Skip to content
This repository has been archived by the owner on Sep 18, 2024. It is now read-only.

change SIGKILL to SIGTERM in local mode cancel trial job #3173

Merged
merged 7 commits into from
Dec 15, 2020
Merged

change SIGKILL to SIGTERM in local mode cancel trial job #3173

Changes from 4 commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
533d8bd
change SIGKILL to SIGTERM in local cancel trial job
Dec 10, 2020
ed87bc1
fix bug
Dec 10, 2020
fe65dfa
fix eslint
Dec 10, 2020
36a0932
reduce delay duration
Dec 10, 2020
41a40ec
change to loop check alive
Dec 14, 2020
ae8bf16
add break
J-shang Dec 15, 2020
c2ff7e6
fix eslint
J-shang Dec 15, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 10 additions & 1 deletion ts/nni_manager/training_service/local/localTrainingService.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -253,7 +253,16 @@ class LocalTrainingService implements TrainingService {

return Promise.resolve();
}
tkill(trialJob.pid, 'SIGKILL');
tkill(trialJob.pid, 'SIGTERM');
const pid = trialJob.pid;
setTimeout(((pid: number): void => {
tkill(pid, 'SIGKILL', (err) => {
if (err){
this.log.warning(`cancel trial job {pid: ${pid}} failed: ${err.message}`);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I suddenly thought of a problem.
If the pid get reused between SIGTERM and SIGKILL, this will kill a random process.
If you don't want to elegantly solve the problem in this release, I suggest to reduce the delay to minimize risk.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, and other place use kill pid also has this risk, maybe elegant way is let them kill themselves, need discuss. This release reduce the delay to 1 second.

}
});
}).bind(this), 1000, pid);

this.setTrialJobStatus(trialJob, getJobCancelStatus(isEarlyStopped));

return Promise.resolve();
Expand Down