From 67e941432daff9ede8dc8246334d26bea649e658 Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Tue, 19 Sep 2006 18:56:42 +0000
Subject: [PATCH] Replace a bunch of kill_srun calls with slow_kill $pid calls.
---
testsuite/expect/test1.1 | 1 -
testsuite/expect/test1.10 | 1 -
testsuite/expect/test1.11 | 2 --
testsuite/expect/test1.12 | 1 -
testsuite/expect/test1.13 | 2 --
testsuite/expect/test1.14 | 1 -
testsuite/expect/test1.15 | 1 -
testsuite/expect/test1.16 | 5 ++---
testsuite/expect/test1.17 | 5 ++---
testsuite/expect/test1.18 | 8 ++++----
testsuite/expect/test1.19 | 20 ++++++++++----------
testsuite/expect/test1.2 | 1 -
testsuite/expect/test1.20 | 10 ++++------
testsuite/expect/test1.21 | 5 ++---
testsuite/expect/test1.22 | 10 ++++------
testsuite/expect/test1.23 | 30 ++++++++++++------------------
testsuite/expect/test1.24 | 5 ++---
testsuite/expect/test1.25 | 5 ++---
testsuite/expect/test1.26 | 34 +++++++++++++---------------------
testsuite/expect/test1.27 | 5 ++---
testsuite/expect/test1.28 | 5 ++---
testsuite/expect/test1.29 | 5 ++---
testsuite/expect/test1.3 | 1 -
testsuite/expect/test1.30 | 7 +++----
testsuite/expect/test1.31 | 5 ++---
testsuite/expect/test1.32 | 8 ++------
testsuite/expect/test1.33 | 8 ++------
testsuite/expect/test1.34 | 6 +++---
testsuite/expect/test1.35 | 9 +++------
testsuite/expect/test1.36 | 5 ++---
testsuite/expect/test1.37 | 13 ++++++-------
testsuite/expect/test1.38 | 14 ++++----------
testsuite/expect/test1.39 | 12 ++++++++----
testsuite/expect/test1.4 | 1 -
testsuite/expect/test1.5 | 1 -
testsuite/expect/test1.6 | 1 -
testsuite/expect/test1.7 | 20 ++++++++------------
testsuite/expect/test1.8 | 10 ++++------
testsuite/expect/test1.9 | 5 ++---
testsuite/expect/test15.10 | 6 ++----
40 files changed, 114 insertions(+), 180 deletions(-)
diff --git a/testsuite/expect/test1.1 b/testsuite/expect/test1.1
index c4dc9727633..b71a9ae45fe 100755
--- a/testsuite/expect/test1.1
+++ b/testsuite/expect/test1.1
@@ -81,7 +81,6 @@ expect {
send_user "\nFAILURE: srun not responding\n"
slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.10 b/testsuite/expect/test1.10
index 2eb02babb4e..1e7cfb4c47b 100755
--- a/testsuite/expect/test1.10
+++ b/testsuite/expect/test1.10
@@ -54,7 +54,6 @@ expect {
send_user "\nFAILURE: srun not responding\n"
slow_kill $srun_pid
set exit_code 1
- exp_continue
}
}
diff --git a/testsuite/expect/test1.11 b/testsuite/expect/test1.11
index 5449706987d..5bc44158016 100755
--- a/testsuite/expect/test1.11
+++ b/testsuite/expect/test1.11
@@ -78,7 +78,6 @@ expect {
send_user "\nFAILURE: srun not responding\n"
slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
@@ -189,7 +188,6 @@ expect {
send_user "\nFAILURE: srun not responding\n"
slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.12 b/testsuite/expect/test1.12
index efe66e7bab1..200c1827209 100755
--- a/testsuite/expect/test1.12
+++ b/testsuite/expect/test1.12
@@ -68,7 +68,6 @@ expect {
send_user "\nFAILURE: srun not responding\n"
slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.13 b/testsuite/expect/test1.13
index 9bf489e21a7..ee38994e831 100755
--- a/testsuite/expect/test1.13
+++ b/testsuite/expect/test1.13
@@ -53,7 +53,6 @@ expect {
send_user "\nFAILURE: srun not responding\n"
slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
@@ -85,7 +84,6 @@ expect {
send_user "\nFAILURE: srun not responding\n"
slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.14 b/testsuite/expect/test1.14
index 56f6a86a6ef..fa81dbdeaee 100755
--- a/testsuite/expect/test1.14
+++ b/testsuite/expect/test1.14
@@ -52,7 +52,6 @@ expect {
send_user "\nFAILURE: srun not responding\n"
slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.15 b/testsuite/expect/test1.15
index c49e9292663..6257c34b691 100755
--- a/testsuite/expect/test1.15
+++ b/testsuite/expect/test1.15
@@ -69,7 +69,6 @@ expect {
send_user "\nFAILURE: srun not responding\n"
slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.16 b/testsuite/expect/test1.16
index fb5f06b4a45..e706b9d89d9 100755
--- a/testsuite/expect/test1.16
+++ b/testsuite/expect/test1.16
@@ -49,7 +49,7 @@ exec $bin_touch $scratch_file
# The --unbuffered option will send the message which lacks a '\n'
#
set timeout $max_job_delay
-spawn $srun --unbuffered --verbose -t1 $bin_rm -i $scratch_file
+set srun_pid [spawn $srun --unbuffered --verbose -t1 $bin_rm -i $scratch_file]
expect {
-re "launching ($number).0 on host" {
set job_id $expect_out(1,string)
@@ -62,9 +62,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.17 b/testsuite/expect/test1.17
index ab8d10783a0..b5689a0bdeb 100755
--- a/testsuite/expect/test1.17
+++ b/testsuite/expect/test1.17
@@ -44,7 +44,7 @@ print_header $test_id
#
set timeout $max_job_delay
set match 0
-spawn $srun --allocate --verbose -t1
+set srun_pid [spawn $srun --allocate --verbose -t1]
expect {
-re "jobid ($number).*" {
set job_id $expect_out(1,string)
@@ -68,9 +68,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.18 b/testsuite/expect/test1.18
index ce22cf2a27e..bb769e4b4c9 100755
--- a/testsuite/expect/test1.18
+++ b/testsuite/expect/test1.18
@@ -53,7 +53,7 @@ exec $bin_chmod 700 $file_prog
# Spawn initial program via srun
#
set timeout $max_job_delay
-spawn $srun -N1-4 -v -t5 $file_prog
+set srun_pid [spawn $srun -N1-4 -v -t5 -l $file_prog]
set init_id $spawn_id
expect {
-i $init_id
@@ -66,6 +66,7 @@ expect {
}
timeout {
send_user "\nFAILURE: srun (launch) not responding\n"
+ slow_kill $srun_pid
set exit_code 1
}
eof {
@@ -86,7 +87,7 @@ if {$matches == 0} {
#
set matches 0
set timeout 10
-spawn $srun -vv --attach=$job_id --join
+set attach_pid [spawn $srun -vv -l --attach=$job_id --join]
set attach_id $spawn_id
expect {
-i $attach_id
@@ -96,9 +97,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun (attach) not responding\n"
- kill_srun
+ slow_kill $attach_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.19 b/testsuite/expect/test1.19
index cb74d07fc16..7e396042987 100755
--- a/testsuite/expect/test1.19
+++ b/testsuite/expect/test1.19
@@ -54,7 +54,7 @@ for {set task_id 0} {$task_id < $task_cnt} {incr task_id} {
exec $bin_rm -f $file_out_t_glob
}
set timeout $max_job_delay
-spawn $srun --output=$file_out_t -N1 -n$task_cnt -O -v -t1 $bin_id
+set srun_pid [spawn $srun --output=$file_out_t -N1 -n$task_cnt -O -v -t1 $bin_id]
expect {
-re "jobid ($number).*" {
set job_id $expect_out(1,string)
@@ -62,7 +62,7 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
exit 1
}
eof {
@@ -94,7 +94,7 @@ if {$file_cnt != $task_cnt} {
#
set job_id 0
set srun_exit 0
-spawn $srun --error=$file_err_j --output=/dev/null -N1 -n$task_cnt -O -v -t1 $bin_sleep aaa
+set srun_pid [spawn $srun --error=$file_err_j --output=/dev/null -N1 -n$task_cnt -O -v -t1 $bin_sleep aaa]
expect {
-re "jobid ($number).*" {
set job_id $expect_out(1,string)
@@ -107,7 +107,7 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
exit 1
}
eof {
@@ -136,7 +136,7 @@ if {[wait_for_file $file_err_j_glob] == 0} {
# file name and confirm it is created
#
set job_id 0
-spawn $srun --output=$file_out_J -N1 -v -t1 $bin_hostname
+set srun_pid [spawn $srun --output=$file_out_J -N1 -v -t1 $bin_hostname]
expect {
-re "jobid ($number).*" {
set job_id $expect_out(1,string)
@@ -144,7 +144,7 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
exit 1
}
eof {
@@ -173,7 +173,7 @@ set file_out_n_glob "test$test_id.n.$node_id.output"
exec $bin_rm -f $file_out_n_glob
set job_id 0
-spawn $srun --output=$file_out_n -N1 -n2 -O -v -t1 $bin_hostname
+set srun_pid [spawn $srun --output=$file_out_n -N1 -n2 -O -v -t1 $bin_hostname]
expect {
-re "jobid ($number).*" {
set job_id $expect_out(1,string)
@@ -181,7 +181,7 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
exit 1
}
eof {
@@ -241,7 +241,7 @@ if { [test_bluegene] } {
}
set job_id 0
-spawn $srun --batch --output=/dev/null -N$node_cnt -n$task_cnt -O -t1 $file_in
+set srun_pid [spawn $srun --batch --output=/dev/null -N$node_cnt -n$task_cnt -O -t1 $file_in]
expect {
-re "jobid ($number) submitted" {
set job_id $expect_out(1,string)
@@ -249,7 +249,7 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
exit 1
}
eof {
diff --git a/testsuite/expect/test1.2 b/testsuite/expect/test1.2
index d6b90a0a487..2bde95c6349 100755
--- a/testsuite/expect/test1.2
+++ b/testsuite/expect/test1.2
@@ -53,7 +53,6 @@ expect {
send_user "\nFAILURE: srun not responding\n"
slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.20 b/testsuite/expect/test1.20
index 661875ce48b..f04116e20a4 100755
--- a/testsuite/expect/test1.20
+++ b/testsuite/expect/test1.20
@@ -41,7 +41,7 @@ print_header $test_id
# Spawn a program via srun with stdout forwarding disabled
#
set timeout $max_job_delay
-spawn $srun --output=none -t1 $bin_id
+set srun_pid [spawn $srun --output=none -t1 $bin_id]
expect {
-re "uid=" {
send_user "\nFAILURE: srun improperly forwarded stdout\n"
@@ -55,9 +55,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
@@ -68,7 +67,7 @@ expect {
# Spawn a program via srun with stderr forwarding disabled
#
set matches 0
-spawn $srun --error=none -t1 $bin_sleep aaa
+set srun_pid [spawn $srun --error=none -t1 $bin_sleep aaa]
expect {
-re "invalid" {
send_user "\nFAILURE: srun improperly forwarded stderr\n"
@@ -82,9 +81,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.21 b/testsuite/expect/test1.21
index 794b7268dda..401d2f0719c 100755
--- a/testsuite/expect/test1.21
+++ b/testsuite/expect/test1.21
@@ -44,7 +44,7 @@ print_header $test_id
# Spawn a shell via srun with stdout forwarding disabled
#
set timeout $max_job_delay
-spawn $srun --input=$task_id -N1 -n10 -O -v --wait=2 -t1 $bin_bash
+set srun_pid [spawn $srun --input=$task_id -N1 -n10 -O -v --wait=2 -t1 $bin_bash]
expect {
-re "launching ($number).0 on host" {
set job_id $expect_out(1,string)
@@ -68,9 +68,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.22 b/testsuite/expect/test1.22
index 9ab76e5903a..2cf22b19a2d 100755
--- a/testsuite/expect/test1.22
+++ b/testsuite/expect/test1.22
@@ -53,7 +53,7 @@ if { [test_bluegene] } {
}
}
-spawn $srun -N$node_cnt -l --threads=1 -t1 $bin_hostname
+set srun_pid [spawn $srun -N$node_cnt -l --threads=1 -t1 $bin_hostname]
expect {
-re "0: ($alpha_numeric)" {
set host_0 $expect_out(1,string)
@@ -61,9 +61,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
@@ -83,7 +82,7 @@ if {[string compare $host_0 ""] == 0} {
#
set host_0 ""
set timeout $max_job_delay
-spawn $srun -N$node_cnt -n32 -O -l --threads=32 -t1 $bin_hostname
+set srun_pid [spawn $srun -N$node_cnt -n32 -O -l --threads=32 -t1 $bin_hostname]
expect {
-re "0: ($alpha_numeric)" {
set host_0 $expect_out(1,string)
@@ -91,9 +90,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.23 b/testsuite/expect/test1.23
index ea1f5588a5e..0eefd473748 100755
--- a/testsuite/expect/test1.23
+++ b/testsuite/expect/test1.23
@@ -44,7 +44,7 @@ print_header $test_id
set err_msg 0
set host_0 ""
set timeout $max_job_delay
-spawn $srun -N1 -l --mincpus=999999 -t1 $bin_hostname
+set srun_pid [spawn $srun -N1 -l --mincpus=999999 -t1 $bin_hostname]
expect {
-re "configuration is not available" {
send_user "This error is expected, no worries\n"
@@ -57,9 +57,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
@@ -80,7 +79,7 @@ if {$err_msg != 1} {
#
set host_0 ""
set timeout $max_job_delay
-spawn $srun -N1 -l --mincpus=1 -t1 $bin_hostname
+set srun_pid [spawn $srun -N1 -l --mincpus=1 -t1 $bin_hostname]
expect {
-re "0: ($alpha_numeric)" {
set host_0 $expect_out(1,string)
@@ -88,9 +87,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
@@ -108,7 +106,7 @@ if {[string compare $host_0 ""] == 0} {
set err_msg 0
set host_0 ""
set timeout $max_job_delay
-spawn $srun -N1 -l --mem=999999 -t1 $bin_hostname
+set srun_pid [spawn $srun -N1 -l --mem=999999 -t1 $bin_hostname]
expect {
-re "configuration is not available" {
send_user "This error is expected, no worries\n"
@@ -121,9 +119,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
@@ -144,7 +141,7 @@ if {$err_msg != 1} {
#
set host_0 ""
set timeout $max_job_delay
-spawn $srun -N1 -l --mem=1 -t1 $bin_hostname
+set srun_pid [spawn $srun -N1 -l --mem=1 -t1 $bin_hostname]
expect {
-re "0: ($alpha_numeric)" {
set host_0 $expect_out(1,string)
@@ -152,9 +149,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
@@ -172,7 +168,7 @@ if {[string compare $host_0 ""] == 0} {
set err_msg 0
set host_0 ""
set timeout $max_job_delay
-spawn $srun -N1 -l --tmp=999999999 -t1 $bin_hostname
+set srun_pid [spawn $srun -N1 -l --tmp=999999999 -t1 $bin_hostname]
expect {
-re "configuration is not available" {
send_user "This error is expected, no worries\n"
@@ -185,9 +181,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
@@ -208,7 +203,7 @@ if {$err_msg != 1} {
#
set host_0 ""
set timeout $max_job_delay
-spawn $srun -N1 -l --tmp=1 -t1 $bin_hostname
+set srun_pid [spawn $srun -N1 -l --tmp=1 -t1 $bin_hostname]
expect {
-re "0: ($alpha_numeric)" {
set host_0 $expect_out(1,string)
@@ -216,9 +211,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.24 b/testsuite/expect/test1.24
index 03a3783546c..829cf84e80e 100755
--- a/testsuite/expect/test1.24
+++ b/testsuite/expect/test1.24
@@ -43,7 +43,7 @@ print_header $test_id
set err_msg 0
set host_0 ""
set timeout $max_job_delay
-spawn $srun -N1 -l --constraint=invalid,constraint -t1 $bin_hostname
+set srun_pid [spawn $srun -N1 -l --constraint=invalid,constraint -t1 $bin_hostname]
expect {
-re "configuration is not available" {
send_user "This error is expected, no worries\n"
@@ -56,9 +56,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.25 b/testsuite/expect/test1.25
index 7c6d06f5892..c590dbe5308 100755
--- a/testsuite/expect/test1.25
+++ b/testsuite/expect/test1.25
@@ -54,7 +54,7 @@ if { [test_bluegene] } {
}
}
-spawn $srun -N$node_cnt --no-kill -t1 $bin_sleep $sleep_time
+set srun_pid [spawn $srun -N$node_cnt --no-kill -t1 $bin_sleep $sleep_time]
expect {
-re "error" {
send_user "\nFAILURE: some error occurred\n"
@@ -63,9 +63,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.26 b/testsuite/expect/test1.26
index bf90382ddfc..9e3d5c53db2 100755
--- a/testsuite/expect/test1.26
+++ b/testsuite/expect/test1.26
@@ -54,7 +54,7 @@ if {[is_super_user] == 0} {
set host_0 ""
set nodelist_name ""
set timeout $max_job_delay
-spawn $srun -v -N1 -l $bin_printenv SLURMD_NODENAME
+set srun_pid [spawn $srun -v -N1 -l $bin_printenv SLURMD_NODENAME]
expect {
-re "on host ($alpha_numeric)," {
set nodelist_name $expect_out(1,string)
@@ -70,7 +70,7 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
}
eof {
@@ -89,7 +89,7 @@ if {[string compare $nodelist_name ""] == 0} {
send_user "\nFAILURE: Did not get nodelist_name of task 0\n"
exit 1
}
-set include_node $nodelist_name
+set include_node $host_0
#
# Submit a job directly to that node
@@ -97,7 +97,7 @@ set include_node $nodelist_name
set host_1 ""
set slurm_user 1
set timeout 10
-spawn $srun -N1 -l --nodelist=$include_node --no-allocate -t1 $bin_printenv SLURMD_NODENAME
+set srun_pid [spawn $srun -N1 -l --nodelist=$include_node --no-allocate -t1 $bin_printenv SLURMD_NODENAME]
expect {
-re "Invalid job credential" {
send_user "\nWARNING: Not SlurmUser or root.\n"
@@ -124,6 +124,7 @@ expect {
timeout {
send_user "\nWARNING: srun not responding, "
send_user "expected if not SlurmUser or root.\n"
+ slow_kill $srun_pid
set slurm_user 0
}
eof {
@@ -133,23 +134,11 @@ expect {
if {$slurm_user == 0} {
exit 0;
}
-if {[string compare $host_0 $host_1]} {
- send_user "\nFAILURE: host name value wrong $host_0 $host_1\n"
+if {[string compare $host_1 $include_node]} {
+ send_user "\nFAILURE: Allocation lacked an included node\n"
set exit_code 1
}
-#
-# Verify node count and name
-#
-if {[string compare $host_0 ""] == 0} {
- send_user "\nFAILURE: Did not get hostname of task 0\n"
- set exit_code 1
-}
-#if {[string compare $host_0 $include_node] != 0} {
-# send_user "\nFAILURE: Allocation lacked an included node\n"
-# set exit_code 1
-#}
-
#
# Run three tasks at a time on some node and do so repeatedly
# This checks for slurmd race conditions
@@ -164,13 +153,13 @@ for {set inx 0} {$inx < $interations} {incr inx} {
exec $bin_usleep 250000
set failures 0
- spawn $srun -N1 --nodelist=$include_node -t1 -l $bin_printenv SLURMD_NODENAME
+ set srun_pid [spawn $srun -N1 --nodelist=$include_node -t1 -l $bin_printenv SLURMD_NODENAME]
set alloc $spawn_id
- spawn $srun -N1 --nodelist=$include_node -Z $bin_usleep 500000
+ set srun_pid1 [spawn $srun -N1 --nodelist=$include_node -Z $bin_usleep 500000]
set noalloc1 $spawn_id
- spawn $srun -N1 --nodelist=$include_node -Z $bin_usleep 250000
+ set srun_pid2 [spawn $srun -N1 --nodelist=$include_node -Z $bin_usleep 250000]
set noalloc2 $spawn_id
set timeout 20
@@ -203,6 +192,7 @@ for {set inx 0} {$inx < $interations} {incr inx} {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
+ slow_kill $srun_pid2
set failures 1
}
eof {
@@ -239,6 +229,7 @@ for {set inx 0} {$inx < $interations} {incr inx} {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
+ slow_kill $srun_pid1
set failures 1
}
eof {
@@ -278,6 +269,7 @@ for {set inx 0} {$inx < $interations} {incr inx} {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
+ slow_kill $srun_pid
set failures 1
}
eof {
diff --git a/testsuite/expect/test1.27 b/testsuite/expect/test1.27
index 6d8fe3a0b5b..40df8849375 100755
--- a/testsuite/expect/test1.27
+++ b/testsuite/expect/test1.27
@@ -63,7 +63,7 @@ array set good_vars {
# Spawn a job via srun to print environment variables
#
set timeout $max_job_delay
-spawn $srun -N1 -n1 -t1 $bin_env
+set srun_pid [spawn $srun -N1 -n1 -t1 $bin_env]
expect {
-re "(SLURM_$alpha_under)=($alpha_numeric)" {
set found_vars($expect_out(1,string)) "$expect_out(2,string)"
@@ -71,9 +71,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.28 b/testsuite/expect/test1.28
index 2aae4a492bc..3bd362a7110 100755
--- a/testsuite/expect/test1.28
+++ b/testsuite/expect/test1.28
@@ -47,7 +47,7 @@ print_header $test_id
#
set timeout $max_job_delay
set env($test_env_name) $test_env_val
-spawn $srun -N1 -t1 $bin_env
+set srun_pid [spawn $srun -N1 -t1 $bin_env]
expect {
-re "$test_env_name=($number)" {
if {$expect_out(1,string) == $test_env_val} {
@@ -57,9 +57,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.29 b/testsuite/expect/test1.29
index 5ed500f6efe..44730f107f3 100755
--- a/testsuite/expect/test1.29
+++ b/testsuite/expect/test1.29
@@ -133,7 +133,7 @@ make_bash_script $file_in "
set timeout $max_job_delay
-spawn $srun --batch --output=$file_out --error=$file_err -t1 ./$file_in
+set srun_pid [spawn $srun --batch --output=$file_out --error=$file_err -t1 ./$file_in]
expect {
-re "jobid ($number) submitted" {
set job_id $expect_out(1,string)
@@ -141,9 +141,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.3 b/testsuite/expect/test1.3
index 23838281a66..57ec23bb0cb 100755
--- a/testsuite/expect/test1.3
+++ b/testsuite/expect/test1.3
@@ -50,7 +50,6 @@ expect {
send_user "\nFAILURE: srun not responding\n"
slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.30 b/testsuite/expect/test1.30
index bb149952d1d..30337c0c675 100755
--- a/testsuite/expect/test1.30
+++ b/testsuite/expect/test1.30
@@ -55,7 +55,7 @@ if { [test_bluegene] } {
#
set timeout $max_job_delay
for {set node_cnt 1} {$node_cnt > 0} {set node_cnt [expr $node_cnt * 2]} {
- spawn $srun --immediate -N$node_cnt $srun_opts -t2 $bin_hostname
+ set srun_pid [spawn $srun --immediate -N$node_cnt $srun_opts -t2 $bin_hostname]
expect {
-re "Immediate execution impossible" {
send_user "This error is expected, no worries\n"
@@ -85,9 +85,8 @@ for {set node_cnt 1} {$node_cnt > 0} {set node_cnt [expr $node_cnt * 2]} {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- set exit_code 1
- kill_srun
- exp_continue
+ slow_kill $srun_pid
+ set exit_code 1
}
eof {
wait
diff --git a/testsuite/expect/test1.31 b/testsuite/expect/test1.31
index fca0bfaaa39..d4693de681b 100755
--- a/testsuite/expect/test1.31
+++ b/testsuite/expect/test1.31
@@ -77,7 +77,7 @@ exec $bin_rm -f $env_valu_stdout
#
# Spawn a job via srun using these environment variables
#
-spawn $srun -t1 $bin_env
+set srun_pid [spawn $srun -t1 $bin_env]
expect {
-re "jobid ($number)" {
set job_id $expect_out(1,string)
@@ -86,9 +86,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.32 b/testsuite/expect/test1.32
index 6fd60f4feed..197a05ccca9 100755
--- a/testsuite/expect/test1.32
+++ b/testsuite/expect/test1.32
@@ -79,13 +79,10 @@ expect {
# Expect have difficulties handling unbuffered srun output
#
set timeout $max_job_delay
-spawn $srun -N1 -t1 --unbuffered $file_prog
-set srun_pid [exp_pid]
+set srun_pid [spawn $srun -N1 -t1 --unbuffered $file_prog]
exec $bin_sleep 5
exec $bin_kill -USR1 $srun_pid
-#exec $bin_pkill -USR1 -n -u $uid srun
exec $bin_kill -USR2 $srun_pid
-#exec $bin_pkill -USR2 -n -u $uid srun
expect {
-re "WAITING" {
incr matches
@@ -105,9 +102,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
send_user "\nEOF\n"
diff --git a/testsuite/expect/test1.33 b/testsuite/expect/test1.33
index 63f6d778c52..a42c1bfb4a8 100755
--- a/testsuite/expect/test1.33
+++ b/testsuite/expect/test1.33
@@ -47,7 +47,6 @@ make_bash_script $exit_script "
make_bash_script $test_script "
$srun -N1 -n2 -O $exit_script
-# RC=\$?
echo srun_exit_code_\$?
"
@@ -56,7 +55,7 @@ make_bash_script $test_script "
#
set sum 0
set timeout $max_job_delay
-spawn $srun -N1 -n2 -O $exit_script
+set srun_pid [spawn $srun -N1 -n2 -O $exit_script]
expect {
-re "exit code ($number)" {
send_user "This error is expected, no worries\n"
@@ -65,9 +64,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
@@ -104,7 +102,5 @@ if {$matches != 1} {
if {$exit_code == 0} {
exec $bin_rm -f $exit_script $test_script
send_user "\nSUCCESS\n"
-} else {
- kill_srun
}
exit $exit_code
diff --git a/testsuite/expect/test1.34 b/testsuite/expect/test1.34
index b97deefd891..13f9dee3632 100755
--- a/testsuite/expect/test1.34
+++ b/testsuite/expect/test1.34
@@ -58,7 +58,7 @@ make_bash_script $file_in "$bin_echo \$1,\$2"
# Spawn a srun batch job with arguments
#
set timeout $max_job_delay
-spawn $srun --batch --output=$file_out --error=$file_err -t1 $file_in $arg1 $arg2
+set srun_pid [spawn $srun --batch --output=$file_out --error=$file_err -t1 $file_in $arg1 $arg2]
expect {
-re "jobid ($number) submitted" {
set job_id $expect_out(1,string)
@@ -66,9 +66,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
@@ -85,6 +84,7 @@ if {$job_id == 0} {
#
if {[wait_for_job $job_id "DONE"] != 0} {
send_user "\nFAILURE: waiting for job to complete\n"
+ cancel_job $job_id
set exit_code 1
}
diff --git a/testsuite/expect/test1.35 b/testsuite/expect/test1.35
index ba277667aba..6d76bfb9de4 100755
--- a/testsuite/expect/test1.35
+++ b/testsuite/expect/test1.35
@@ -6,9 +6,6 @@
# Output: "TEST: #.#" followed by "SUCCESS" if test was successful, OR
# "FAILURE: ..." otherwise with an explanation of the failure, OR
# anything else indicates a failure mode that must be investigated.
-#
-# Note: This script generates and then deletes files in the working directory
-# named test1.35.input, test1.35.output, and test1.35.error
############################################################################
# Copyright (C) 2002-2006 The Regents of the University of California.
# Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
@@ -73,7 +70,7 @@ if { [test_bluegene] } {
}
}
-spawn $srun --batch -N$node_cnt --output=$file_out --error=$file_err -t1 $file_in
+set srun_pid [spawn $srun --batch -N$node_cnt --output=$file_out --error=$file_err -t1 $file_in]
expect {
-re "jobid ($number) submitted" {
set job_id $expect_out(1,string)
@@ -81,9 +78,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
@@ -100,6 +96,7 @@ if {$job_id == 0} {
#
if {[wait_for_job $job_id "DONE"] != 0} {
send_user "\nFAILURE: waiting for job to complete\n"
+ cancel_job $job_id
set exit_code 1
}
diff --git a/testsuite/expect/test1.36 b/testsuite/expect/test1.36
index e1cd00443ed..3e3632a1772 100755
--- a/testsuite/expect/test1.36
+++ b/testsuite/expect/test1.36
@@ -58,7 +58,7 @@ if { [test_bluegene] } {
}
}
-spawn $srun -N$node_cnt -n$task_cnt -O -t1 $srun -l -n$mult -O $bin_id
+set srun_pid [spawn $srun -N$node_cnt -n$task_cnt -O -t1 $srun -l -n$mult -O $bin_id]
expect {
-re "($number): uid=" {
incr task_output
@@ -66,9 +66,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.37 b/testsuite/expect/test1.37
index bf8e57c10af..5915a40baf3 100755
--- a/testsuite/expect/test1.37
+++ b/testsuite/expect/test1.37
@@ -48,7 +48,7 @@ print_header $test_id
# Submit a job and get the node's NodeName from the nodelist
#
set timeout $max_job_delay
-spawn $srun -v -N1 -l -t1 $bin_hostname
+set srun_pid [spawn $srun -v -N1 -l -t1 $bin_hostname]
expect {
-re "on host ($alpha_numeric)," {
set nodelist_name $expect_out(1,string)
@@ -60,9 +60,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
@@ -89,7 +88,7 @@ make_bash_script $file_in "$srun $bin_sleep 5"
# with sharing permitted. Insure the first job completes before the
# second job is started.
#
-spawn $srun --batch -N1 --exclusive --nodelist=$nodelist_name -t1 --output=$file_out --error=$file_err $file_in
+set srun_pid [spawn $srun --batch -N1 --exclusive --nodelist=$nodelist_name -t1 --output=$file_out --error=$file_err $file_in]
expect {
-re "jobid ($number) submitted" {
set job_id1 $expect_out(1,string)
@@ -97,7 +96,7 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
exit 1
}
eof {
@@ -113,7 +112,7 @@ if {$job_id1 == 0} {
set partition "dummy"
set waited 1
set timeout [expr $timeout + 5]
-spawn $srun -N1 --nodelist=$nodelist_name -t1 --share $scontrol -o show job $job_id1
+set srun_pid [spawn $srun -N1 --nodelist=$nodelist_name -t1 --share $scontrol -o show job $job_id1]
expect {
-re "Partition=($alpha_numeric)" {
set partition $expect_out(1,string)
@@ -125,7 +124,7 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
exit 1
}
eof {
diff --git a/testsuite/expect/test1.38 b/testsuite/expect/test1.38
index 07a093b1ce3..e3084acf3f2 100755
--- a/testsuite/expect/test1.38
+++ b/testsuite/expect/test1.38
@@ -74,8 +74,7 @@ expect {
set timeout $max_job_delay
set matches 0
set job_id 0
-spawn $srun -v -N1 -t1 --unbuffered $file_in
-set srun_pid [exp_pid]
+set srun_pid [spawn $srun -v -N1 -t1 --unbuffered $file_in]
expect {
-re "launching ($number).0" {
set job_id $expect_out(1,string)
@@ -84,7 +83,6 @@ expect {
-re "WAITING" {
incr matches
exec $bin_kill -INT $srun_pid
-# exec $bin_pkill -INT -n -u $uid srun
exp_continue
}
-re "srun: interrupt" {
@@ -98,9 +96,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
@@ -118,8 +115,7 @@ if {$matches != 3} {
#
set matches 0
set job_id 0
-spawn $srun -v -N1 -t1 --unbuffered --quit-on-interrupt $file_in
-set srun_pid [exp_pid]
+set srun_pid [spawn $srun -v -N1 -t1 --unbuffered --quit-on-interrupt $file_in]
expect {
-re "launching ($number).0" {
set job_id $expect_out(1,string)
@@ -128,7 +124,6 @@ expect {
-re "WAITING" {
incr matches
exec $bin_kill -INT $srun_pid
-# exec $bin_pkill -INT -n -u $uid srun
exp_continue
}
-re "srun: interrupt" {
@@ -145,9 +140,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.39 b/testsuite/expect/test1.39
index 34885f0eda6..4d8cfd1351d 100755
--- a/testsuite/expect/test1.39
+++ b/testsuite/expect/test1.39
@@ -44,12 +44,17 @@ print_header $test_id
# Test if "light" core file supported
#
set matches 0
-spawn $srun --core=list
+set srun_pid [spawn $srun --core=list]
expect {
-re " light " {
set matches 1
exp_continue
}
+ timeout {
+ send_user "\nFAILURE: srun not responding\n"
+ slow_kill $srun_pid
+ exit 1
+ }
eof {
wait
}
@@ -77,7 +82,7 @@ foreach filename [glob -nocomplain *core*] {
# Spawn initial program via srun
#
set timeout $max_job_delay
-spawn $srun -N1 -t1 --core=light $file_prog
+set srun_pid [spawn $srun -N1 -t1 --core=light $file_prog]
expect {
-re "Segmentation fault" {
send_user "\nNo worries, this error is expected.\n"
@@ -85,9 +90,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.4 b/testsuite/expect/test1.4
index 177f1edb896..cd70baea80e 100755
--- a/testsuite/expect/test1.4
+++ b/testsuite/expect/test1.4
@@ -50,7 +50,6 @@ expect {
send_user "\nFAILURE: srun not responding\n"
slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.5 b/testsuite/expect/test1.5
index 179da4a4245..fae47de2425 100755
--- a/testsuite/expect/test1.5
+++ b/testsuite/expect/test1.5
@@ -58,7 +58,6 @@ expect {
send_user "\nFAILURE: srun not responding\n"
slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.6 b/testsuite/expect/test1.6
index 2863c290832..f4311e80ff2 100755
--- a/testsuite/expect/test1.6
+++ b/testsuite/expect/test1.6
@@ -59,7 +59,6 @@ expect {
send_user "\nFAILURE: srun not responding\n"
slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.7 b/testsuite/expect/test1.7
index 3e1c12230ed..ed0515be2b9 100755
--- a/testsuite/expect/test1.7
+++ b/testsuite/expect/test1.7
@@ -6,9 +6,6 @@
# Output: "TEST: #.#" followed by "SUCCESS" if test was successful, OR
# "FAILURE: ..." otherwise with an explanation of the failure, OR
# anything else indicates a failure mode that must be investigated.
-#
-# Note: This script generates and then deletes files in the working directory
-# named test1.7.input, test1.7.output, and test1.7.error
############################################################################
# Copyright (C) 2002-2006 The Regents of the University of California.
# Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
@@ -110,7 +107,7 @@ make_bash_script $file_in "
#
set timeout [expr $max_job_delay + $sleep_time]
set timed_out 0
-spawn $srun -t1 $bin_sleep $sleep_time
+set srun_pid [spawn $srun -t1 $bin_sleep $sleep_time]
expect {
-re "job exceeded timelimit" {
set timed_out 1
@@ -126,9 +123,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
@@ -142,7 +138,7 @@ if {$timed_out == 1} {
}
set completions 0
-spawn $srun -t4 $bin_sleep $sleep_time
+set srun_pid [spawn $srun -t4 $bin_sleep $sleep_time]
expect {
-re "job exceeded timelimit" {
set completions -1
@@ -162,9 +158,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
@@ -180,7 +175,7 @@ if {$completions != 1} {
# Spawn a srun batch job with arguments
#
set timeout $max_job_delay
-spawn $srun --batch --output=$file_out --error=$file_err -t4 $file_in
+set srun_pid [spawn $srun --batch --output=$file_out --error=$file_err -t4 ./$file_in]
expect {
-re "jobid ($number) submitted" {
set job_id $expect_out(1,string)
@@ -188,7 +183,7 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
}
eof {
@@ -224,7 +219,8 @@ if {[wait_for_file $file_out] == 0} {
}
if {$output_fini == 0} {
- send_user "\nFAILURE: Unexpected batch job output, possible premature job termination\n"
+ send_user "\nFAILURE: Unexpected batch job output, "
+ send_user "possible premature job termination\n"
set exit_code 1
}
diff --git a/testsuite/expect/test1.8 b/testsuite/expect/test1.8
index 29b9d88c676..ee9303b4906 100755
--- a/testsuite/expect/test1.8
+++ b/testsuite/expect/test1.8
@@ -63,7 +63,7 @@ exec $bin_chmod 700 $file_in
# Spawn a shell via srun that uses stdin/out/err and confirm their contents
#
set timeout $max_job_delay
-spawn $srun --input=$file_in --output=$file_out --error=$file_err -t1 $bin_bash
+set srun_pid [spawn $srun --input=$file_in --output=$file_out --error=$file_err -t1 $bin_bash]
expect {
-re "Unable to contact" {
send_user "\nFAILURE: slurm appears to be down\n"
@@ -71,9 +71,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
@@ -150,7 +149,7 @@ if {$got_sleep_err == 0} {
#
# Spawn a program to run for a while with no input, output or error
#
-spawn $srun --input=none --output=none --error=none -t1 -N1 od -c $srun
+set srun_pid [spawn $srun --input=none --output=none --error=none -t1 -N1 od -c $srun]
expect {
-re "Unable to contact" {
send_user "\nFAILURE: slurm appears to be down\n"
@@ -168,9 +167,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test1.9 b/testsuite/expect/test1.9
index e91341879e9..22be915a3e7 100755
--- a/testsuite/expect/test1.9
+++ b/testsuite/expect/test1.9
@@ -43,7 +43,7 @@ print_header $test_id
# Submit a slurm job that will execute 'id' on 1 node and over task_cnt tasks
#
set timeout $max_job_delay
-spawn $srun -N1 -n$task_cnt -O -v -t1 $bin_id
+set srun_pid [spawn $srun -N1 -n$task_cnt -O -v -t1 $bin_id]
expect {
-re "jobid" {
set verbosity 1
@@ -59,9 +59,8 @@ expect {
}
timeout {
send_user "\nFAILURE: srun not responding\n"
- kill_srun
+ slow_kill $srun_pid
set exit_code 1
- exp_continue
}
eof {
wait
diff --git a/testsuite/expect/test15.10 b/testsuite/expect/test15.10
index 27c715d21ad..fc880c863ae 100755
--- a/testsuite/expect/test15.10
+++ b/testsuite/expect/test15.10
@@ -59,13 +59,11 @@ for {set inx 1} {$inx < 4} {set inx [expr $inx * 2]} {
}
timeout {
send_user "\nFAILURE: salloc not responding\n"
- if {$job_id == 0} {
- slow_kill $salloc_pid
- } else {
+ if {$job_id != 0} {
cancel_job $job_id
}
+ slow_kill $salloc_pid
set exit_code 1
- exp_continue
}
eof {
if {$tasks_get < $tasks_set} {
--
GitLab