[18.0] Expose jobrunner metrics by yvaucher · Pull Request #1029 · OCA/queue

yvaucher · 2026-11-04T16:53:36Z

This is based on #976

This turns out to be fairly simple. A few things worth noting:

The http server runs in the same thread and event loop as listen/notify job events, so there are no concurrency issues while accessing the channels stats

I added the set of jobs waiting for dependencies on each channel for completeness and convenient access to that stat, similar to what is done for failed jobs.

I used the prometheus-client external library. It's currently required even if the metrics server is not configured. This could be improved by mocking the lib and making it mandatory only for users enabling the metrics server. Or make it a required dependency (which is my preference, for simplicity).

This still needs tests and docs.

Test by setting ODOO_QUEUE_JOB_METRICS_PORT=8080 before running Odoo then watch curl -s http://localhost:8080.

oca-cla-bot · 2026-11-04T16:53:42Z

All commit authors signed the Contributor License Agreement.

OCA-git-bot · 2026-11-04T16:53:43Z

Hi @guewen, @sbidoul,
some modules you are maintaining are being modified, check this out!

sbidoul

Thanks! A few thoughts after a first read.

sbidoul · 2026-11-04T17:19:03Z

+        """Register gauges for a specific database in this channel."""
+        self._unregister_metrics()
        self._metrics_labels = {
+            "db": self.db_name,


Note to self: I think self.db_name can be None if this is the global ChannelManager. Test that this works.

It renders db="None". Not ideal. So maybe __all__?

sbidoul · 2026-11-04T17:34:27Z

+
+    def _capacity_gauge(self) -> float:
+        if self.capacity is None:
+            return math.inf


Note to self: test that prometheus client renders math.inf correctly

It renders +Inf. Now need to see what prometheus and grafana do with that.

sbidoul · 2026-11-05T11:22:34Z

        self._channel_manager_by_db[db_name] = channel_manager
        self._channel_managers = [self._server_wide_channel_manager]
        self._notify_existing_jobs(db, channel_manager)
+        self.register_metrics()


This one should not be here. Because the server wide channel manager is a singleton without db_name, we don't need to re-register metrics for each db. It should move somewhere near the creation of the server wide ChannelManager

sbidoul · 2026-11-05T11:42:31Z

Can you also add prometheus-client to the manifest external deps?

amh-mw · 2026-11-05T12:19:03Z

Can you also add prometheus-client to the manifest external deps?

Late to the party here and not familiar with prometheus, but is it appropriate to hardcode a metrics collection agent and additional python dependencies into the main queue_job module? I'm running on AWS, won't use this, and would prefer not to install a python package that won't get used at best, and will steal CPU cycles at worst.

amh-mw · 2026-11-05T12:23:36Z

Late to the party here and not familiar with prometheus, but is it appropriate to hardcode a metrics collection agent and additional python dependencies into the main queue_job module? I'm running on AWS, won't use this, and would prefer not to install a python package that won't get used at best, and will steal CPU cycles at worst.

It looks like Prometheus metrics collected by the CloudWatch agent, so that removes some decent portion of my argument against.

yvaucher · 2026-11-05T14:03:12Z

@amh-mw on our wish list we have the extraction of the metrics in a separate module and the possibility to plug other observability tools such as opentelemetry.

The state of this PR is still in draft. Your input are welcome nevertheless, and I agree with you that we might not want to enforce the usage of Prometheus.

This adds a 'db' label to channel gauges to track which databases have jobs in each channel. This allows for better monitoring when multiple databases share the same channel structure.

Not sure that's useful. Let's wait for the actual use case.

OCA-git-bot added mod:queue_job Module queue_job series:18.0 labels Nov 4, 2026

yvaucher force-pushed the 18.0-metrics-sbi-ypt branch from 9669136 to b7c12b7 Compare November 4, 2026 16:57

sbidoul reviewed Nov 4, 2026

View reviewed changes

sbidoul mentioned this pull request Nov 4, 2026

Expose jobrunner metrics #976

Closed

yvaucher force-pushed the 18.0-metrics-sbi-ypt branch 2 times, most recently from 3ddb459 to 672ba4d Compare November 5, 2026 10:37