X-Git-Url: http://git.tuebingen.mpg.de/?p=dss.git;a=blobdiff_plain;f=dss.c;h=07a60425170ddab063ee6071d4901993a9db8c62;hp=237208b7da0a0f3a8f14c2e86c4cbce7cb1fe735;hb=05e75054398c9d39f62f8c4b9be7b874a2019a3c;hpb=20f8c7d5e2b7985137e89143005f4b7b15ae59d4

diff --git a/dss.c b/dss.c
index 237208b..07a6042 100644
--- a/dss.c
+++ b/dss.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2008-2011 Andre Noll <maan@systemlinux.org>
+ * Copyright (C) 2008-2011 Andre Noll <maan@tuebingen.mpg.de>
  *
  * Licensed under the GPL v2. For licencing details see COPYING.
  */
@@ -45,6 +45,8 @@ static int signal_pipe;
 static pid_t create_pid;
 /** Whether the pre-create-hook/rsync/post-create-hook is currently stopped. */
 static int create_process_stopped;
+/** How many times in a row the rsync command failed. */
+static int num_consecutive_rsync_errors;
 /** Process id of current pre-remove/rm/post-remove process. */
 static pid_t remove_pid;
 /** When the next snapshot is due. */
@@ -125,6 +127,7 @@ static void dump_dss_config(const char *msg)
 		"reference_snapshot: %s\n"
 		"snapshot_creation_status: %s\n"
 		"snapshot_removal_status: %s\n"
+		"num_consecutive_rsync_errors: %d\n"
 		,
 		(int) getpid(),
 		logfile? conf.logfile_arg : "stderr",
@@ -135,7 +138,8 @@ static void dump_dss_config(const char *msg)
 		name_of_reference_snapshot?
 			name_of_reference_snapshot : "(none)",
 		hook_status_description[snapshot_creation_status],
-		hook_status_description[snapshot_removal_status]
+		hook_status_description[snapshot_removal_status],
+		num_consecutive_rsync_errors
 	);
 	if (create_pid != 0)
 		fprintf(log,
@@ -531,17 +535,25 @@ static struct snapshot *find_outdated_snapshot(struct snapshot_list *sl)
 
 static struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
 {
-	int i;
-	struct snapshot *s;
+	int i, num_complete;
+	struct snapshot *s, *ref = NULL;
+
+	num_complete = num_complete_snapshots(sl);
+	if (num_complete <= conf.min_complete_arg)
+		return NULL;
 	FOR_EACH_SNAPSHOT(s, i, sl) {
 		if (snapshot_is_being_created(s))
 			continue;
-		if (is_reference_snapshot(s))
+		if (is_reference_snapshot(s)) { /* avoid this one */
+			ref = s;
 			continue;
+		}
 		DSS_INFO_LOG(("oldest removable snapshot: %s\n", s->name));
 		return s;
 	}
-	return NULL;
+	assert(ref);
+	DSS_WARNING_LOG(("removing reference snapshot %s\n", ref->name));
+	return ref;
 }
 
 static int rename_incomplete_snapshot(int64_t start)
@@ -836,23 +848,23 @@ static int handle_rsync_exit(int status)
 	es = WEXITSTATUS(status);
 	/*
 	 * Restart rsync on non-fatal errors:
-	 * 12: Error in rsync protocol data stream
-	 * 13: Errors with program diagnostics
+	 * 24: Partial transfer due to vanished source files
 	 */
-	if (es == 12 || es == 13) {
-		DSS_WARNING_LOG(("rsync process %d returned %d -- restarting\n",
-			(int)create_pid, es));
+	if (es != 0 && es != 24) {
+		DSS_WARNING_LOG(("rsync exit code %d, error count %d\n",
+			es, ++num_consecutive_rsync_errors));
+		if (num_consecutive_rsync_errors > conf.max_rsync_errors_arg) {
+			ret = -E_TOO_MANY_RSYNC_ERRORS;
+			snapshot_creation_status = HS_READY;
+			goto out;
+		}
+		DSS_WARNING_LOG(("restarting rsync process\n"));
 		snapshot_creation_status = HS_NEEDS_RESTART;
 		next_snapshot_time = get_current_time() + 60;
 		ret = 1;
 		goto out;
 	}
-	if (es != 0 && es != 23 && es != 24) {
-		DSS_ERROR_LOG(("rsync process %d returned %d\n", (int)create_pid, es));
-		ret = -E_BAD_EXIT_CODE;
-		snapshot_creation_status = HS_READY;
-		goto out;
-	}
+	num_consecutive_rsync_errors = 0;
 	ret = rename_incomplete_snapshot(current_snapshot_creation_time);
 	if (ret < 0)
 		goto out;
@@ -1038,6 +1050,13 @@ static int handle_sighup(void)
 	return change_to_dest_dir();
 }
 
+static void kill_children(void)
+{
+	restart_create_process();
+	dss_kill(create_pid, SIGTERM, NULL);
+	dss_kill(remove_pid, SIGTERM, NULL);
+}
+
 static int handle_signal(void)
 {
 	int sig, ret = next_signal();
@@ -1048,9 +1067,7 @@ static int handle_signal(void)
 	switch (sig) {
 	case SIGINT:
 	case SIGTERM:
-		restart_create_process();
-		dss_kill(create_pid, SIGTERM, NULL);
-		dss_kill(remove_pid, SIGTERM, NULL);
+		kill_children();
 		ret = -E_SIGNAL;
 		break;
 	case SIGHUP:
@@ -1319,6 +1336,7 @@ static int com_run(void)
 	ret = select_loop();
 	if (ret >= 0) /* impossible */
 		ret = -E_BUG;
+	kill_children();
 	exit_hook(ret);
 	return ret;
 }
@@ -1506,6 +1524,7 @@ int main(int argc, char **argv)
 	if (ret < 0)
 		goto out;
 	ret = call_command_handler();
+	signal_shutdown();
 out:
 	if (ret < 0)
 		DSS_EMERG_LOG(("%s\n", dss_strerror(-ret)));