Jason918 commented on issue #12812: URL: https://github.com/apache/pulsar/issues/12812#issuecomment-972459284
> * Newer ZK versions have better performance (eg: better txn pipelining in leader) and ways to configure group commit on the transaction log to improve throughput. > * We should do a test with a 3/5 nodes ZK cluster, not just a single server My bad, we did this perf test on zk version 3.6.3, with 3 nodes cluster. Zk 3.4.6 is the version we use in production. > * It would be good to simulate a mix write/read scenario, perhaps adding a YCSB driver (https://github.com/brianfrankcooper/YCSB) based on metadata store API. Totally agreed. But it's a bit tricky to simulate perfect mix write/read scenario. And we have a perf cluster ready for million topic scenario. So We plan to just test it with a quick implementation of `BatchedMetadataStore`. > Finally, we should consider what happens when batching reads, of which we don't know upfront the collective size. If we are exceeding the ZK max frame size, we should be able to automatically switch back to non-batched read. Thanks for the reminding, automatic fallback mechanism will be added. -- 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]
