allthingssecurity commented on PR #30137: URL: https://github.com/apache/camel/pull/30137#issuecomment-6913256340
Thanks for checking it against the Curator sources, Claus. Pushed f8ebff579e7c. - **Process, JIRAs for the three cases:** agreed, they look real on main. The JIRAs are being filed (the two znode cases together, as they have the same cause and fix, and `startStartedRoute` as a core-level one, as you suggest), with fixes in preparation. A follow-up commit will name the keys in the world Javadoc, next to the three `dst.zookeeper.*` properties and in test-dst.adoc; this commit already says in the Javadoc and the property comments what each case does. - **[Suggestion] "Running the tests" in test-dst.adoc:** a paragraph for camel-zookeeper-master now gives the `-pl` module, the past fixes of the self-check (`MasterConsumer` of e670025a49a4^ for CAMEL-25315, `MasterRoutePolicy` of 706beec6d5ab^ for CAMEL-26540, 136ffd929263 for CAMEL-26802 and 28f3170dcc6a for CAMEL-26887 reverted, and the `unregisterAsListener()` of 5ecab86a58ba for CAMEL-26870 taken out by hand) and the three properties. Catalog copy updated too. It is a paragraph of its own, so it merges more easily with the other open worlds that touch that section. - **[Nit] `checkInterrupted`:** I first changed the fake to apply a write made after the interrupt while the client is connected (`ClientCnxn.submitRequest` queues the packet before it waits), and the default set did not stay unaffected: 4 of 5000 seeds failed. In those seeds the close interrupts a `getChildren` of the group's thread, the thread goes on to a create, and that create is applied after the group left, so the znode stays for the life of the session. That is the same bug as `stopDuringCreate`, reached another way. So the fake now has `writesAfterInterrupt`, and the world turns it on with `-Ddst.zookeeper.stopDuringCreate=true` only: the default set is as before (5000 of 5000), and with that property the fake also models this second way to the leftover znode (I have not rerun that strict mode for this commit). The Javadoc of `interrupt` says this, and that Curator queues a guaranteed delete again on an `InterruptedException`, which the fake's `IllegalStateException` does not show. - **[Nit] `SessionExpiredException`:** added to "Not modelled". The fake's client gets its new session as soon as the connection is back, so no operation sees the expired one. - **[Nit] A write that never reached the server:** not added. Under the timing rule of the harness a write takes effect at the start of its latency, and a write lost before it reaches the server looks to the client like a call made while the connection is down, which the world already drives (it fails with a connection loss and is retried). A seed choice would add a second way to the same outcome. Happy to add it if you see a case it would catch. Tests: `ZooKeeperMasterDstTest` passes 5000 of 5000 seeds, and the camel-zookeeper-master unit tests pass (5040 run with those seeds, 0 failures). _Claude Code on behalf of allthingssecurity_ -- 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]
