported to this
* function as -1. Pacemaker 2.0+ does not support rolling upgrades from
* those versions or processing of saved CIB files from those versions,
* so we do not need to care much about this case.
*/
result = PCMK_LRM_OP_ERROR;
crm_warn("Expected result not found for %s on %s (corrupt or obsolete CIB?)",
key, node->details->uname);
} else if (target_rc != rc) {
result = PCMK_LRM_OP_ERROR;
pe_rsc_debug(rsc, "%s on %s: expected %d (%s), got %d (%s%s%s)",
key, node->details->uname,
target_rc, services_ocf_exitcode_str(target_rc),
rc, services_ocf_exitcode_str(rc),
(*exit_reason? ": " : ""), exit_reason);
}
switch (rc) {
case PCMK_OCF_OK:
if (is_probe && (target_rc == PCMK_OCF_NOT_RUNNING)) {
result = PCMK_LRM_OP_DONE;
pe_rsc_info(rsc, "Probe found %s active on %s at %s",
rsc->id, node->details->uname,
last_change_str(xml_op));
}
break;
case PCMK_OCF_NOT_RUNNING:
if (is_probe || (target_rc == rc)
|| !pcmk_is_set(rsc->flags, pe_rsc_managed)) {
result = PCMK_LRM_OP_DONE;
rsc->role = RSC_ROLE_STOPPED;
/* clear any previous failure actions */
*on_fail = action_fail_ignore;
pe__set_next_role(rsc, RSC_ROLE_UNKNOWN, "not running");
}
break;
case PCMK_OCF_RUNNING_PROMOTED:
if (is_probe && (rc != target_rc)) {
result = PCMK_LRM_OP_DONE;
pe_rsc_info(rsc,
"Probe found %s active and promoted on %s at %s",
rsc->id, node->details->uname,
last_change_str(xml_op));
}
rsc->role = RSC_ROLE_PROMOTED;
break;
case PCMK_OCF_DEGRADED_PROMOTED:
case PCMK_OCF_FAILED_PROMOTED:
rsc->role = RSC_ROLE_PROMOTED;
result = PCMK_LRM_OP_ERROR;
break;
case PCMK_OCF_NOT_CONFIGURED:
result = PCMK_LRM_OP_ERROR_FATAL;
break;
case PCMK_OCF_UNIMPLEMENT_FEATURE:
if (interval_ms > 0) {
result = PCMK_LRM_OP_NOTSUPPORTED;
break;
}
// fall through
case PCMK_OCF_NOT_INSTALLED:
case PCMK_OCF_INVALID_PARAM:
case PCMK_OCF_INSUFFICIENT_PRIV:
if (!pe_can_fence(data_set, node)
&& !strcmp(task, CRMD_ACTION_STOP)) {
/* If a stop fails and we can't fence, there's nothing else we can do */
pe_proc_err("No further recovery can be attempted for %s "
"because %s on %s failed (%s%s%s) at %s "
CRM_XS " rc=%d id=%s", rsc->id, task,
node->details->uname, services_ocf_exitcode_str(rc),
(*exit_reason? ": " : ""), exit_reason,
last_change_str(xml_op), rc, ID(xml_op));
pe__clear_resource_flags(rsc, pe_rsc_managed);
pe__set_resource_flags(rsc, pe_rsc_block);
}
result = PCMK_LRM_OP_ERROR_HARD;
break;
default:
if (result == PCMK_LRM_OP_DONE) {
crm_info("Treating unknown exit status %d from %s of %s "
"on %s at %s as failure",
rc, task, rsc->id, node->details->uname,
last_change_str(xml_op));
result = PCMK_LRM_OP_ERROR;
}
break;
}
return result;
}
// return TRUE if start or monitor last failure but parameters changed
static bool
should_clear_for_param_change(xmlNode *xml_op, const char *task,
pe_resource_t *rsc, pe_node_t *node,
pe_working_set_t *data_set)
{
if (!strcmp(task, "start") || !strcmp(task, "monitor")) {
if (pe__bundle_needs_remote_name(rsc, data_set)) {
/* We haven't allocated resources yet, so we can't reliably
* substitute addr parameters for the REMOTE_CONTAINER_HACK.
* When that's needed, defer the check until later.
*/
pe__add_param_check(xml_op, rsc, node, pe_check_last_failure,
data_set);
} else {
op_digest_cache_t *digest_data = NULL;
digest_data = rsc_action_digest_cmp(rsc, xml_op, node, data_set);
switch (digest_data->rc) {
case RSC_DIGEST_UNKNOWN:
crm_trace("Resource %s history entry %s on %s"
" has no digest to compare",
rsc->id, get_op_key(xml_op), node->details->id);
break;
case RSC_DIGEST_MATCH:
break;
default:
return TRUE;
}
}
}
return FALSE;
}
// Order action after fencing of remote node, given connection rsc
static void
order_after_remote_fencing(pe_action_t *action, pe_resource_t *remote_conn,
pe_working_set_t *data_set)
{
pe_node_t *remote_node = pe_find_node(data_set->nodes, remote_conn->id);
if (remote_node) {
pe_action_t *fence = pe_fence_op(remote_node, NULL, TRUE, NULL,
FALSE, data_set);
order_actions(fence, action, pe_order_implies_then);
}
}
static bool
should_ignore_failure_timeout(pe_resource_t *rsc, xmlNode *xml_op,
const char *task, guint interval_ms,
bool is_last_failure, pe_working_set_t *data_set)
{
/* Clearing failures of recurring monitors has special concerns. The
* executor reports only changes in the monitor result, so if the
* monitor is still active and still getting the same failure result,
* that will go undetected after the failure is cleared.
*
* Also, the operation history will have the time when the recurring
* monitor result changed to the given code, not the time when the
* result last happened.
*
* @TODO We probably should clear such failures only when the failure
* timeout has passed since the last occurrence of the failed result.
* However we don't record that information. We could maybe approximate
* that by clearing only if there is a more recent successful monitor or
* stop result, but we don't even have that information at this point
* since we are still unpacking the resource's operation history.
*
* This is especially important for remote connection resources with a
* reconnect interval, so in that case, we skip clearing failures
* if the remote node hasn't been fenced.
*/
if (rsc->remote_reconnect_ms
&& pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)
&& (interval_ms != 0) && pcmk__str_eq(task, CRMD_ACTION_STATUS, pcmk__str_casei)) {
pe_node_t *remote_node = pe_find_node(data_set->nodes, rsc->id);
if (remote_node && !remote_node->details->remote_was_fenced) {
if (is_last_failure) {
crm_info("Waiting to clear monitor failure for remote node %s"
" until fencing has occurred", rsc->id);
}
return TRUE;
}
}
return FALSE;
}
/*!
* \internal
* \brief Check operation age and schedule failure clearing when appropriate
*
* This function has two distinct purposes. The first is to check whether an
* operation history entry is expired (i.e. the resource has a failure timeout,
* the entry is older than the timeout, and the resource either has no fail
* count or its fail count is entirely older than the timeout). The second is to
* schedule fail count clearing when appropriate (i.e. the operation is expired
* and either the resource has an expired fail count or the operation is a
* last_failure for a remote connection resource with a reconnect interval,
* or the operation is a last_failure for a start or monitor operation and the
* resource's parameters have changed since the operation).
*
* \param[in] rsc Resource that operation happened to
* \param[in] node Node that operation happened on
* \param[in] rc Actual result of operation
* \param[in] xml_op Operation history entry XML
* \param[in] data_set Current working set
*
* \return TRUE if operation history entry is expired, FALSE otherwise
*/
static bool
check_operation_expiry(pe_resource_t *rsc, pe_node_t *node, int rc,
xmlNode *xml_op, pe_working_set_t *data_set)
{
bool expired = FALSE;
bool is_last_failure = pcmk__ends_with(ID(xml_op), "_last_failure_0");
time_t last_run = 0;
guint interval_ms = 0;
int unexpired_fail_count = 0;
const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK);
const char *clear_reason = NULL;
crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms);
if ((rsc->failure_timeout > 0)
&& (crm_element_value_epoch(xml_op, XML_RSC_OP_LAST_CHANGE,
&last_run) == 0)) {
// Resource has a failure-timeout, and history entry has a timestamp
time_t now = get_effective_time(data_set);
time_t last_failure = 0;
// Is this particular operation history older than the failure timeout?
if ((now >= (last_run + rsc->failure_timeout))
&& !should_ignore_failure_timeout(rsc, xml_op, task, interval_ms,
is_last_failure, data_set)) {
expired = TRUE;
}
// Does the resource as a whole have an unexpired fail count?
unexpired_fail_count = pe_get_failcount(node, rsc, &last_failure,
pe_fc_effective, xml_op,
data_set);
// Update scheduler recheck time according to *last* failure
crm_trace("%s@%lld is %sexpired @%lld with unexpired_failures=%d timeout=%ds"
" last-failure@%lld",
ID(xml_op), (long long) last_run, (expired? "" : "not "),
(long long) now, unexpired_fail_count, rsc->failure_timeout,
(long long) last_failure);
last_failure += rsc->failure_timeout + 1;
if (unexpired_fail_count && (now < last_failure)) {
pe__update_recheck_time(last_failure, data_set);
}
}
if (expired) {
if (pe_get_failcount(node, rsc, NULL, pe_fc_default, xml_op, data_set)) {
// There is a fail count ignoring timeout
if (unexpired_fail_count == 0) {
// There is no fail count considering timeout
clear_reason = "it expired";
} else {
/* This operation is old, but there is an unexpired fail count.
* In a properly functioning cluster, this should only be
* possible if this operation is not a failure (otherwise the
* fail count should be expired too), so this is really just a
* failsafe.
*/
expired = FALSE;
}
} else if (is_last_failure && rsc->remote_reconnect_ms) {
/* Clear any expired last failure when reconnect interval is set,
* even if there is no fail count.
*/
clear_reason = "reconnect interval is set";
}
}
if (!expired && is_last_failure
&& should_clear_for_param_change(xml_op, task, rsc, node, data_set)) {
clear_reason = "resource parameters have changed";
}
if (clear_reason != NULL) {
// Schedule clearing of the fail count
pe_action_t *clear_op = pe__clear_failcount(rsc, node, clear_reason,
data_set);
if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)
&& rsc->remote_reconnect_ms) {
/* If we're clearing a remote connection due to a reconnect
* interval, we want to wait until any scheduled fencing
* completes.
*
* We could limit this to remote_node->details->unclean, but at
* this point, that's always true (it won't be reliable until
* after unpack_node_history() is done).
*/
crm_info("Clearing %s failure will wait until any scheduled "
"fencing of %s completes", task, rsc->id);
order_after_remote_fencing(clear_op, rsc, data_set);
}
}
if (expired && (interval_ms == 0) && pcmk__str_eq(task, CRMD_ACTION_STATUS, pcmk__str_casei)) {
switch(rc) {
case PCMK_OCF_OK:
case PCMK_OCF_NOT_RUNNING:
case PCMK_OCF_RUNNING_PROMOTED:
case PCMK_OCF_DEGRADED:
case PCMK_OCF_DEGRADED_PROMOTED:
// Don't expire probes that return these values
expired = FALSE;
break;
}
}
return expired;
}
int pe__target_rc_from_xml(xmlNode *xml_op)
{
int target_rc = 0;
const char *key = crm_element_value(xml_op, XML_ATTR_TRANSITION_KEY);
if (key == NULL) {
return -1;
}
decode_transition_key(key, NULL, NULL, NULL, &target_rc);
return target_rc;
}
static enum action_fail_response
get_action_on_fail(pe_resource_t *rsc, const char *key, const char *task, pe_working_set_t * data_set)
{
enum action_fail_response result = action_fail_recover;
pe_action_t *action = custom_action(rsc, strdup(key), task, NULL, TRUE, FALSE, data_set);
result = action->on_fail;
pe_free_action(action);
return result;
}
static void
update_resource_state(pe_resource_t * rsc, pe_node_t * node, xmlNode * xml_op, const char * task, int rc,
xmlNode * last_failure, enum action_fail_response * on_fail, pe_working_set_t * data_set)
{
gboolean clear_past_failure = FALSE;
CRM_ASSERT(rsc);
CRM_ASSERT(xml_op);
if (rc == PCMK_OCF_NOT_RUNNING) {
clear_past_failure = TRUE;
} else if (rc == PCMK_OCF_NOT_INSTALLED) {
rsc->role = RSC_ROLE_STOPPED;
} else if (pcmk__str_eq(task, CRMD_ACTION_STATUS, pcmk__str_casei)) {
if (last_failure) {
const char *op_key = get_op_key(xml_op);
const char *last_failure_key = get_op_key(last_failure);
if (pcmk__str_eq(op_key, last_failure_key, pcmk__str_casei)) {
clear_past_failure = TRUE;
}
}
if (rsc->role < RSC_ROLE_STARTED) {
set_active(rsc);
}
} else if (pcmk__str_eq(task, CRMD_ACTION_START, pcmk__str_casei)) {
rsc->role = RSC_ROLE_STARTED;
clear_past_failure = TRUE;
} else if (pcmk__str_eq(task, CRMD_ACTION_STOP, pcmk__str_casei)) {
rsc->role = RSC_ROLE_STOPPED;
clear_past_failure = TRUE;
} else if (pcmk__str_eq(task, CRMD_ACTION_PROMOTE, pcmk__str_casei)) {
rsc->role = RSC_ROLE_PROMOTED;
clear_past_failure = TRUE;
} else if (pcmk__str_eq(task, CRMD_ACTION_DEMOTE, pcmk__str_casei)) {
if (*on_fail == action_fail_demote) {
// Demote clears an error only if on-fail=demote
clear_past_failure = TRUE;
}
rsc->role = RSC_ROLE_UNPROMOTED;
} else if (pcmk__str_eq(task, CRMD_ACTION_MIGRATED, pcmk__str_casei)) {
rsc->role = RSC_ROLE_STARTED;
clear_past_failure = TRUE;
} else if (pcmk__str_eq(task, CRMD_ACTION_MIGRATE, pcmk__str_casei)) {
unpack_migrate_to_success(rsc, node, xml_op, data_set);
} else if (rsc->role < RSC_ROLE_STARTED) {
pe_rsc_trace(rsc, "%s active on %s", rsc->id, node->details->uname);
set_active(rsc);
}
/* clear any previous failure actions */
if (clear_past_failure) {
switch (*on_fail) {
case action_fail_stop:
case action_fail_fence:
case action_fail_migrate:
case action_fail_standby:
pe_rsc_trace(rsc, "%s.%s is not cleared by a completed stop",
rsc->id, fail2text(*on_fail));
break;
case action_fail_block:
case action_fail_ignore:
case action_fail_demote:
case action_fail_recover:
case action_fail_restart_container:
*on_fail = action_fail_ignore;
pe__set_next_role(rsc, RSC_ROLE_UNKNOWN, "clear past failures");
break;
case action_fail_reset_remote:
if (rsc->remote_reconnect_ms == 0) {
/* With no reconnect interval, the connection is allowed to
* start again after the remote node is fenced and
* completely stopped. (With a reconnect interval, we wait
* for the failure to be cleared entirely before attempting
* to reconnect.)
*/
*on_fail = action_fail_ignore;
pe__set_next_role(rsc, RSC_ROLE_UNKNOWN,
"clear past failures and reset remote");
}
break;
}
}
}
/*!
* \internal
* \brief Remap informational monitor results to usual values
*
* Certain OCF result codes are for providing extended information to the
* user about services that aren't yet failed but not entirely healthy either.
* These must be treated as the "normal" result by pacemaker.
*
* \param[in] rc Actual result of a monitor action
* \param[in] xml_op Operation history XML
* \param[in] node Node that operation happened on
* \param[in] rsc Resource that operation happened to
* \param[in] data_set Cluster working set
*
* \return Result code that pacemaker should use
*
* \note If the result is remapped, and the node is not shutting down or failed,
* the operation will be recorded in the data set's list of failed
* operations, to highlight it for the user.
*/
static int
remap_monitor_rc(int rc, xmlNode *xml_op, const pe_node_t *node,
const pe_resource_t *rsc, pe_working_set_t *data_set)
{
int remapped_rc = pcmk__effective_rc(rc);
if (rc != remapped_rc) {
crm_trace("Remapping monitor result %d to %d", rc, remapped_rc);
if (!node->details->shutdown || node->details->online) {
record_failed_op(xml_op, node, rsc, data_set);
}
}
return remapped_rc;
}
static void
unpack_rsc_op(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op,
xmlNode **last_failure, enum action_fail_response *on_fail,
pe_working_set_t *data_set)
{
int rc = 0;
int task_id = 0;
int target_rc = 0;
int status = PCMK_LRM_OP_UNKNOWN;
guint interval_ms = 0;
const char *task = NULL;
const char *task_key = NULL;
const char *exit_reason = NULL;
bool expired = FALSE;
pe_resource_t *parent = rsc;
enum action_fail_response failure_strategy = action_fail_recover;
CRM_CHECK(rsc && node && xml_op, return);
target_rc = pe__target_rc_from_xml(xml_op);
task_key = get_op_key(xml_op);
task = crm_element_value(xml_op, XML_LRM_ATTR_TASK);
exit_reason = crm_element_value(xml_op, XML_LRM_ATTR_EXIT_REASON);
if (exit_reason == NULL) {
exit_reason = "";
}
crm_element_value_int(xml_op, XML_LRM_ATTR_RC, &rc);
crm_element_value_int(xml_op, XML_LRM_ATTR_CALLID, &task_id);
crm_element_value_int(xml_op, XML_LRM_ATTR_OPSTATUS, &status);
crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms);
CRM_CHECK(task != NULL, return);
CRM_CHECK(status <= PCMK_LRM_OP_INVALID, return);
CRM_CHECK(status >= PCMK_LRM_OP_PENDING, return);
if (!strcmp(task, CRMD_ACTION_NOTIFY) ||
!strcmp(task, CRMD_ACTION_METADATA)) {
/* safe to ignore these */
return;
}
if (!pcmk_is_set(rsc->flags, pe_rsc_unique)) {
parent = uber_parent(rsc);
}
pe_rsc_trace(rsc, "Unpacking task %s/%s (call_id=%d, status=%d, rc=%d) on %s (role=%s)",
task_key, task, task_id, status, rc, node->details->uname, role2text(rsc->role));
if (node->details->unclean) {
pe_rsc_trace(rsc, "Node %s (where %s is running) is unclean."
" Further action depends on the value of the stop's on-fail attribute",
node->details->uname, rsc->id);
}
/* It should be possible to call remap_monitor_rc() first then call
* check_operation_expiry() only if rc != target_rc, because there should
* never be a fail count without at least one unexpected result in the
* resource history. That would be more efficient by avoiding having to call
* check_operation_expiry() for expected results.
*
* However, we do have such configurations in the scheduler regression
* tests, even if it shouldn't be possible with the current code. It's
* probably a good idea anyway, but that would require updating the test
* inputs to something currently possible.
*/
if ((status != PCMK_LRM_OP_NOT_INSTALLED)
&& check_operation_expiry(rsc, node, rc, xml_op, data_set)) {
expired = TRUE;
}
if (!strcmp(task, CRMD_ACTION_STATUS)) {
rc = remap_monitor_rc(rc, xml_op, node, rsc, data_set);
}
if (expired && (rc != target_rc)) {
const char *magic = crm_element_value(xml_op, XML_ATTR_TRANSITION_MAGIC);
if (interval_ms == 0) {
crm_notice("Ignoring expired %s failure on %s "
CRM_XS " actual=%d expected=%d magic=%s",
task_key, node->details->uname, rc, target_rc, magic);
goto done;
} else if(node->details->online && node->details->unclean == FALSE) {
/* Reschedule the recurring monitor. CancelXmlOp() won't work at
* this stage, so as a hacky workaround, forcibly change the restart
* digest so check_action_definition() does what we want later.
*
* @TODO We should skip this if there is a newer successful monitor.
* Also, this causes rescheduling only if the history entry
* has an op-digest (which the expire-non-blocked-failure
* scheduler regression test doesn't, but that may not be a
* realistic scenario in production).
*/
crm_notice("Rescheduling %s after failure expired on %s "
CRM_XS " actual=%d expected=%d magic=%s",
task_key, node->details->uname, rc, target_rc, magic);
crm_xml_add(xml_op, XML_LRM_ATTR_RESTART_DIGEST, "calculated-failure-timeout");
goto done;
}
}
/* If the executor reported an operation status of anything but done or
* error, consider that final. But for done or error, we know better whether
* it should be treated as a failure or not, because we know the expected
* result.
*/
if(status == PCMK_LRM_OP_DONE || status == PCMK_LRM_OP_ERROR) {
status = determine_op_status(rsc, rc, target_rc, node, xml_op, on_fail, data_set);
pe_rsc_trace(rsc, "Remapped %s status to %d", task_key, status);
}
switch (status) {
case PCMK_LRM_OP_CANCELLED:
// Should never happen
pe_err("Resource history contains cancellation '%s' "
"(%s of %s on %s at %s)",
ID(xml_op), task, rsc->id, node->details->uname,
last_change_str(xml_op));
break;
case PCMK_LRM_OP_PENDING:
if (!strcmp(task, CRMD_ACTION_START)) {
pe__set_resource_flags(rsc, pe_rsc_start_pending);
set_active(rsc);
} else if (!strcmp(task, CRMD_ACTION_PROMOTE)) {
rsc->role = RSC_ROLE_PROMOTED;
} else if (!strcmp(task, CRMD_ACTION_MIGRATE) && node->details->unclean) {
/* If a pending migrate_to action is out on a unclean node,
* we have to force the stop action on the target. */
const char *migrate_target = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_TARGET);
pe_node_t *target = pe_find_node(data_set->nodes, migrate_target);
if (target) {
stop_action(rsc, target, FALSE);
}
}
if (rsc->pending_task == NULL) {
if ((interval_ms != 0) || strcmp(task, CRMD_ACTION_STATUS)) {
rsc->pending_task = strdup(task);
rsc->pending_node = node;
} else {
/* Pending probes are not printed, even if pending
* operations are requested. If someone ever requests that
* behavior, enable the below and the corresponding part of
* native.c:native_pending_task().
*/
#if 0
rsc->pending_task = strdup("probe");
rsc->pending_node = node;
#endif
}
}
break;
case PCMK_LRM_OP_DONE:
pe_rsc_trace(rsc, "%s of %s on %s completed at %s " CRM_XS " id=%s",
task, rsc->id, node->details->uname,
last_change_str(xml_op), ID(xml_op));
update_resource_state(rsc, node, xml_op, task, rc, *last_failure, on_fail, data_set);
break;
case PCMK_LRM_OP_NOT_INSTALLED:
failure_strategy = get_action_on_fail(rsc, task_key, task, data_set);
if (failure_strategy == action_fail_ignore) {
crm_warn("Cannot ignore failed %s of %s on %s: "
"Resource agent doesn't exist "
CRM_XS " status=%d rc=%d id=%s",
task, rsc->id, node->details->uname, status, rc,
ID(xml_op));
/* Also for printing it as "FAILED" by marking it as pe_rsc_failed later */
*on_fail = action_fail_migrate;
}
resource_location(parent, node, -INFINITY, "hard-error", data_set);
unpack_rsc_op_failure(rsc, node, rc, xml_op, last_failure, on_fail, data_set);
break;
case PCMK_LRM_OP_NOT_CONNECTED:
if (pe__is_guest_or_remote_node(node)
&& pcmk_is_set(node->details->remote_rsc->flags, pe_rsc_managed)) {
/* We should never get into a situation where a managed remote
* connection resource is considered OK but a resource action
* behind the connection gets a "not connected" status. But as a
* fail-safe in case a bug or unusual circumstances do lead to
* that, ensure the remote connection is considered failed.
*/
pe__set_resource_flags(node->details->remote_rsc,
pe_rsc_failed|pe_rsc_stop);
}
// fall through
case PCMK_LRM_OP_ERROR:
case PCMK_LRM_OP_ERROR_HARD:
case PCMK_LRM_OP_ERROR_FATAL:
case PCMK_LRM_OP_TIMEOUT:
case PCMK_LRM_OP_NOTSUPPORTED:
case PCMK_LRM_OP_INVALID:
failure_strategy = get_action_on_fail(rsc, task_key, task, data_set);
if ((failure_strategy == action_fail_ignore)
|| (failure_strategy == action_fail_restart_container
&& !strcmp(task, CRMD_ACTION_STOP))) {
crm_warn("Pretending failed %s (%s%s%s) of %s on %s at %s "
"succeeded " CRM_XS " rc=%d id=%s",
task, services_ocf_exitcode_str(rc),
(*exit_reason? ": " : ""), exit_reason, rsc->id,
node->details->uname, last_change_str(xml_op), rc,
ID(xml_op));
update_resource_state(rsc, node, xml_op, task, target_rc, *last_failure, on_fail, data_set);
crm_xml_add(xml_op, XML_ATTR_UNAME, node->details->uname);
pe__set_resource_flags(rsc, pe_rsc_failure_ignored);
record_failed_op(xml_op, node, rsc, data_set);
if ((failure_strategy == action_fail_restart_container)
&& cmp_on_fail(*on_fail, action_fail_recover) <= 0) {
*on_fail = failure_strategy;
}
} else {
unpack_rsc_op_failure(rsc, node, rc, xml_op, last_failure, on_fail, data_set);
if(status == PCMK_LRM_OP_ERROR_HARD) {
do_crm_log(rc != PCMK_OCF_NOT_INSTALLED?LOG_ERR:LOG_NOTICE,
"Preventing %s from restarting on %s because "
"of hard failure (%s%s%s)" CRM_XS " rc=%d id=%s",
parent->id, node->details->uname,
services_ocf_exitcode_str(rc),
(*exit_reason? ": " : ""), exit_reason,
rc, ID(xml_op));
resource_location(parent, node, -INFINITY, "hard-error", data_set);
} else if(status == PCMK_LRM_OP_ERROR_FATAL) {
crm_err("Preventing %s from restarting anywhere because "
"of fatal failure (%s%s%s) " CRM_XS " rc=%d id=%s",
parent->id, services_ocf_exitcode_str(rc),
(*exit_reason? ": " : ""), exit_reason,
rc, ID(xml_op));
resource_location(parent, NULL, -INFINITY, "fatal-error", data_set);
}
}
break;
}
done:
pe_rsc_trace(rsc, "Resource %s after %s: role=%s, next=%s",
rsc->id, task, role2text(rsc->role),
role2text(rsc->next_role));
}
static void
add_node_attrs(xmlNode *xml_obj, pe_node_t *node, bool overwrite,
pe_working_set_t *data_set)
{
const char *cluster_name = NULL;
pe_rule_eval_data_t rule_data = {
.node_hash = NULL,
.role = RSC_ROLE_UNKNOWN,
.now = data_set->now,
.match_data = NULL,
.rsc_data = NULL,
.op_data = NULL
};
g_hash_table_insert(node->details->attrs,
strdup(CRM_ATTR_UNAME), strdup(node->details->uname));
g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_ID),
strdup(node->details->id));
if (pcmk__str_eq(node->details->id, data_set->dc_uuid, pcmk__str_casei)) {
data_set->dc_node = node;
node->details->is_dc = TRUE;
g_hash_table_insert(node->details->attrs,
strdup(CRM_ATTR_IS_DC), strdup(XML_BOOLEAN_TRUE));
} else {
g_hash_table_insert(node->details->attrs,
strdup(CRM_ATTR_IS_DC), strdup(XML_BOOLEAN_FALSE));
}
cluster_name = g_hash_table_lookup(data_set->config_hash, "cluster-name");
if (cluster_name) {
g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_CLUSTER_NAME),
strdup(cluster_name));
}
pe__unpack_dataset_nvpairs(xml_obj, XML_TAG_ATTR_SETS, &rule_data,
node->details->attrs, NULL, overwrite, data_set);
if (pe_node_attribute_raw(node, CRM_ATTR_SITE_NAME) == NULL) {
const char *site_name = pe_node_attribute_raw(node, "site-name");
if (site_name) {
g_hash_table_insert(node->details->attrs,
strdup(CRM_ATTR_SITE_NAME),
strdup(site_name));
} else if (cluster_name) {
/* Default to cluster-name if unset */
g_hash_table_insert(node->details->attrs,
strdup(CRM_ATTR_SITE_NAME),
strdup(cluster_name));
}
}
}
static GList *
extract_operations(const char *node, const char *rsc, xmlNode * rsc_entry, gboolean active_filter)
{
int counter = -1;
int stop_index = -1;
int start_index = -1;
xmlNode *rsc_op = NULL;
GList *gIter = NULL;
GList *op_list = NULL;
GList *sorted_op_list = NULL;
/* extract operations */
op_list = NULL;
sorted_op_list = NULL;
for (rsc_op = pcmk__xe_first_child(rsc_entry);
rsc_op != NULL; rsc_op = pcmk__xe_next(rsc_op)) {
if (pcmk__str_eq((const char *)rsc_op->name, XML_LRM_TAG_RSC_OP,
pcmk__str_none)) {
crm_xml_add(rsc_op, "resource", rsc);
crm_xml_add(rsc_op, XML_ATTR_UNAME, node);
op_list = g_list_prepend(op_list, rsc_op);
}
}
if (op_list == NULL) {
/* if there are no operations, there is nothing to do */
return NULL;
}
sorted_op_list = g_list_sort(op_list, sort_op_by_callid);
/* create active recurring operations as optional */
if (active_filter == FALSE) {
return sorted_op_list;
}
op_list = NULL;
calculate_active_ops(sorted_op_list, &start_index, &stop_index);
for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) {
xmlNode *rsc_op = (xmlNode *) gIter->data;
counter++;
if (start_index < stop_index) {
crm_trace("Skipping %s: not active", ID(rsc_entry));
break;
} else if (counter < start_index) {
crm_trace("Skipping %s: old", ID(rsc_op));
continue;
}
op_list = g_list_append(op_list, rsc_op);
}
g_list_free(sorted_op_list);
return op_list;
}
GList *
find_operations(const char *rsc, const char *node, gboolean active_filter,
pe_working_set_t * data_set)
{
GList *output = NULL;
GList *intermediate = NULL;
xmlNode *tmp = NULL;
xmlNode *status = find_xml_node(data_set->input, XML_CIB_TAG_STATUS, TRUE);
pe_node_t *this_node = NULL;
xmlNode *node_state = NULL;
for (node_state = pcmk__xe_first_child(status); node_state != NULL;
node_state = pcmk__xe_next(node_state)) {
if (pcmk__str_eq((const char *)node_state->name, XML_CIB_TAG_STATE, pcmk__str_none)) {
const char *uname = crm_element_value(node_state, XML_ATTR_UNAME);
if (node != NULL && !pcmk__str_eq(uname, node, pcmk__str_casei)) {
continue;
}
this_node = pe_find_node(data_set->nodes, uname);
if(this_node == NULL) {
CRM_LOG_ASSERT(this_node != NULL);
continue;
} else if (pe__is_guest_or_remote_node(this_node)) {
determine_remote_online_status(data_set, this_node);
} else {
determine_online_status(node_state, this_node, data_set);
}
if (this_node->details->online
|| pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) {
/* offline nodes run no resources...
* unless stonith is enabled in which case we need to
* make sure rsc start events happen after the stonith
*/
xmlNode *lrm_rsc = NULL;
tmp = find_xml_node(node_state, XML_CIB_TAG_LRM, FALSE);
tmp = find_xml_node(tmp, XML_LRM_TAG_RESOURCES, FALSE);
for (lrm_rsc = pcmk__xe_first_child(tmp); lrm_rsc != NULL;
lrm_rsc = pcmk__xe_next(lrm_rsc)) {
if (pcmk__str_eq((const char *)lrm_rsc->name,
XML_LRM_TAG_RESOURCE, pcmk__str_none)) {
const char *rsc_id = crm_element_value(lrm_rsc, XML_ATTR_ID);
if (rsc != NULL && !pcmk__str_eq(rsc_id, rsc, pcmk__str_casei)) {
continue;
}
intermediate = extract_operations(uname, rsc_id, lrm_rsc, active_filter);
output = g_list_concat(output, intermediate);
}
}
}
}
}
return output;
}