Whamcloud - gitweb
LU-3326 tests: limit the disk space used by run_*.sh
[fs/lustre-release.git] / lustre / tests / run_tar.sh
index 7502c24..823c781 100755 (executable)
@@ -1,68 +1,73 @@
 #!/bin/bash
-set -x
 
 TMP=${TMP:-/tmp}
 
-TESTSUITELOG=${TESTSUITELOG:-$TMP/recovery-mds-scale}
-LOG=${TESTSUITELOG}_$(basename $0)-$(hostname)
-DEBUGLOG=${LOG}.debug
+TESTLOG_PREFIX=${TESTLOG_PREFIX:-$TMP/recovery-mds-scale}
+TESTNAME=${TESTNAME:-""}
+[ -n "$TESTNAME" ] && TESTLOG_PREFIX=$TESTLOG_PREFIX.$TESTNAME
+
+LOG=$TESTLOG_PREFIX.$(basename $0 .sh)_stdout.$(hostname -s).log
+DEBUGLOG=$(echo $LOG | sed 's/\(.*\)stdout/\1debug/')
 
 mkdir -p ${LOG%/*}
 
 rm -f $LOG $DEBUGLOG
 exec 2>$DEBUGLOG
+set -x
 
-if [ -z "$MOUNT" -o -z "$END_RUN_FILE" -o -z "$LOAD_PID_FILE" ]; then
-    echo "The following must be set: MOUNT END_RUN_FILE LOAD_PID_FILE"
-    exit 1
-fi
-
-echoerr () { echo "$@" 1>&2 ; }
+. $(dirname $0)/functions.sh
 
-signaled() {
-    echoerr "$(date +'%F %H:%M:%S'): client load was signaled to terminate"
-    kill -TERM -$PPID
-    sleep 5
-    kill -KILL -$PPID
-}
+assert_env MOUNT END_RUN_FILE LOAD_PID_FILE LFS CLIENT_COUNT
 
 trap signaled TERM
 
-# recovery-mds-scale uses this to signal the client loads to die
+# recovery-*-scale scripts use this to signal the client loads to die
 echo $$ >$LOAD_PID_FILE
 
-TESTDIR=$MOUNT/tar-$(hostname)
+TESTDIR=$MOUNT/d0.tar-$(hostname)
+
+do_tar() {
+    tar cf - /etc | tar xf - >$LOG 2>&1
+    return ${PIPESTATUS[1]}
+}
 
 CONTINUE=true
 while [ ! -e "$END_RUN_FILE" ] && $CONTINUE; do
-    echoerr "$(date +'%F %H:%M:%S'): tar run starting"
-    mkdir -p $TESTDIR
-    cd $TESTDIR
-    tar cf - /etc | tar xf - 2>&1 | tee $LOG &
-    load_pid=$!
-ps -e f -o "pid ppid pgrp comm" >$TMP/client-load.ps-list
-    wait $load_pid
-    RC=${PIPESTATUS[0]}
-    PREV_ERRORS=$(grep "exit delayed from previous errors" $LOG) || true
-    if [ $RC -ne 0 -a "$ERRORS_OK" -a "$PREV_ERRORS" ]; then
-        echoerr "$(date +'%F %H:%M:%S'): tar errors earlier, ignoring"
-        RC=0
-    fi
-    if [ $RC -eq 0 ]; then
-       echoerr "$(date +'%F %H:%M:%S'): tar succeeded"
-       cd $TMP
-       rm -rf $TESTDIR
-       echoerr "$(date +'%F %H:%M:%S'): tar run finished"
-    else
-       echoerr "$(date +'%F %H:%M:%S'): tar failed"
-       if [ -z "$ERRORS_OK" ]; then
-           echo $(hostname) >> $END_RUN_FILE
+       echoerr "$(date +'%F %H:%M:%S'): tar run starting"
+       mkdir -p $TESTDIR
+       USAGE=$(du -s /etc | awk '{print $1}')
+       FREE_SPACE=$($LFS df $TESTDIR | awk '/filesystem summary:/ {print $5}')
+       AVAIL=$((FREE_SPACE * 9 / 10 / CLIENT_COUNT))
+       if [ $AVAIL -lt $USAGE ]; then
+               echoerr "no enough free disk space: need $USAGE, avail $AVAIL"
+               echo $(hostname) >> $END_RUN_FILE
+               break
+       fi
+
+       cd $TESTDIR
+       do_tar &
+       wait $!
+       RC=$?
+       PREV_ERRORS=$(grep "exit delayed from previous errors" $LOG) || true
+       if [ $RC -ne 0 -a "$ERRORS_OK" -a "$PREV_ERRORS" ]; then
+               echoerr "$(date +'%F %H:%M:%S'): tar errors earlier, ignoring"
+               RC=0
        fi
-       if [ $BREAK_ON_ERROR ]; then
-           # break
-            CONTINUE=false
+       if [ $RC -eq 0 ]; then
+               echoerr "$(date +'%F %H:%M:%S'): tar succeeded"
+               cd $TMP
+               rm -rf $TESTDIR
+               echoerr "$(date +'%F %H:%M:%S'): tar run finished"
+       else
+               echoerr "$(date +'%F %H:%M:%S'): tar failed"
+               if [ -z "$ERRORS_OK" ]; then
+                       echo $(hostname) >> $END_RUN_FILE
+               fi
+               if [ $BREAK_ON_ERROR ]; then
+                       # break
+                       CONTINUE=false
+               fi
        fi
-    fi
 done
 
 echoerr "$(date +'%F %H:%M:%S'): tar run exiting"