b=17873

[fs/lustre-release.git] / lustre / tests / insanity.sh

diff --git a/lustre/tests/insanity.sh b/lustre/tests/insanity.sh
index 06b5ca3..690322c 100755 (executable)
--- a/lustre/tests/insanity.sh
+++ b/lustre/tests/insanity.sh
@@ -42,6 +42,7 @@ remote_ost_nodsh && skip "remote OST with nodsh" && exit 0
 FAIL_CLIENTS=$(echo " $FAIL_CLIENTS " | sed -re "s/\s+$LIVE_CLIENT\s+/ /g")
 
 DIR=${DIR:-$MOUNT}
+TESTDIR=$DIR/d0.$(basename $0 .sh)
 
 #####
 # fail clients round robin
@@ -111,29 +112,28 @@ client_touch() {
     file=$1
     for c in $LIVE_CLIENT $FAIL_CLIENTS;  do
        if echo $DOWN_CLIENTS | grep -q $c; then continue; fi
-       $PDSH $c touch $MOUNT/${c}_$file || return 1
+       $PDSH $c touch $TESTDIR/${c}_$file || return 1
     done
 }
 
 client_rm() {
     file=$1
     for c in $LIVE_CLIENT $FAIL_CLIENTS;  do
-       $PDSH $c rm $MOUNT/${c}_$file
+       $PDSH $c rm $TESTDIR/${c}_$file
     done
 }
 
 client_mkdirs() {
     for c in $LIVE_CLIENT $FAIL_CLIENTS;  do
-       echo "$c mkdir $MOUNT/$c"
-       $PDSH $c "mkdir $MOUNT/$c"
-       $PDSH $c "ls -l $MOUNT/$c" 
+       echo "$c mkdir $TESTDIR/$c"
+       $PDSH $c "mkdir $TESTDIR/$c && ls -l $TESTDIR/$c"
     done
 }
 
 client_rmdirs() {
     for c in $LIVE_CLIENT $FAIL_CLIENTS;  do
-       echo "rmdir $MOUNT/$c"
-       $PDSH $LIVE_CLIENT "rmdir $MOUNT/$c"
+       echo "rmdir $TESTDIR/$c"
+       $PDSH $LIVE_CLIENT "rmdir $TESTDIR/$c"
     done
 }
 
@@ -144,18 +144,17 @@ clients_recover_osts() {
 
 check_and_setup_lustre
 
+rm -rf $TESTDIR
+mkdir -p $TESTDIR
+
 # 9 Different Failure Modes Combinations
 echo "Starting Test 17 at `date`"
 
 test_0() {
-    facet_failover $SINGLEMDS
-    echo "Waiting for df pid: $DFPID"
-    wait $DFPID || { echo "df returned $?" && return 1; }
+    fail $SINGLEMDS
 
     for i in $(seq $OSTCOUNT) ; do
-        facet_failover ost$i || return 4
-        echo "Waiting for df pid: $DFPID"
-        wait $DFPID || { echo "df returned $?" && return 3; }
+        fail ost$i
     done
     return 0
 }
@@ -215,8 +214,7 @@ test_3() {
     [ -z "$(mounted_lustre_filesystems)" ] && error "Lustre is not running"
     
     #MDS Portion
-    facet_failover $SINGLEMDS
-    wait $DFPID || echo df failed: $?
+    fail $SINGLEMDS
     #Check FS
 
     echo "Test Lustre stability after MDS failover"
@@ -286,7 +284,7 @@ run_test 4 "Fourth Failure Mode: OST/MDS `date`"
 
 ############### Fifth Failure Mode ###############
 test_5() {
-    [ $OSTCOUNT -lt 2 ] && skip "$OSTCOUNT < 2, not enough OSTs" && return 0
+    [ $OSTCOUNT -lt 2 ] && skip_env "$OSTCOUNT < 2, not enough OSTs" && return 0
 
     echo "Fifth Failure Mode: OST/OST `date`"
 
@@ -404,8 +402,8 @@ test_7() {
     #Check FS
     echo "Test Lustre stability after CLIENTs failure"
     client_df
-    $PDSH $LIVE_CLIENT "ls -l $MOUNT"
-    $PDSH $LIVE_CLIENT "rm -f $MOUNT/*_testfile"
+    $PDSH $LIVE_CLIENT "ls -l $TESTDIR"
+    $PDSH $LIVE_CLIENT "rm -f $TESTDIR/*_testfile"
     
     #Sleep
     echo "Wait 1 minutes"
@@ -419,13 +417,10 @@ test_7() {
     client_rm testfile
 
     #MDS Portion
-    facet_failover $SINGLEMDS
+    fail $SINGLEMDS
 
-    #Check FS
-    echo "Test Lustre stability after MDS failover"
-    wait $DFPID || echo "df on down clients fails " || return 1
-    $PDSH $LIVE_CLIENT "ls -l $MOUNT"
-    $PDSH $LIVE_CLIENT "rm -f $MOUNT/*_testfile"
+    $PDSH $LIVE_CLIENT "ls -l $TESTDIR"
+    $PDSH $LIVE_CLIENT "rm -f $TESTDIR/*_testfile"
 
     #Reintegration
     echo "Reintegrating CLIENTs"
@@ -458,8 +453,8 @@ test_8() {
     #Check FS
     echo "Test Lustre stability after CLIENTs failure"
     client_df
-    $PDSH $LIVE_CLIENT "ls -l $MOUNT"
-    $PDSH $LIVE_CLIENT "rm -f $MOUNT/*_testfile"
+    $PDSH $LIVE_CLIENT "ls -l $TESTDIR"
+    $PDSH $LIVE_CLIENT "rm -f $TESTDIR/*_testfile"
 
     #Sleep
     echo "Wait 1 minutes"
@@ -483,8 +478,8 @@ test_8() {
     DFPID=$!
     sleep 5
     #non-failout hangs forever here
-    #$PDSH $LIVE_CLIENT "ls -l $MOUNT"
-    #$PDSH $LIVE_CLIENT "rm -f $MOUNT/*_testfile"
+    #$PDSH $LIVE_CLIENT "ls -l $TESTDIR"
+    #$PDSH $LIVE_CLIENT "rm -f $TESTDIR/*_testfile"
     
     #Reintegration
     echo "Reintegrating CLIENTs/OST"
@@ -521,8 +516,8 @@ test_9() {
     #Check FS
     echo "Test Lustre stability after CLIENTs failure"
     client_df
-    $PDSH $LIVE_CLIENT "ls -l $MOUNT" || return 1
-    $PDSH $LIVE_CLIENT "rm -f $MOUNT/*_testfile" || return 2
+    $PDSH $LIVE_CLIENT "ls -l $TESTDIR" || return 1
+    $PDSH $LIVE_CLIENT "rm -f $TESTDIR/*_testfile" || return 2
 
     #Sleep
     echo "Wait 1 minutes"
@@ -540,8 +535,8 @@ test_9() {
     #Check FS
     echo "Test Lustre stability after CLIENTs failure"
     client_df
-    $PDSH $LIVE_CLIENT "ls -l $MOUNT" || return 5
-    $PDSH $LIVE_CLIENT "rm -f $MOUNT/*_testfile" || return 6
+    $PDSH $LIVE_CLIENT "ls -l $TESTDIR" || return 5
+    $PDSH $LIVE_CLIENT "rm -f $TESTDIR/*_testfile" || return 6
 
     #Reintegration
     echo "Reintegrating  CLIENTs/CLIENTs"