From: Zack Cerza <zack@redhat.com>
Date: Mon, 21 Mar 2022 23:44:34 +0000 (-0600)
Subject: supervisor: Return job's return code
X-Git-Tag: 1.2.0~190^2~14
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=78304b776f69af384b72e895aa95180529030950;p=teuthology.git

supervisor: Return job's return code

Signed-off-by: Zack Cerza <zack@redhat.com>
---

diff --git a/scripts/dispatcher.py b/scripts/dispatcher.py
index 669f12511..8d955e48c 100644
--- a/scripts/dispatcher.py
+++ b/scripts/dispatcher.py
@@ -24,10 +24,11 @@ standard arguments:
 """
 
 import docopt
+import sys
 
 import teuthology.dispatcher
 
 
 def main():
     args = docopt.docopt(__doc__)
-    teuthology.dispatcher.main(args)
+    sys.exit(teuthology.dispatcher.main(args))
diff --git a/teuthology/dispatcher/__init__.py b/teuthology/dispatcher/__init__.py
index f6eea469d..5902f4035 100644
--- a/teuthology/dispatcher/__init__.py
+++ b/teuthology/dispatcher/__init__.py
@@ -88,6 +88,7 @@ def main(args):
     fetch_qa_suite('master')
 
     keep_running = True
+    job_procs = set()
     while keep_running:
         # Check to see if we have a teuthology-results process hanging around
         # and if so, read its return code so that it can exit.
@@ -107,6 +108,8 @@ def main(args):
         if job is None:
             continue
 
+        job_procs = set(filter(lambda p: p.poll() is None, job_procs))
+
         # bury the job so it won't be re-run if it fails
         job.bury()
         job_id = job.jid
@@ -154,6 +157,7 @@ def main(args):
 
         try:
             job_proc = subprocess.Popen(run_args)
+            job_procs.add(job_proc)
             log.info('Job supervisor PID: %s', job_proc.pid)
         except Exception:
             error_message = "Saw error while trying to spawn supervisor."
@@ -171,6 +175,12 @@ def main(args):
         except Exception:
             log.exception("Saw exception while trying to delete job")
 
+    returncodes = set([0])
+    for proc in job_procs:
+        if proc.returncode is not None:
+            returncodes.add(proc.returncode)
+    return max(returncodes)
+
 
 def lock_machines(job_config):
     report.try_push_job_info(job_config, dict(status='running'))
diff --git a/teuthology/dispatcher/supervisor.py b/teuthology/dispatcher/supervisor.py
index 3036d0335..f52e37938 100644
--- a/teuthology/dispatcher/supervisor.py
+++ b/teuthology/dispatcher/supervisor.py
@@ -54,14 +54,14 @@ def main(args):
             yaml.safe_dump(job_config, f, default_flow_style=False)
 
     try:
-        run_job(
+        return run_job(
             job_config,
             teuth_bin_path,
             archive_dir,
             verbose
         )
     except SkipJob:
-        return
+        return 0
 
 
 def run_job(job_config, teuth_bin_path, archive_dir, verbose):
@@ -152,6 +152,7 @@ def run_job(job_config, teuth_bin_path, archive_dir, verbose):
         log.info('Success!')
     if 'targets' in job_config:
         unlock_targets(job_config)
+    return p.returncode
 
 def failure_is_reimage(failure_reason):
     if not failure_reason: