b=16437

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

diff --git a/lustre/tests/insanity.sh b/lustre/tests/insanity.sh
index 54ae83a..f4b7754 100755 (executable)
--- a/lustre/tests/insanity.sh
+++ b/lustre/tests/insanity.sh
@@ -8,22 +8,36 @@ LUSTRE=${LUSTRE:-`dirname $0`/..}
 
 init_test_env $@
 
-. ${CONFIG:=$LUSTRE/tests/cfg/insanity-local.sh}
+. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
+#              
+ALWAYS_EXCEPT="10  $INSANITY_EXCEPT"
+
+if [ "$FAILURE_MODE" = "HARD" ]; then
+    mixed_ost_devs && CONFIG_EXCEPTIONS="0 2 4 5 6 8" && \
+        echo -n "Several ost services on one ost node are used with FAILURE_MODE=$FAILURE_MODE. " && \
+        echo "Except the tests: $CONFIG_EXCEPTIONS" && \
+        ALWAYS_EXCEPT="$ALWAYS_EXCEPT $CONFIG_EXCEPTIONS"
+fi
 
-ALWAYS_EXCEPT="10 $INSANITY_EXCEPT"
+#
+[ "$SLOW" = "no" ] && EXCEPT_SLOW=""
 
-SETUP=${SETUP:-"setup"}
-CLEANUP=${CLEANUP:-"cleanup"}
+SETUP=${SETUP:-""}
+CLEANUP=${CLEANUP:-""}
 
 build_test_filter
 
+SINGLECLIENT=${SINGLECLIENT:-$HOSTNAME}
+LIVE_CLIENT=${LIVE_CLIENT:-$SINGLECLIENT}
+FAIL_CLIENTS=${FAIL_CLIENTS:-$RCLIENTS}
+
 assert_env mds_HOST MDS_MKFS_OPTS
 assert_env ost_HOST OST_MKFS_OPTS OSTCOUNT
 assert_env LIVE_CLIENT FSNAME
 
 
 # This can be a regexp, to allow more clients
-CLIENTS=${CLIENTS:-"`comma_list $LIVE_CLIENT $FAIL_CLIENTS $EXTRA_CLIENTS`"}
+CLIENTS=${CLIENTS:-"`comma_list $LIVE_CLIENT $FAIL_CLIENTS`"}
 
 DIR=${DIR:-$MOUNT}
 
@@ -66,11 +80,14 @@ reboot_node() {
 
 fail_clients() {
     num=$1
+
+    log "Request clients to fail: ${num}. Num of clients to fail: ${FAIL_NUM}, already failed: $DOWN_NUM"
     if [ -z "$num"  ] || [ "$num" -gt $((FAIL_NUM - DOWN_NUM)) ]; then
        num=$((FAIL_NUM - DOWN_NUM)) 
     fi
     
     if [ -z "$num" ] || [ "$num" -le 0 ]; then
+        log "No clients failed!"
         return
     fi
 
@@ -106,22 +123,6 @@ start_ost() {
     start ost$1 `ostdevname $1` $OST_MOUNT_OPTS
 }
 
-setup() {
-    cleanup
-    rm -rf logs/*
-    formatall
-    setupall
-
-    while ! do_node $CLIENTS "ls -d $LUSTRE" > /dev/null; do sleep 5; done
-    grep " $MOUNT " /proc/mounts || zconf_mount $CLIENTS $MOUNT
-}
-
-cleanup() {
-    zconf_umount $CLIENTS $MOUNT
-    cleanupall
-    cleanup_check
-}
-
 trap exit INT
 
 client_touch() {
@@ -159,21 +160,7 @@ clients_recover_osts() {
 #    do_node $CLIENTS "$LCTL "'--device %OSC_`hostname`_'"${facet}_svc_MNT_client_facet recover"
 }
 
-if [ "$ONLY" == "cleanup" ]; then
-    $CLEANUP
-    exit
-fi
-
-if [ ! -z "$EVAL" ]; then
-    eval "$EVAL"
-    exit $?
-fi
-
-$SETUP
-
-if [ "$ONLY" == "setup" ]; then
-    exit 0
-fi
+cleanup_and_setup_lustre
 
 # 9 Different Failure Modes Combinations
 echo "Starting Test 17 at `date`"
@@ -183,13 +170,11 @@ test_0() {
     echo "Waiting for df pid: $DFPID"
     wait $DFPID || { echo "df returned $?" && return 1; }
 
-    facet_failover ost1 || return 4
-    echo "Waiting for df pid: $DFPID"
-    wait $DFPID || { echo "df returned $?" && return 2; }
-
-    facet_failover ost2 || return 5
-    echo "Waiting for df pid: $DFPID"
-    wait $DFPID || { echo "df returned $?" && return 3; }
+    for i in $(seq $OSTCOUNT) ; do
+        facet_failover ost$i || return 4
+        echo "Waiting for df pid: $DFPID"
+        wait $DFPID || { echo "df returned $?" && return 3; }
+    done
     return 0
 }
 run_test 0 "Fail all nodes, independently"
@@ -205,6 +190,8 @@ run_test 1 "MDS/MDS failure"
 ############### Second Failure Mode ###############
 test_2() {
     echo "Verify Lustre filesystem is up and running"
+    [ -z "$(mounted_lustre_filesystems)" ] && error "Lustre is not running"
+
     client_df
 
     shutdown_facet $SINGLEMDS
@@ -243,6 +230,7 @@ run_test 2 "Second Failure Mode: MDS/OST `date`"
 test_3() {
     #Create files
     echo "Verify Lustre filesystem is up and running"
+    [ -z "$(mounted_lustre_filesystems)" ] && error "Lustre is not running"
     
     #MDS Portion
     facet_failover $SINGLEMDS
@@ -316,10 +304,14 @@ 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
+
     echo "Fifth Failure Mode: OST/OST `date`"
 
     #Create files
     echo "Verify Lustre filesystem is up and running"
+    [ -z "$(mounted_lustre_filesystems)" ] && error "Lustre is not running"
+
     client_df
     
     #OST Portion
@@ -366,7 +358,9 @@ test_6() {
 
     #Create files
     echo "Verify Lustre filesystem is up and running"
-    client_df || return 1
+    [ -z "$(mounted_lustre_filesystems)" ] && error "Lustre is not running"
+
+    client_df
     client_touch testfile || return 2
        
     #OST Portion
@@ -399,7 +393,8 @@ test_6() {
     wait $DFPIDA
     wait $DFPIDB
     echo "Verifying mount"
-    client_df || return 3
+    [ -z "$(mounted_lustre_filesystems)" ] && return 3
+    client_df
 }
 run_test 6 "Sixth Failure Mode: OST/CLIENT `date`"
 ###################################################
@@ -411,6 +406,8 @@ test_7() {
 
     #Create files
     echo "Verify Lustre filesystem is up and running"
+    [ -z "$(mounted_lustre_filesystems)" ] && error "Lustre is not running"
+
     client_df
     client_touch testfile  || return 1
 
@@ -430,6 +427,8 @@ test_7() {
 
     #Create files
     echo "Verify Lustre filesystem is up and running"
+    [ -z "$(mounted_lustre_filesystems)" ] && return 2
+
     client_df
     client_rm testfile
 
@@ -444,8 +443,8 @@ test_7() {
 
     #Reintegration
     echo "Reintegrating CLIENTs"
-    reintegrate_clients
-    client_df || return 2
+    reintegrate_clients || return 2
+    client_df
     
     #Sleep
     echo "wait 1 minutes"
@@ -461,6 +460,8 @@ test_8() {
 
     #Create files
     echo "Verify Lustre filesystem is up and running"
+    [ -z "$(mounted_lustre_filesystems)" ] && error "Lustre is not running"
+
     client_df
     client_touch testfile
        
@@ -480,6 +481,8 @@ test_8() {
 
     #Create files
     echo "Verify Lustre filesystem is up and running"
+    [ -z "$(mounted_lustre_filesystems)" ] && error "Lustre is not running"
+
     client_df
     client_touch testfile
 
@@ -499,7 +502,7 @@ test_8() {
     
     #Reintegration
     echo "Reintegrating CLIENTs/OST"
-    reintegrate_clients
+    reintegrate_clients || return 3
     wait_for ost1
     start_ost 1
     wait $DFPID
@@ -520,6 +523,8 @@ test_9() {
 
     #Create files
     echo "Verify Lustre filesystem is up and running"
+    [ -z "$(mounted_lustre_filesystems)" ] && error "Lustre is not running"
+
     client_df
     client_touch testfile || return 1
        
@@ -554,8 +559,8 @@ test_9() {
 
     #Reintegration
     echo "Reintegrating  CLIENTs/CLIENTs"
-    reintegrate_clients
-    client_df || return 7
+    reintegrate_clients || return 7
+    client_df
     
     #Sleep
     echo "Wait 1 minutes"
@@ -573,5 +578,5 @@ test_10() {
 run_test 10 "Running Availability for 6 hours..."
 
 equals_msg `basename $0`: test complete, cleaning up
-$CLEANUP
+check_and_cleanup_lustre
 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true