b=22423 add regression test for reconnect flooding issue

i=dmitry

diff --git a/lustre/tests/conf-sanity.sh b/lustre/tests/conf-sanity.sh
index 69b1e05..40b4551 100644 (file)
--- a/lustre/tests/conf-sanity.sh
+++ b/lustre/tests/conf-sanity.sh
@@ -1313,8 +1313,11 @@ test_35b() { # bug 18674
        # Setting OBD_FAIL_MDS_RESEND=0x136
        do_facet mds "$LCTL set_param fail_loc=0x80000136" || return 2
 
-       log "Creating a test file"
+       $LCTL set_param mdc.${FSNAME}*.stats=clear
+
+       log "Creating a test file and stat it"
        touch $MOUNT/$tdir/$tfile
+       stat $MOUNT/$tdir/$tfile
 
        log "Stop injecting EBUSY on MDS"
        do_facet mds "$LCTL set_param fail_loc=0" || return 3
@@ -1326,6 +1329,8 @@ test_35b() { # bug 18674
 
        $LCTL dk $TMP/lustre-log-$TESTNAME.log
 
+       CONNCNT=`$LCTL get_param mdc.${FSNAME}*.stats | awk '/mds_connect/{print $2}'`
+
        # retrieve from the log if the client has ever tried to
        # contact the fake server after the loss of connection
        FAILCONN=`awk "BEGIN {ret = 0;}
@@ -1345,6 +1350,12 @@ test_35b() { # bug 18674
                log "ERROR: The client tried to reconnect to the failover server while the primary was busy" && \
                return 5
 
+       # When OBD_FAIL_MDS_RESEND is hit, we sleep for 2 * obd_timeout
+        # Reconnects are supposed to be rate limited to one every 5s
+       [ $CONNCNT -gt $((2 * $TIMEOUT / 5 + 1)) ] && \
+               log "ERROR: Too many reconnects $CONNCNT" && \
+               return 6
+
         cleanup
 }
 run_test 35b "Continue reconnection retries, if the active server is busy"