uint32_t monitor_interval;
uint32_t tickle_update_interval;
uint32_t script_timeout;
+ uint32_t script_ban_count; /* ban after this many consec timeouts*/
uint32_t recovery_grace_period;
uint32_t recovery_ban_period;
uint32_t database_hash_size;
struct ctdb_monitor_state *monitor;
struct ctdb_log_state *log;
int start_as_disabled;
+ uint32_t event_script_timeouts; /* counting how many consecutive times an eventscript has timedout */
TALLOC_CTX *eventscripts_ctx; /* a context to hold data for the RUN_EVENTSCRIPTS control */
};
{ "MonitorInterval", 15, offsetof(struct ctdb_tunable, monitor_interval) },
{ "TickleUpdateInterval",20, offsetof(struct ctdb_tunable, tickle_update_interval) },
{ "EventScriptTimeout", 20, offsetof(struct ctdb_tunable, script_timeout) },
+ { "EventScriptBanCount", 5, offsetof(struct ctdb_tunable, script_ban_count) },
{ "RecoveryGracePeriod", 60, offsetof(struct ctdb_tunable, recovery_grace_period) },
{ "RecoveryBanPeriod", 300, offsetof(struct ctdb_tunable, recovery_ban_period) },
{ "DatabaseHashSize", 10000, offsetof(struct ctdb_tunable, database_hash_size) },
talloc_set_destructor(state, NULL);
talloc_free(state);
callback(ctdb, status, private_data);
+
+ ctdb->event_script_timeouts = 0;
+}
+
+static void ctdb_ban_self(struct ctdb_context *ctdb, uint32_t ban_period)
+{
+ int ret;
+ struct ctdb_ban_info b;
+ TDB_DATA data;
+
+ b.pnn = ctdb->pnn;
+ b.ban_time = ban_period;
+
+ data.dptr = (uint8_t *)&b;
+ data.dsize = sizeof(b);
+
+ ret = ctdb_daemon_send_message(ctdb, CTDB_BROADCAST_CONNECTED,
+ CTDB_SRVID_BAN_NODE, data);
+ if (ret != 0) {
+ DEBUG(DEBUG_ERR,(__location__ " Failed to send ban message\n"));
+ }
}
void *private_data = state->private_data;
struct ctdb_context *ctdb = state->ctdb;
- DEBUG(DEBUG_ERR,("event script timed out : %s\n", state->options));
+ DEBUG(DEBUG_ERR,("Event script timed out : %s count : %u\n", state->options, ctdb->event_script_timeouts));
+
talloc_free(state);
callback(ctdb, -1, private_data);
+
+ ctdb->event_script_timeouts++;
+ if (ctdb->event_script_timeouts > ctdb->tunable.script_ban_count) {
+ ctdb->event_script_timeouts = 0;
+ DEBUG(DEBUG_ERR, ("Maximum timeout count reached for eventscript. Banning self for %d seconds\n", ctdb->tunable.recovery_ban_period));
+ ctdb_ban_self(ctdb, ctdb->tunable.recovery_ban_period);
+ }
}
/*
if (!timeval_is_zero(&timeout)) {
event_add_timed(ctdb->ev, state, timeout, ctdb_event_script_timeout, state);
} else {
- DEBUG(DEBUG_ERR, (__location__ " eventscript %s called with no timeout\n", fmt));
+ DEBUG(DEBUG_ERR, (__location__ " eventscript %s called with no timeout\n", state->options));
}
return 0;