X-Git-Url: http://git.tuebingen.mpg.de/?p=dss.git;a=blobdiff_plain;f=dss.c;h=0bc891373bd7e6096063ebc45aa953582433d68c;hp=006cd27c333e097231ae140374fffdf3b8d689f6;hb=bf4beb6878cfcf692755807486f95e835f807b72;hpb=15eb99dcf9a30f8c8d488d5bc3c92130cf828d60

diff --git a/dss.c b/dss.c
index 006cd27..0bc8913 100644
--- a/dss.c
+++ b/dss.c
@@ -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,
@@ -602,9 +606,18 @@ static int try_to_free_disk_space(void)
 		if (next_snapshot_is_due())
 			return 0;
 	}
+	/*
+	 * Idle and --keep_redundant not given, or low disk space. Look at
+	 * existing snapshots.
+	 */
 	dss_get_snapshot_list(&sl);
 	ret = 0;
-	if (!low_disk_space && sl.num_snapshots <= 1)
+	/*
+	 * Don't remove anything if there is free space and we have fewer
+	 * snapshots than configured, plus one. This way there is always one
+	 * snapshot that can be recycled.
+	 */
+	if (!low_disk_space && sl.num_snapshots <= 1 << conf.num_intervals_arg)
 		goto out;
 	why = "outdated";
 	victim = find_outdated_snapshot(&sl);
@@ -614,13 +627,13 @@ static int try_to_free_disk_space(void)
 	victim = find_redundant_snapshot(&sl);
 	if (victim)
 		goto remove;
-	/* try harder only if disk space is low */
-	if (!low_disk_space)
-		goto out;
 	why = "orphaned";
 	victim = find_orphaned_snapshot(&sl);
 	if (victim)
 		goto remove;
+	/* try harder only if disk space is low */
+	if (!low_disk_space)
+		goto out;
 	DSS_WARNING_LOG(("disk space low and nothing obvious to remove\n"));
 	victim = find_oldest_removable_snapshot(&sl);
 	if (victim)
@@ -844,23 +857,27 @@ 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 (conf.create_given) {
+			ret = -E_BAD_EXIT_CODE;
+			goto out;
+		}
+		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;
@@ -1476,12 +1493,6 @@ static int setup_signal_handling(void)
 	return install_sighandler(SIGCHLD);
 }
 
-/**
- * The main function of dss.
- *
- * \param argc Usual argument count.
- * \param argv Usual argument vector.
- */
 int main(int argc, char **argv)
 {
 	int ret;
@@ -1520,6 +1531,7 @@ 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)));