b=20237

i=Manoj.Joseph
i=Robert.Read
gather and archive the logs

diff --git a/lustre/tests/cfg/local.sh b/lustre/tests/cfg/local.sh
index c4baef8..7b6436a 100644 (file)
--- a/lustre/tests/cfg/local.sh
+++ b/lustre/tests/cfg/local.sh
@@ -124,3 +124,5 @@ FAIL_ON_ERROR=${FAIL_ON_ERROR:-true}
 
 MPIRUN=$(which mpirun 2>/dev/null) || true
 MPI_USER=${MPI_USER:-mpiuser}
+SHARED_DIR_LOGS=${SHARED_DIR_LOGS:-""}
+

diff --git a/lustre/tests/recovery-mds-scale.sh b/lustre/tests/recovery-mds-scale.sh
index 1c7cc84..675988c 100644 (file)
--- a/lustre/tests/recovery-mds-scale.sh
+++ b/lustre/tests/recovery-mds-scale.sh
@@ -15,8 +15,11 @@ init_test_env $@
 
 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
 
-TESTSUITELOG=${TESTSUITELOG:-$TMP/recovery-mds-scale}
+TESTSUITELOG=${TESTSUITELOG:-$TMP/$(basename $0 .sh)}
 DEBUGLOG=$TESTSUITELOG.debug
+
+cleanup_logs
+
 exec 2>$DEBUGLOG
 echo "--- env ---" >&2
 env >&2
@@ -28,7 +31,7 @@ set -x
 
 [ -n "$CLIENTS" ] || { skip "$0 Need two or more remote clients" && exit 0; }
 [ $CLIENTCOUNT -ge 3 ] || \
-    { skip "$0 Need two or more clients, have $CLIENTCOUNT" && exit 0; }
+    { skip "$0 Need two or more remote clients, have $CLIENTCOUNT" && exit 0; }
 
 END_RUN_FILE=${END_RUN_FILE:-$SHARED_DIRECTORY/end_run_file}
 LOAD_PID_FILE=${LOAD_PID_FILE:-$TMP/client-load.pid}
@@ -145,6 +148,15 @@ Status: $result: rc=$rc"
         sleep 5
         kill -9 $CLIENT_LOAD_PIDS || true
     fi
+    if [ $rc -ne 0 ]; then
+        # we are interested in only on failed clients and servers
+        local failedclients=$(cat $END_RUN_FILE | grep -v $0)
+        # FIXME: need ostfailover-s nodes also for FLAVOR=OST
+        local product=$(gather_logs $(comma_list $(osts_nodes) \
+                               $(mdts_nodes) $mdsfailover_HOST $failedclients))
+        echo logs files $product
+    fi
+
     [ $rc -eq 0 ] && zconf_mount $(hostname) $MOUNT
 
     exit $rc

diff --git a/lustre/tests/test-framework.sh b/lustre/tests/test-framework.sh
index 7111182..9f0bac0 100644 (file)
--- a/lustre/tests/test-framework.sh
+++ b/lustre/tests/test-framework.sh
@@ -3014,3 +3014,45 @@ oos_full() {
         done
         return $OSCFULL
 }
+
+gather_logs () {
+    local list=$1
+
+    local ts=$(date +%s)
+
+    # bug 20237, comment 11
+    # It would also be useful to provide the option
+    # of writing the file to an NFS directory so it doesn't need to be copied.
+    local tmp=$TMP
+    local docp=true
+    [ -d "$SHARED_DIR_LOGS" ] && tmp=$SHARED_DIR_LOGS && docp=false
+
+    # dump lustre logs, dmesg
+    do_nodes $list "log=$tmp/\\\$(hostname)-debug-$ts.log ;
+lctl dk \\\$log >/dev/null;
+log=$tmp/\\\$(hostname)-dmesg-$ts.log;
+dmesg > \\\$log; "
+
+    # FIXME: does it make sense to collect the logs for $ts only, but all
+    # TESTSUITE logs?
+    # rsync $TMP/*${TESTSUITE}* to gather the logs dumped by error fn
+    local logs=$TMP/'*'${TESTSUITE}'*'
+    if $docp; then
+        logs=$logs' '$tmp/'*'$ts'*'
+    fi
+    for node in ${list//,/ }; do
+        rsync -az $node:"$logs" $TMP 
+    done
+
+    local archive=$TMP/${TESTSUITE}-$ts.tar.bz2
+    tar -jcf $archive $tmp/*$ts* $TMP/*${TESTSUITE}*
+
+    echo $archive
+}
+
+cleanup_logs () {
+    local list=${1:-$(comma_list $(nodes_list))}
+
+    [ -n ${TESTSUITE} ] && do_nodes $list "rm -f $TMP/*${TESTSUITE}*" || true
+}
+