Don't wait for likely leaked transactions on exit

db/db_tunables.c

@@ -480,6 +480,7 @@ extern int gbl_timer_warn_interval;
 int gbl_incoherent_clnt_wait = 0;
 int gbl_new_leader_duration = 3;
 extern int gbl_transaction_grace_period;
+extern int gbl_transaction_grace_period_timeout;
 
 /*
   =========================================================

db/db_tunables.h

@@ -2497,5 +2497,9 @@ REGISTER_TUNABLE("transaction_grace_period",
                  "Time to wait for connections with pending transactions to go away on exit. (Default: 60)",
                  TUNABLE_INTEGER, &gbl_transaction_grace_period, 0, NULL, NULL, NULL, NULL);
 
+REGISTER_TUNABLE("transaction_grace_period_timeout",
+                 "Ignore transactions at exit older than this (seconds). (Default: 5min)",
+                 TUNABLE_INTEGER, &gbl_transaction_grace_period_timeout, 0, NULL, NULL, NULL, NULL);
+
 
 #endif /* _DB_TUNABLES_H */

db/sql.h

@@ -1208,6 +1208,7 @@ struct connection_info {
     int time_in_state_int;
     enum connection_state state_int;
     int64_t in_transaction;
+    char *argv0;
 };
 
 /* makes master swing verbose */

db/sqlinterfaces.c

@@ -6682,6 +6682,7 @@ static void gather_connection_int(struct connection_info *c, struct sqlclntstate
         c->fingerprint = NULL;
     }
     c->in_transaction = clnt->in_client_trans;
+    c->argv0 = strdup(clnt->argv0);
     Pthread_mutex_unlock(&clnt->state_lk);
 }
 
@@ -6731,6 +6732,7 @@ void free_connection_info(struct connection_info *info, int num_connections)
     for (int i = 0; i < num_connections; i++) {
         if (info[i].sql) free(info[i].sql);
         if (info[i].fingerprint) free(info[i].fingerprint);
+        if (info[i].argv0) free(info[i].argv0);
         /* state is static, don't free */
     }
     free(info);
@@ -7166,15 +7168,18 @@ char *clnt_tzname(struct sqlclntstate *clnt, sqlite3_stmt *stmt)
 }
 
 int gbl_transaction_grace_period = 60;
+int gbl_transaction_grace_period_timeout = 300 ;
 void wait_for_transactions(void) {
     int ntrans;
     int nwaits = 0;
+    static int once = 1;
 
 
     for (nwaits = 0; nwaits < gbl_transaction_grace_period; nwaits++) {
         ntrans = 0;
         struct connection_info *connections;
         int nconnections;
+        int now = comdb2_time_epochms();
 
         int rc = gather_connection_info(&connections, &nconnections);
         if (rc) {
@@ -7183,15 +7188,20 @@ void wait_for_transactions(void) {
         }
         for (int i = 0; i < nconnections; i++) {
             struct connection_info *c = &connections[i];
-            if (c->in_transaction) {
+            int runtime = (now - c->time_in_state_int) / 1000;
+            if (c->in_transaction && runtime < gbl_transaction_grace_period_timeout) {
                 ntrans++;
                 if (nwaits > 10) {
                     if (ntrans == 1)
                         logmsg(LOGMSG_INFO, "waiting for transactions:\n------------------------\n");
-                    logmsg(LOGMSG_INFO, "open transaction on connection from %s pid %d\n", c->host, (int) c->pid);
+                    logmsg(LOGMSG_INFO, "open transaction on connection from %s pid %d argv0 %s age %d seconds\n", c->host, (int) c->pid, c->argv0, runtime);
                 }
             }
+            if (once && c->in_transaction && runtime >= gbl_transaction_grace_period_timeout) {
+                logmsg(LOGMSG_INFO, "old open transaction on connection from %s pid %d argv0 %s age %d seconds\n", c->host, (int) c->pid, c->argv0, runtime);
+            }
         }
+        once = 0;
         free_connection_info(connections, nconnections);
         if (ntrans) {
             if (nwaits > 10)

sqlite/ext/comdb2/connections.c

@@ -97,5 +97,7 @@ int systblConnectionsInit(sqlite3 *db) {
             CDB2_INTEGER, "is_ssl", -1, offsetof(struct connection_info, is_ssl),
             CDB2_INTEGER, "has_cert", -1, offsetof(struct connection_info, has_cert),
             CDB2_CSTRING, "common_name", -1, offsetof(struct connection_info, common_name),
+            CDB2_CSTRING, "argv0", -1, offsetof(struct connection_info, argv0),
+            CDB2_INTEGER, "in_transaction", -1, offsetof(struct connection_info, in_transaction),
             SYSTABLE_END_OF_FIELDS);
 }

tests/tunables.test/t00_all_tunables.expected

@@ -987,6 +987,7 @@
 (name='track_replication_times_max_lsns', description='Track replication times for up to this many transactions.', type='INTEGER', value='50', read_only='N')
 (name='tracked_locklist_init', description='Initial allocation count for tracked locks', type='INTEGER', value='10', read_only='N')
 (name='transaction_grace_period', description='Time to wait for connections with pending transactions to go away on exit. (Default: 60)', type='INTEGER', value='60', read_only='N')
+(name='transaction_grace_period_timeout', description='Ignore transactions at exit older than this (seconds). (Default: 5min)', type='INTEGER', value='300', read_only='N')
 (name='transient_page_reallocation', description='Orphaned pages are maintained locally', type='BOOLEAN', value='OFF', read_only='N')
 (name='typessql', description='Use typessql to attempt to buffer results until all columns are non-null. (Default: off)', type='BOOLEAN', value='OFF', read_only='N')
 (name='typessql_records_max', description='The maximum number of records for typessql to buffer and find non-null types before giving up. (Default: 1000)', type='INTEGER', value='1000', read_only='N')