[Bug] MetadataStoreTopicPoliciesService: policy change events may be lost or transiently return null
- Dominant language
- Java
- Stars
- 15.3k
- Forks
- 3.8k
- Avg merge
- 1d 14h
- Merged PRs (30d)
- 160
Description
### Search before reporting
- [x] I searched in the [issues](https://github.com/apache/pulsar/issues) and found nothing similar.
### Read release policy
- [x] I understand that [unsupported versions](https://pulsar.apache.org/contribute/release-policy/#supported-versions) don't get bug fixes. I will attempt to reproduce the issue on a supported version of Pulsar client and Pulsar broker.
### User environment
built from master
### Issue Description
1. Change events lost after ZK reconnection
The service relies on ZK watches. Any policy update that occurs during a transient ZK disconnection is silently dropped — downstream listeners never receive `onUpdate`. `SystemTopicBasedTopicPoliciesService` avoids this by resuming its `__change_events` reader from the last consumed offset.
2. Transient null during policy update
`handleNotification` calls `cache.invalidate(path)` before re-fetching. During this window, `getTopicPoliciesIfExists` (which calls `MetadataCache.getIfCached`) returns `null` instead of the old value. `SystemTopicBasedTopicPoliciesService` always overwrites in-memory entries atomically, so callers see old or new — never null.
In single-cluster deployments (`metadataStoreUrl == configurationMetadataStoreUrl`), both listeners fire on the same store instance, calling `invalidate` twice and doubling the null window.
### Error messages
```text
```
### Reproducing the issue
Any update to the `MetadataStoreTopicPoliciesService` topic policies.
### Additional information
_No response_
### Are you willing to submit a PR?
- [x] I'm willing to submit a PR!
Contributor guide
Research direction
Start by reading MetadataStoreTopicPoliciesService, especially handleNotification and getTopicPoliciesIfExists, then compare its behavior with SystemTopicBasedTopicPoliciesService. Reproduce a topic-policy update across a ZooKeeper reconnection and in a single-cluster deployment; done means updates are not lost and callers do not transiently observe null.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- java
- Domain
- backend, distributed-systems
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Active
- Clarity
- Mostly clear
- Newbie friendliness
- 48/100