fix: handle job array cancellation for arrays that have never started…

… (issue #141)
neilmunday · Sep 1, 2024 · 3311467 · 3311467
1 parent e4c7b66
commit 3311467
Show file tree

Hide file tree

Showing 2 changed files with 20 additions and 5 deletions.
diff --git a/src/slurmmail/cli.py b/src/slurmmail/cli.py
@@ -221,7 +221,15 @@ def __process_spool_file(
                 for i in range(field_num):
                     sacct_dict[fields[i]] = data[i]
 
-                if not re.match(r"^([0-9]+|[0-9]+_[0-9]+|[0-9]+\+[0-9]+)$", sacct_dict["JobId"]):
+                # possible job ID formats:
+                # [0-9]+
+                # [0-9]+_[0-9]+             --> job array
+                # [0-9]+_\[[0-9]+-[0-9]+\]  --> job array, not started
+                # [0-9]+\+[0-9]+            --> HET job
+                if not re.match(
+                    r"^([0-9]+|[0-9]+_[0-9]+|[0-9]+_\[[0-9]+-[0-9]+\]|[0-9]+\+[0-9]+)$",
+                    sacct_dict["JobId"]
+                ):
                     logging.debug("job ID %s failed reg ex match", sacct_dict["JobId"])
                     # grab MaxRSS value
                     if (
@@ -356,7 +364,7 @@ def __process_spool_file(
                 job.save()
                 jobs.append(job)
 
-    if array_summary or len(jobs) == 1:
+    if (array_summary and len(jobs) > 0) or len(jobs) == 1:
         jobs = [jobs[0]]
 
     if not array_summary and 0 < options.array_max_notifications < len(jobs):

diff --git a/src/slurmmail/slurm.py b/src/slurmmail/slurm.py
@@ -73,6 +73,7 @@ class Job:
     """
 
     GECOS_NAME_FIELD: int = 0
+    JOB_ARRAY_NOT_STARTED_RE = re.compile(r"([0-9]+)_\[[0-9]+-[0-9]+\]")
 
     def __init__(
         self, datetime_format: str,
@@ -114,9 +115,15 @@ def __init__(
         self.workdir: Optional[str] = None
 
         if "_" in job_id:
-            array_id, index = job_id.split("_")
-            self.array_id = int(array_id)
-            self.index = int(index)
+
+            # has the job array started?
+            match = Job.JOB_ARRAY_NOT_STARTED_RE.match(job_id)
+            if match:
+                self.array_id = match.group(1)
+            else:
+                array_id, index = job_id.split("_")
+                self.array_id = int(array_id)
+                self.index = int(index)
         elif "+" in job_id:
             hetjob_id, index = job_id.split("+")
             self.hetjob_id = int(hetjob_id)