Is your feature request related to a problem? Please describe.
Currently, the EventProcessor in azure-eventhub auto-generates an owner_id using str(uuid.uuid4()) when calling receive/receive_batch. This becomes problematic in environments with frequent pod/worker restarts (e.g., K8s), as each restart results in a new owner_id, forcing unnecessary partition rebalancing and checkpoint contention.
Describe the solution you'd like
Add an optional parameter (owner_id: Optional[str] = None) to the said methods. If provided, this value would override the auto-generated UUID. This allows users to pass a persistent identifier (e.g., pod FQDN) to:
- Reduce partition ownership juggling during brief restarts.
- Improve traceability with meaningful IDs.
Example usage:
client.receive_batch(
on_event_batch=on_event_batch,
on_error=on_error,
owner_id=socket.getfqdn(),
)Describe alternatives you've considered
As for now, I monkey patch EventProcessor.
Additional context
In my deployment, workers restart hourly (due to a custom proxy requiring credentials refresh) and use a custom MongoDB (is Web scale) checkpoint store with the following behavior:
- Graceful shutdowns do not release partitions.
- On restart, the same worker reclaims its previous partitions. But it only works if the
owner_id is stable.
Explicit control over owner_id would greatly help in this solution.
Also, as I understand it, the Java SDK allows you to explicitly specify owner_id.
I’d be happy to implement this myself if the feature is approved.
Thank you in advance.
Is your feature request related to a problem? Please describe.
Currently, the
EventProcessorinazure-eventhubauto-generates anowner_idusingstr(uuid.uuid4())when callingreceive/receive_batch. This becomes problematic in environments with frequent pod/worker restarts (e.g., K8s), as each restart results in a newowner_id, forcing unnecessary partition rebalancing and checkpoint contention.Describe the solution you'd like
Add an optional parameter (
owner_id: Optional[str] = None) to the said methods. If provided, this value would override the auto-generated UUID. This allows users to pass a persistent identifier (e.g., pod FQDN) to:Example usage:
Describe alternatives you've considered
As for now, I monkey patch
EventProcessor.Additional context
In my deployment, workers restart hourly (due to a custom proxy requiring credentials refresh) and use a custom MongoDB
(is Web scale)checkpoint store with the following behavior:owner_idis stable.Explicit control over
owner_idwould greatly help in this solution.Also, as I understand it, the Java SDK allows you to explicitly specify
owner_id.I’d be happy to implement this myself if the feature is approved.
Thank you in advance.