diff options
| author | Thomas White <taw@physics.org> | 2020-09-02 12:11:28 +0200 |
|---|---|---|
| committer | Thomas White <taw@physics.org> | 2020-09-02 12:11:28 +0200 |
| commit | dbe2734d875f9dec5a3b28b9619eb5eeb9446121 (patch) | |
| tree | eabbcd12c24b582584ed34c2c6082eb6533e36b4 /src/gui_backend_slurm.c | |
| parent | 5bd6c58c82eba7109cb6365440aa43f4d577ed67 (diff) | |
Fix job tracking logic
Diffstat (limited to 'src/gui_backend_slurm.c')
| -rw-r--r-- | src/gui_backend_slurm.c | 22 |
1 files changed, 18 insertions, 4 deletions
diff --git a/src/gui_backend_slurm.c b/src/gui_backend_slurm.c index f51af9e9..504bf3d4 100644 --- a/src/gui_backend_slurm.c +++ b/src/gui_backend_slurm.c @@ -97,7 +97,20 @@ static int job_running(uint32_t job_id) /* FIXME: Distinguish error cond from job complete */ } - /* FIXME: Check that job is actually still running */ + switch ( job_info->job_array[0].job_state & JOB_STATE_BASE ) { + + /* Only the following states are reasons to keep on watching + * the job */ + case JOB_PENDING : + case JOB_RUNNING : + case JOB_SUSPENDED : + running = 1; + break; + + default : + running = 0; + break; + } slurm_free_job_info_msg(job_info); @@ -118,9 +131,9 @@ static int get_task_status(void *job_priv, n_proc += read_number_processed(job->stderr_filenames[i]); - if ( (job->job_ids[i] != 0) - && !job_running(job->job_ids[i]) ) - { + if ( job->job_ids[i] == 0 ) continue; + + if ( !job_running(job->job_ids[i]) ) { job->job_ids[i] = 0; } else { all_complete = 0; @@ -138,6 +151,7 @@ static void cancel_task(void *job_priv) int i; struct slurm_job *job = job_priv; for ( i=0; i<job->n_blocks; i++) { + if ( job->job_ids[i] == 0 ) continue; STATUS("Stopping SLURM job %i\n", job->job_ids[i]); if ( slurm_kill_job(job->job_ids[i], SIGINT, 0) ) { ERROR("Couldn't stop job: %s\n", |