Skip to contentSkip to conversation

cgroup_fj_stress: Avoid killall

PATCH New Clemens Famulla-Conrad 2 participants 4 messages Nov 5, 2019 – Nov 6, 2019

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 cgroup_fj_stress: Avoid killall

We discovered problems that killall didn't catched all processes. With this patch, we collect the pids manually and kill them one after the other.

---
 testcases/kernel/controllers/cgroup_fj/cgroup_fj_stress.sh | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)
testcases/kernel/controllers/cgroup_fj/cgroup_fj_stress.sh +6 -3
index 698aa4979..27ea7634a 100755
@@ -74,6 +74,7 @@ setup
 export TMPFILE=./tmp_tasks.$$
 
 count=0
+collected_pids=""
 
 build_subgroups()
 {
@@ -107,6 +108,7 @@ attach_task()
     if [ -z "$ppid" ]; then
         cgroup_fj_proc&
         pid=$!
+        collected_pids="$collected_pids $pid"
     else
         pid="$ppid"
     fi
@@ -148,9 +150,10 @@ case $attach_operation in
 "each" )
     tst_resm TINFO "Attaching task to each subgroup"
     attach_task "$start_path" 0
-    ROD killall -9 "cgroup_fj_proc"
-    # Wait for attached tasks to terminate
-    wait
+    for pid in $collected_pids; do
+        ROD kill -9 "$pid"
+        wait "$pid"
+    done
     ;;
 *  )
     ;;
Signature · 1 line
2.16.4
#1 lore raw

PATCH 1/1 cgroup_fj_stress: Avoid killall

Hi Clements,

Quoted 3 lines

We discovered problems that killall didn't catched all processes. With this patch, we collect the pids manually and kill them one after the other.

LGTM.
I wonder if we also want to kill cgroup_fj_proc this way (see cgroup_fj_common.sh).

I guess you're not planning to create minimal reproducer to prove the problem of left processes after killall, are you?

Quoted 10 lines
---
 testcases/kernel/controllers/cgroup_fj/cgroup_fj_stress.sh | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)
testcases/kernel/controllers/cgroup_fj/cgroup_fj_stress.sh +0 -0
index 698aa4979..27ea7634a 100755

testcases/kernel/controllers/cgroup_fj/cgroup_fj_stress.sh setup · line 77

Quoted 4 lines
testcases/kernel/controllers/cgroup_fj/cgroup_fj_stress.sh +1 -0
@@ -74,6 +74,7 @@ setup
 export TMPFILE=./tmp_tasks.$$
 
 count=0
+collected_pids=""

nit:
collected_pids=

...

Kind regards,
Petr

#2 lore raw

PATCH 1/1 cgroup_fj_stress: Avoid killall

Hi Petr,

On Tue, 2019-11-05 at 14:20 +0100, Petr Vorel wrote:
<snip>

Quoted 2 lines

I wonder if we also want to kill cgroup_fj_proc this way (see cgroup_fj_common.sh).

I'm not sure if I understand you. We do kill cgroup_fj_proc this way. The killall -9 cgroup_fj_proc call in cgrouip_fj_common.sh looks for me like a cleaner and there is no wait or similar afterwards, so I would guess we are not facing the problem here. And I would keep killall here.
As far as I can see, all other cgroup_fj_proc& calls already kill them separately.

Quoted 3 lines

I guess you're not planning to create minimal reproducer to prove the problem of
left processes after killall, are you?

Sure nice idea, I can give it a try. But not within this patchset.

Thanks
Clemens

#3 lore raw

PATCH 1/1 cgroup_fj_stress: Avoid killall

Hi Clemens,

Quoted 10 lines
Quoted 2 lines

I wonder if we also want to kill cgroup_fj_proc this way (see cgroup_fj_common.sh).

I'm not sure if I understand you. We do kill cgroup_fj_proc this way. The killall -9 cgroup_fj_proc call in cgrouip_fj_common.sh looks for me like a cleaner and there is no wait or similar afterwards, so I would guess we are not facing the problem here. And I would keep killall here.
As far as I can see, all other cgroup_fj_proc& calls already kill them separately.

OK, merged :).
Thanks!

Quoted 5 lines
Quoted 3 lines

I guess you're not planning to create minimal reproducer to prove the problem of
left processes after killall, are you?

Sure nice idea, I can give it a try. But not within this patchset.

Thanks!

Kind regards,
Petr