diff --git a/daemons/pacemakerd/pacemakerd.c b/daemons/pacemakerd/pacemakerd.c index acec357b45a..07f1f1eded9 100644 --- a/daemons/pacemakerd/pacemakerd.c +++ b/daemons/pacemakerd/pacemakerd.c @@ -259,6 +259,8 @@ main(int argc, char **argv) pcmk_ipc_api_t *old_instance = NULL; qb_ipcs_service_t *ipcs = NULL; + subdaemon_check_progress = time(NULL); + crm_log_preinit(NULL, argc, argv); mainloop_add_signal(SIGHUP, pcmk_ignore); mainloop_add_signal(SIGQUIT, pcmk_sigquit); diff --git a/daemons/pacemakerd/pacemakerd.h b/daemons/pacemakerd/pacemakerd.h index 7c541bbf9e8..38e986b88a1 100644 --- a/daemons/pacemakerd/pacemakerd.h +++ b/daemons/pacemakerd/pacemakerd.h @@ -21,6 +21,7 @@ extern unsigned int shutdown_complete_state_reported_to; extern gboolean shutdown_complete_state_reported_client_closed; extern crm_trigger_t *shutdown_trigger; extern crm_trigger_t *startup_trigger; +extern time_t subdaemon_check_progress; gboolean mcp_read_config(void); diff --git a/daemons/pacemakerd/pcmkd_messages.c b/daemons/pacemakerd/pcmkd_messages.c index 0439986ecfd..abb0cfb67b3 100644 --- a/daemons/pacemakerd/pcmkd_messages.c +++ b/daemons/pacemakerd/pcmkd_messages.c @@ -25,7 +25,6 @@ pcmk_handle_ping_request(pcmk__client_t *c, xmlNode *msg, uint32_t id) const char *value = NULL; xmlNode *ping = NULL; xmlNode *reply = NULL; - time_t pinged = time(NULL); const char *from = crm_element_value(msg, F_CRM_SYS_FROM); /* Pinged for status */ @@ -36,7 +35,8 @@ pcmk_handle_ping_request(pcmk__client_t *c, xmlNode *msg, uint32_t id) value = crm_element_value(msg, F_CRM_SYS_TO); crm_xml_add(ping, XML_PING_ATTR_SYSFROM, value); crm_xml_add(ping, XML_PING_ATTR_PACEMAKERDSTATE, pacemakerd_state); - crm_xml_add_ll(ping, XML_ATTR_TSTAMP, (long long) pinged); + crm_xml_add_ll(ping, XML_ATTR_TSTAMP, + (long long) subdaemon_check_progress); crm_xml_add(ping, XML_PING_ATTR_STATUS, "ok"); reply = create_reply(msg, ping); free_xml(ping); diff --git a/daemons/pacemakerd/pcmkd_subdaemons.c b/daemons/pacemakerd/pcmkd_subdaemons.c index 9556d7409cd..5a22a40ca77 100644 --- a/daemons/pacemakerd/pcmkd_subdaemons.c +++ b/daemons/pacemakerd/pcmkd_subdaemons.c @@ -32,14 +32,16 @@ typedef struct pcmk_child_s { const char *command; const char *endpoint; /* IPC server name */ bool needs_cluster; + int check_count; /* Anything below here will be dynamically initialized */ bool needs_retry; bool active_before_startup; } pcmk_child_t; -#define PCMK_PROCESS_CHECK_INTERVAL 5 -#define SHUTDOWN_ESCALATION_PERIOD 180000 /* 3m */ +#define PCMK_PROCESS_CHECK_INTERVAL 1 +#define PCMK_PROCESS_CHECK_RETRIES 5 +#define SHUTDOWN_ESCALATION_PERIOD 180000 /* 3m */ /* Index into the array below */ #define PCMK_CHILD_CONTROLD 5 @@ -82,6 +84,7 @@ static char *opts_vgrind[] = { NULL, NULL, NULL, NULL, NULL }; crm_trigger_t *shutdown_trigger = NULL; crm_trigger_t *startup_trigger = NULL; +time_t subdaemon_check_progress = 0; /* When contacted via pacemakerd-api by a client having sbd in * the name we assume it is sbd-daemon which wants to know @@ -103,7 +106,6 @@ gboolean running_with_sbd = FALSE; /* local copy */ GMainLoop *mainloop = NULL; static gboolean fatal_error = FALSE; -static bool global_keep_tracking = false; static gboolean check_active_before_startup_processes(gpointer user_data); static int child_liveness(pcmk_child_t *child); @@ -127,44 +129,94 @@ pcmkd_cluster_connected(void) static gboolean check_active_before_startup_processes(gpointer user_data) { - gboolean keep_tracking = FALSE; - - for (int i = 0; i < PCMK__NELEM(pcmk_children); i++) { - if (!pcmk_children[i].active_before_startup) { - /* we are already tracking it as a child process. */ - continue; - } else { - int rc = child_liveness(&pcmk_children[i]); - - switch (rc) { - case pcmk_rc_ok: - break; - case pcmk_rc_ipc_unresponsive: - case pcmk_rc_ipc_pid_only: // This case: it was previously OK - if (pcmk_children[i].respawn) { - crm_err("%s[%lld] terminated%s", pcmk_children[i].name, - (long long) PCMK__SPECIAL_PID_AS_0(pcmk_children[i].pid), - (rc == pcmk_rc_ipc_pid_only)? " as IPC server" : ""); - } else { - /* orderly shutdown */ - crm_notice("%s[%lld] terminated%s", pcmk_children[i].name, - (long long) PCMK__SPECIAL_PID_AS_0(pcmk_children[i].pid), - (rc == pcmk_rc_ipc_pid_only)? " as IPC server" : ""); - } - pcmk_process_exit(&(pcmk_children[i])); - continue; - default: - crm_exit(CRM_EX_FATAL); - break; /* static analysis/noreturn */ + static int next_child = 0; + int rc = child_liveness(&pcmk_children[next_child]); + + crm_trace("%s[%lld] checked as %d", + pcmk_children[next_child].name, + (long long) PCMK__SPECIAL_PID_AS_0( + pcmk_children[next_child].pid), + rc); + + switch (rc) { + case pcmk_rc_ok: + pcmk_children[next_child].check_count = 0; + next_child++; + subdaemon_check_progress = time(NULL); + break; + case pcmk_rc_ipc_pid_only: // This case: it was previously OK + pcmk_children[next_child].check_count++; + if (pcmk_children[next_child].check_count >= PCMK_PROCESS_CHECK_RETRIES) { + crm_err("%s[%lld] is unresponsive to ipc after %d tries but " + "we found the pid so have it killed that we can restart", + pcmk_children[next_child].name, + (long long) PCMK__SPECIAL_PID_AS_0( + pcmk_children[next_child].pid), + pcmk_children[next_child].check_count); + stop_child(&pcmk_children[next_child], SIGKILL); + if (pcmk_children[next_child].respawn) { + /* as long as the respawn-limit isn't reached + give it another round of check retries + */ + pcmk_children[next_child].check_count = 0; + } + } else { + crm_notice("%s[%lld] is unresponsive to ipc after %d tries", + pcmk_children[next_child].name, + (long long) PCMK__SPECIAL_PID_AS_0( + pcmk_children[next_child].pid), + pcmk_children[next_child].check_count); + if (pcmk_children[next_child].respawn) { + /* as long as the respawn-limit isn't reached + and we haven't run out of connect retries + we account this as progress we are willing + to tell to sbd + */ + subdaemon_check_progress = time(NULL); + } } - } - /* at least one of the processes found at startup - * is still going, so keep this recurring timer around */ - keep_tracking = TRUE; + /* go to the next child and see if + we can make progress there + */ + next_child++; + break; + case pcmk_rc_ipc_unresponsive: + if (pcmk_children[next_child].respawn) { + crm_err("%s[%lld] terminated", + pcmk_children[next_child].name, + (long long) PCMK__SPECIAL_PID_AS_0( + pcmk_children[next_child].pid)); + } else { + /* orderly shutdown */ + crm_notice("%s[%lld] terminated", + pcmk_children[next_child].name, + (long long) PCMK__SPECIAL_PID_AS_0( + pcmk_children[next_child].pid)); + } + pcmk_process_exit(&(pcmk_children[next_child])); + if (!pcmk_children[next_child].respawn) { + /* if a subdaemon is down and we don't want it + to be restarted this is a success during + shutdown. if it isn't restarted anymore + due to MAX_RESPAWN it is + rather no success. + */ + if (pcmk_children[next_child].respawn_count <= MAX_RESPAWN) { + subdaemon_check_progress = time(NULL); + } + next_child++; + } + break; + default: + crm_exit(CRM_EX_FATAL); + break; /* static analysis/noreturn */ } - global_keep_tracking = keep_tracking; - return keep_tracking; + if (next_child >= PCMK__NELEM(pcmk_children)) { + next_child = 0; + } + + return G_SOURCE_CONTINUE; } static gboolean @@ -257,11 +309,6 @@ pcmk_process_exit(pcmk_child_t * child) child->name, child->endpoint); /* need to monitor how it evolves, and start new process if badly */ child->active_before_startup = true; - if (!global_keep_tracking) { - global_keep_tracking = true; - g_timeout_add_seconds(PCMK_PROCESS_CHECK_INTERVAL, - check_active_before_startup_processes, NULL); - } } else { if (child->needs_cluster && !pcmkd_cluster_connected()) { @@ -647,7 +694,6 @@ child_liveness(pcmk_child_t *child) int find_and_track_existing_processes(void) { - bool tracking = false; bool wait_in_progress; int rc; size_t i, rounds; @@ -715,7 +761,6 @@ find_and_track_existing_processes(void) pcmk_children[i].pid)); pcmk_children[i].respawn_count = -1; /* 0~keep watching */ pcmk_children[i].active_before_startup = true; - tracking = true; break; case pcmk_rc_ipc_pid_only: if (pcmk_children[i].respawn_count == WAIT_TRIES) { @@ -750,10 +795,8 @@ find_and_track_existing_processes(void) pcmk_children[i].respawn_count = 0; /* restore pristine state */ } - if (tracking) { - g_timeout_add_seconds(PCMK_PROCESS_CHECK_INTERVAL, + g_timeout_add_seconds(PCMK_PROCESS_CHECK_INTERVAL, check_active_before_startup_processes, NULL); - } return pcmk_rc_ok; }