poorbarcode commented on PR #132: URL: https://github.com/apache/flink-connector-pulsar/pull/132#issuecomment-6946422367
@ferenc-csaky I addressed points 2 and 4 from [your review](https://github.com/apache/flink-connector-pulsar/pull/132#issuecomment-6922710497) in 5e20116. **2. Broker compatibility.** The sink checks the broker version before interpreting a message’s transaction headers. Pulsar 3.x, 4.0.0–4.0.13, 4.1.x and 4.2.0–4.2.4 use the original coordinator commit handling. During a rolling upgrade, the topic owner can differ from the broker reporting the version, so the sink also checks the actual response for the header introduced with per-transaction semantics. I tested the case you described on 4.0.13 and 4.0.14: an earlier transaction stayed open on the same topic while a later committed transaction was restored. Recovery completed on both, and the committed record was delivered. **4. Admin calls on normal commits.** You were right about the previous implementation. A fresh transaction now goes directly to the coordinator. A successful first commit makes no Admin HTTP request. For transactions with saved message IDs, an ambiguous commit result or a checkpoint replay triggers verification on supported brokers. An already committed transaction is skipped. The exactly-once integration test now runs with an unreachable Admin URL and separate writer and committer operators. It passed and all expected records were delivered. The recovery lookup still requires access to the Admin endpoint and consume permission on the topic. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
