From 52828321b9100778e982bbe4b967732b9951d879 Mon Sep 17 00:00:00 2001
From: grev <grev>
Date: Sat, 4 Apr 2009 02:05:17 +0000
Subject: [PATCH] b=18914 i=Brian check_client_load: try to connect several
 times to avoid ENFILE

---
 lustre/tests/test-framework.sh | 32 ++++++++++++++++++++++++++++----
 1 file changed, 28 insertions(+), 4 deletions(-)

diff --git a/lustre/tests/test-framework.sh b/lustre/tests/test-framework.sh
index 84050a4..e9d8965 100644
--- a/lustre/tests/test-framework.sh
+++ b/lustre/tests/test-framework.sh
@@ -699,12 +699,36 @@ check_client_load () {
     local TESTLOAD=run_${!var}.sh
 
     ps auxww | grep -v grep | grep $client | grep -q "$TESTLOAD" || return 1
+    
+    # bug 18914: try to connect several times not only when
+    # check ps, but  while check_catastrophe also
+    local tries=3
+    local RC=254
+    while [ $RC = 254 -a $tries -gt 0 ]; do
+        let tries=$tries-1
+        # assume success
+        RC=0
+        if ! check_catastrophe $client; then
+            RC=${PIPESTATUS[0]}
+            if [ $RC -eq 254 ]; then
+                # FIXME: not sure how long we shuold sleep here
+                sleep 10
+                continue
+            fi
+            echo "check catastrophe failed: RC=$RC "
+            return $RC
+        fi
+    done
 
-    check_catastrophe $client || return 2
+    # We can continue try to connect if RC=254
+    # Just print the warning about this
+    if [ $RC = 254 ]; then
+        echo "got a return status of $RC from do_node while checking catastrophe on $client"
+    fi
 
     # see if the load is still on the client
-    local tries=3
-    local RC=254
+    tries=3
+    RC=254
     while [ $RC = 254 -a $tries -gt 0 ]; do
         let tries=$tries-1
         # assume success
@@ -715,7 +739,7 @@ check_client_load () {
         fi
     done
     if [ $RC = 254 ]; then
-        echo "got a return status of $RC from do_node while checking (i.e. with 'ps') the client load on the remote system"
+        echo "got a return status of $RC from do_node while checking (catastrophe and 'ps') the client load on $client"
         # see if we can diagnose a bit why this is
     fi
 
-- 
1.8.3.1