From: Vsevolod Stakhov <vsevolod@highsecure.ru>
Date: Mon, 15 Aug 2016 10:49:34 +0000 (+0100)
Subject: [Fix] Try to deal with multiple workers terminated
X-Git-Tag: 1.3.3~1
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=e67ced158a0cac42f6b89a18552dbd69c8f67c54;p=thirdparty%2Frspamd.git

[Fix] Try to deal with multiple workers terminated
---

diff --git a/src/rspamd.c b/src/rspamd.c
index 9bea4b5c03..5692e315b1 100644
--- a/src/rspamd.c
+++ b/src/rspamd.c
@@ -917,73 +917,74 @@ rspamd_cld_handler (gint signo, short what, gpointer arg)
 	/* Turn off locking for logger */
 	rspamd_log_nolock (rspamd_main->logger);
 
-	msg_debug_main ("catch SIGCHLD signal, finding terminated worker");
+	msg_debug_main ("catch SIGCHLD signal, finding terminated workers");
 	/* Remove dead child form children list */
-	wrk = waitpid (0, &res, 0);
-	if ((cur =
-				 g_hash_table_lookup (rspamd_main->workers,
-						 GSIZE_TO_POINTER (wrk))) != NULL) {
-		/* Unlink dead process from queue and hash table */
-
-		g_hash_table_remove (rspamd_main->workers, GSIZE_TO_POINTER (
-				wrk));
-
-		if (WIFEXITED (res) && WEXITSTATUS (res) == 0) {
-			/* Normal worker termination, do not fork one more */
-			msg_info_main ("%s process %P terminated normally",
-					g_quark_to_string (cur->type),
-					cur->pid);
-		}
-		else {
-			if (WIFSIGNALED (res)) {
+	while ((wrk = waitpid (0, &res, WNOHANG)) > 0) {
+		if ((cur =
+				g_hash_table_lookup (rspamd_main->workers,
+						GSIZE_TO_POINTER (wrk))) != NULL) {
+			/* Unlink dead process from queue and hash table */
+
+			g_hash_table_remove (rspamd_main->workers, GSIZE_TO_POINTER (
+					wrk));
+
+			if (WIFEXITED (res) && WEXITSTATUS (res) == 0) {
+				/* Normal worker termination, do not fork one more */
+				msg_info_main ("%s process %P terminated normally",
+						g_quark_to_string (cur->type),
+						cur->pid);
+			}
+			else {
+				if (WIFSIGNALED (res)) {
 #ifdef WCOREDUMP
-				if (WCOREDUMP (res)) {
+					if (WCOREDUMP (res)) {
+						msg_warn_main (
+								"%s process %P terminated abnormally by signal: %d"
+								" and created core file",
+								g_quark_to_string (cur->type),
+								cur->pid,
+								WTERMSIG (res));
+					}
+					else {
+						msg_warn_main (
+								"%s process %P terminated abnormally by signal: %d"
+								" but NOT created core file",
+								g_quark_to_string (cur->type),
+								cur->pid,
+								WTERMSIG (res));
+					}
+#else
 					msg_warn_main (
-							"%s process %P terminated abnormally by signal: %d"
-							" and created core file",
+							"%s process %P terminated abnormally by signal: %d",
 							g_quark_to_string (cur->type),
 							cur->pid,
 							WTERMSIG (res));
+#endif
 				}
 				else {
-					msg_warn_main (
-							"%s process %P terminated abnormally by signal: %d"
-							" but NOT created core file",
+					msg_warn_main ("%s process %P terminated abnormally "
+							"with exit code %d",
 							g_quark_to_string (cur->type),
 							cur->pid,
-							WTERMSIG (res));
+							WEXITSTATUS (res));
 				}
-#else
-				msg_warn_main (
-						"%s process %P terminated abnormally by signal: %d",
-						g_quark_to_string (cur->type),
-						cur->pid,
-						WTERMSIG (res));
-#endif
-			}
-			else {
-				msg_warn_main ("%s process %P terminated abnormally "
-						"with exit code %d",
-						g_quark_to_string (cur->type),
-						cur->pid,
-						WEXITSTATUS (res));
+				/* Fork another worker in replace of dead one */
+				rspamd_check_core_limits (rspamd_main);
+				rspamd_fork_delayed (cur->cf, cur->index, rspamd_main);
 			}
-			/* Fork another worker in replace of dead one */
-			rspamd_check_core_limits (rspamd_main);
-			rspamd_fork_delayed (cur->cf, cur->index, rspamd_main);
-		}
 
-		event_del (&cur->srv_ev);
-		/* We also need to clean descriptors left */
-		close (cur->control_pipe[0]);
-		close (cur->srv_pipe[0]);
-		g_free (cur);
-	}
-	else {
-		for (i = 0; i < other_workers->len; i++) {
-			if (g_array_index (other_workers, pid_t, i) == wrk) {
-				g_array_remove_index_fast (other_workers, i);
-				msg_info_main ("related process %P terminated", wrk);
+			event_del (&cur->srv_ev);
+			/* We also need to clean descriptors left */
+			close (cur->control_pipe[0]);
+			close (cur->srv_pipe[0]);
+			g_free (cur);
+		}
+		else {
+			for (i = 0; i < other_workers->len; i++) {
+				if (g_array_index (other_workers, pid_t, i) == wrk) {
+					g_array_remove_index_fast (other_workers, i);
+					msg_info_main ("related process %P terminated", wrk);
+				}
 			}
 		}
 	}