b=21582 make sure not to use cached statfs info in fail()

Use client_up which sleeps for 1s before running statfs.
It grants that we are not using cached statfs info, since patch
from bug 20065 has been landed.

i=grev

diff --git a/lustre/tests/test-framework.sh b/lustre/tests/test-framework.sh
index 24e1632..b326bc5 100644 (file)
--- a/lustre/tests/test-framework.sh
+++ b/lustre/tests/test-framework.sh
@@ -1126,7 +1126,7 @@ ost_evict_client() {
 
 fail() {
     facet_failover $* || error "failover: $?"
-    df $MOUNT || error "post-failover df: $?"
+    clients_up || error "post-failover df: $?"
 }
 
 fail_nodf() {
@@ -1139,9 +1139,8 @@ fail_abort() {
     stop $facet
     change_active $facet
     mount_facet $facet -o abort_recovery
-    df $MOUNT || echo "first df failed: $?"
-    sleep 1
-    df $MOUNT || error "post-failover df: $?"
+    clients_up || echo "first df failed: $?"
+    clients_up || error "post-failover df: $?"
 }
 
 do_lmc() {