All of vdi I/O and dog command freeze during deleting files in .stale dir

[masahiro tsuji]

Hi

I found a problem that all of vdi I/O and dog command freeze during deleting files in .stale dir.

The system take long time(about 10 min) for deleteing .stae files( about 100,000 object files) when recovery has been finished.
And VM got truble during deleting the files.

I make patch for extend time of delete .stale files.
And reproduce the problem.

I suspect sd_read_lock(&md_lock) in for_each_obj_path.

[root@node001 sheep]# diff -u plain_store.c-org plain_store.c
--- plain_store.c-org 2014-10-08 13:40:00.851572835 +0900
+++ plain_store.c 2014-10-08 13:43:20.374507384 +0900
@@ -9,6 +9,7 @@
  * along with this program. If not, see <http://www.gnu.org/licenses/>.
  */

+#include <unistd.h>
 #include <libgen.h>

 #include "sheep_priv.h"
@@ -156,6 +157,9 @@
 static int purge_stale_dir(char *path)
 {
        char p[PATH_MAX];
+ sd_debug("wait 600 sec in pruge_dtale_dir");
+ sleep(600);
+ sd_debug("wait 600 sec in pruge_dtale_dir done");

        snprintf(p, PATH_MAX, "%s/.stale", path);
        return purge_dir(p);

[masahiro tsuji (masa-tsuji)]

I added sleep at top of default_cleanup(void). it was same result.
So I/O is freeze during default_cleanup.