From: manoj <manoj>
Date: Tue, 14 Jul 2009 21:11:44 +0000 (+0000)
Subject: b=19894
X-Git-Tag: v1_9_220~23
X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=commitdiff_plain;h=054ce9ccdefb7fd15aa33e967ca882372b87c29d

b=19894
r=grev
r=nathan.rutman

Kill and cleanup iozone before replicating data.
---

diff --git a/lustre/tests/acceptance-small.sh b/lustre/tests/acceptance-small.sh
index 7645795..320eeeb 100755
--- a/lustre/tests/acceptance-small.sh
+++ b/lustre/tests/acceptance-small.sh
@@ -438,7 +438,6 @@ if [ "$SANITY_GSS" != "no" ]; then
 fi
 
 
-echo replication sanity: $LREPLICATE_TEST
 [ "$LREPLICATE_TEST" != "no" ] && skip_remmds lreplicate-test && LREPLICATE_TEST=no && MSKIPPED=1
 [ "$LREPLICATE_TEST" != "no" ] && skip_remost lreplicate-test && LREPLICATE_TEST=no && OSKIPPED=1
 if [ "$LREPLICATE_TEST" != "no" ]; then
diff --git a/lustre/tests/lreplicate-test.sh b/lustre/tests/lreplicate-test.sh
index 12dbcb9..4e0d261 100644
--- a/lustre/tests/lreplicate-test.sh
+++ b/lustre/tests/lreplicate-test.sh
@@ -392,7 +392,20 @@ test_4() {
 
     $KILL -SIGCONT $child_pid
     sleep 60
-    $KILL -SIGKILL $child_pid
+    $KILL -SIGKILL $(pgrep run_iozone.sh)
+    $KILL -SIGKILL $(pgrep iozone)
+
+    # After killing 'run_iozone.sh', process 'iozone' becomes the
+    # child of PID 1. Hence 'wait' does not wait for it. Killing
+    # iozone first, means more iozone processes are spawned off which
+    # is not desirable. So, after sending a sigkill, the test goes
+    # into a wait loop for iozone to cleanup and exit.
+    wait
+    while [ "$(pgrep "iozone")" != "" ];
+    do
+      ps -ef | grep iozone | grep -v grep
+      sleep 1;
+    done
 
     $LREPLICATE -l $LREPL_LOG
     check_diff $DIR/$tdir $TGT/$tdir
@@ -563,5 +576,5 @@ log "cleanup: ======================================================"
 cd $ORIG_PWD
 check_and_cleanup_lustre
 echo '=========================== finished ==============================='
-[ -f "$REPLOG" ] && cat $REPLLOG && grep -q FAIL $REPLLOG && exit 1 || true
+[ -f "$REPLLOG" ] && cat $REPLLOG && grep -q FAIL $REPLLOG && exit 1 || true
 echo "$0: completed"
diff --git a/lustre/utils/lreplicate.c b/lustre/utils/lreplicate.c
index 1eb775a..b1b64e4 100644
--- a/lustre/utils/lreplicate.c
+++ b/lustre/utils/lreplicate.c
@@ -1352,6 +1352,16 @@ void lr_print_status(struct lr_info *info)
                 printf("Using rsync: %s (%s)\n", rsync, rsync_ver);
 }
 
+DECLARE_CHANGELOG_NAMES;
+
+void lr_print_failure(struct lr_info *info, int rc)
+{
+        fprintf(stderr, "Replication of operation failed(%d):"
+                " %lld %s (%d) %s %s %s\n", rc, info->recno, 
+                changelog_str[info->type], info->type, info->tfid,
+                info->pfid, info->name);
+}
+
 /* Replicate filesystem operations from src_path to target_path */
 int lr_replicate()
 {
@@ -1466,9 +1476,7 @@ int lr_replicate()
                         break;
                 }
                 if (rc && rc != -ENOENT) {
-                        fprintf(stderr, "Replication of operation %d, "
-                                "index %lld failed: %d\n",
-                                info->type, info->recno, rc);
+                        lr_print_failure(info, rc);
                         errors++;
                         if (abort_on_err)
                                 break;