Differentiate between PID and unique cluster ID #414
Conversation
Codecov Report
@@ Coverage Diff @@
## master #414 +/- ##
==========================================
+ Coverage 90.63% 90.66% +0.03%
==========================================
Files 43 43
Lines 2914 2926 +12
==========================================
+ Hits 2641 2653 +12
Misses 273 273 Continue to review full report at Codecov.
|
django_q/tests/test_monitor.py
Outdated
| @@ -10,15 +11,16 @@ | |||
|
|
|||
| @pytest.mark.django_db | |||
| def test_monitor(monkeypatch): | |||
| assert Stat.get(0).sentinel == 0 | |||
| cluster_id = uuid.uuid4() | |||
| assert Stat.get(pid=0, cluster_id=4).sentinel == 0 | |||
There was a problem hiding this comment.
Yeah, totally, it's a typo. Should be cluster_id=cluster_id.
setup.py
Outdated
| @@ -26,7 +26,7 @@ def run(self): | |||
|
|
|||
| setup( | |||
| name='django-q', | |||
| version='1.1.0', | |||
| version='1.2.0', | |||
There was a problem hiding this comment.
I prefer to keep the versioning out of the PR's. It's never clear when it will be released, there could be some package updates in between for example.
|
Great work on this btw. Especially since this is one of those things that just doesn't come up in the way I use it mysefl. |
This reverts commit 3d394b2.
|
PR updated per your requests. Thanks for the quick response - and thanks for writing |
This PR enables
django-qto run correctly in namespaced Docker containers, by differentiating between a randomly assigned unique cluster ID and the cluster's PID.Motivation
Before this PR,
django-qused the PID of the./manage.py qclustercommand as the ID of the cluster. This caused the system to become confused when multiple clusters were running inside namespaced Docker containers. Because all theqclusterprocesses had the same PID.(Namespaced Docker containers have their PID space separate from the PID space of the host OS. So the entrypoint process always has PID 1.)
Implementation
This PR separates a
Cluster.pidfield from aCluster.cluster_idfield. The latter is populated by a random UUID at cluster initialization. Other methods are adapted to depend oncluster_id(which is guaranteed to be unique) rather thanpid(which is very likely to be non-unique in certain deployment schemes).