From 3435b4da94cea33aa288f510660e1f2cb3380b94 Mon Sep 17 00:00:00 2001
From: ericm <ericm>
Date: Mon, 21 Apr 2008 17:11:38 +0000
Subject: [PATCH] branch: b1_6 change at_max according to AT status. b=13659
 r=nathan r=rread r=grev

---
 lustre/tests/recovery-small.sh | 12 ++++++-----
 lustre/tests/replay-single.sh  | 35 ++++++++++++++++++++++++-------
 lustre/tests/test-framework.sh | 47 ++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 82 insertions(+), 12 deletions(-)

diff --git a/lustre/tests/recovery-small.sh b/lustre/tests/recovery-small.sh
index f794e34..13b61ea 100755
--- a/lustre/tests/recovery-small.sh
+++ b/lustre/tests/recovery-small.sh
@@ -196,11 +196,13 @@ test_16() {
 run_test 16 "timeout bulk put, don't evict client (2732)"
 
 test_17() {
+    local at_max_saved=0
+
     # With adaptive timeouts, bulk_get won't expire until adaptive_timeout_max
-    local at_max=$(do_facet ost1 "find /sys/ -name at_max")
-    [ -z "$at_max" ] && skip "missing /sys/.../at_max" && return 0
-    OST_AT_MAX=$(do_facet ost1 "cat $at_max")
-    do_facet ost1 "echo $TIMEOUT >> $at_max"
+    if at_is_valid && at_is_enabled; then
+        at_max_saved=$(at_max_get ost1)
+        at_max_set $TIMEOUT ost1
+    fi
 
     # OBD_FAIL_PTLRPC_BULK_GET_NET 0x0503 | OBD_FAIL_ONCE
     # OST bulk will time out here, client retries
@@ -218,7 +220,7 @@ test_17() {
     # expect cmp to succeed, client resent bulk
     do_facet client "cmp $SAMPLE_FILE $DIR/$tfile" || return 3
     do_facet client "rm $DIR/$tfile" || return 4
-    do_facet ost1 "echo $OST_AT_MAX >> $at_max"
+    [ $at_max_saved -ne 0 ] && $(at_max_set $at_max_saved ost1)
     return 0
 }
 run_test 17 "timeout bulk get, don't evict client (2732)"
diff --git a/lustre/tests/replay-single.sh b/lustre/tests/replay-single.sh
index 6e1717d7..1ce5c2c 100755
--- a/lustre/tests/replay-single.sh
+++ b/lustre/tests/replay-single.sh
@@ -874,13 +874,17 @@ test_43() { # bug 2530
 run_test 43 "mds osc import failure during recovery; don't LBUG"
 
 test_44() {
+    local at_max_saved=0
+
     mdcdev=`awk '/-mdc-/ {print $1}' $LPROC/devices`
     [ "$mdcdev" ] || exit 2
+
     # adaptive timeouts slow this way down
-    local at_max=$(do_facet mds "find /sys/ -name at_max")
-    [ -z "$at_max" ] && skip "missing /sys/.../at_max" && return 0
-    MDS_AT_MAX=$(do_facet mds "cat $at_max")
-    do_facet mds "echo 40 >> $at_max"
+    if at_is_valid && at_is_enabled; then
+        at_max_saved=$(at_max_get mds)
+        at_max_set 40 mds
+    fi
+
     for i in `seq 1 10`; do
 	echo "$i of 10 ($(date +%s))"
 	do_facet mds "grep service $LPROC/mdt/MDS/mds/timeouts"
@@ -889,8 +893,9 @@ test_44() {
 	$LCTL --device $mdcdev recover
 	df $MOUNT
     done
+
     do_facet mds "sysctl -w lustre.fail_loc=0"
-    do_facet mds "echo $MDS_AT_MAX >> $at_max"
+    [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved mds
     return 0
 }
 run_test 44 "race in target handle connect"
@@ -1362,9 +1367,19 @@ test_61c() {
 }
 run_test 61c "test race mds llog sync vs llog cleanup"
 
-#Adaptive Timeouts
-at_start() #bug 3055
+#Adaptive Timeouts (bug 3055)
+AT_MAX_SET=0
+
+at_start()
 {
+    at_is_valid || skip "AT env is invalid"
+
+    if ! at_is_enabled; then
+        echo "AT is disabled, enable it by force temporarily"
+        at_max_set 600 mds ost client
+        AT_MAX_SET=1
+    fi
+
     if [ -z "$ATOLDBASE" ]; then
 	local at_history=$(do_facet mds "find /sys/ -name at_history")
 	[ -z "$at_history" ] && skip "missing /sys/.../at_history " && return 1
@@ -1540,6 +1555,12 @@ if [ -n "$ATOLDBASE" ]; then
     do_facet mds "echo $ATOLDBASE >> $at_history" || true
     do_facet ost1 "echo $ATOLDBASE >> $at_history" || true
 fi
+
+if [ $AT_MAX_SET -ne 0 ]; then
+    echo "restore AT status to be disabled"
+    at_max_set 0 mds ost client
+fi
+
 # end of AT tests includes above lines
 
 equals_msg `basename $0`: test complete, cleaning up
diff --git a/lustre/tests/test-framework.sh b/lustre/tests/test-framework.sh
index 3b7f839..fb824d3 100644
--- a/lustre/tests/test-framework.sh
+++ b/lustre/tests/test-framework.sh
@@ -88,6 +88,7 @@ init_test_env() {
     export NAME=${NAME:-local}
     export LPROC=/proc/fs/lustre
     export DIR2
+    export AT_MAX_PATH
 
     if [ "$ACCEPTOR_PORT" ]; then
         export PORT_OPT="--port $ACCEPTOR_PORT"
@@ -888,6 +889,52 @@ absolute_path() {
 }
 
 ##################################
+# Adaptive Timeouts funcs
+
+at_is_valid() {
+    if [ -z "$AT_MAX_PATH" ]; then
+        AT_MAX_PATH=$(do_facet mds "find /sys/ -name at_max")
+        [ -z "$AT_MAX_PATH" ] && echo "missing /sys/.../at_max " && return 1
+    fi
+    return 0
+}
+
+at_is_enabled() {
+    at_is_valid || error "invalid call"
+
+    # only check mds, we assume at_max is the same on all nodes
+    local at_max=$(do_facet mds "cat $AT_MAX_PATH")
+    if [ $at_max -eq 0 ]; then
+        return 1
+    else
+        return 0
+    fi
+}
+
+at_max_get() {
+    at_is_valid || error "invalid call"
+
+    do_facet $1 "cat $AT_MAX_PATH"
+}
+
+at_max_set() {
+    local at_max=$1
+    shift
+
+    at_is_valid || error "invalid call"
+
+    for facet in $@; do
+        if [ $facet == "ost" ]; then
+            for i in `seq $OSTCOUNT`; do
+                do_facet ost$i "echo $at_max > $AT_MAX_PATH"
+            done
+        else
+            do_facet $facet "echo $at_max > $AT_MAX_PATH"
+        fi
+    done
+}
+
+##################################
 # OBD_FAIL funcs
 
 drop_request() {
-- 
1.8.3.1