Palaiologos1453 commented on issue #10982: URL: https://github.com/apache/rocketmq/issues/10982#issuecomment-6307546128
I would like to take this one. I have a local implementation against develop `57c2dc04` and am finishing the benchmark/registration comparison before opening the PR. The proposed first version is opt-in (`routeInfoCacheMaxBytes=0` by default), with a weighted, bounded cache populated only by queries. Effective route changes advance a revision; broker registration does not walk or eagerly rebuild cached topics. Unchanged registrations/heartbeats retain cached responses. Ordering KV data stays outside the cache and its value is checked on lookup; the two encoding variants are populated separately, and acting-master mode is part of validation. I checked the earlier cache removal in #4246. This version keeps serialization outside the route lock and adds no cache rebuild to the registration path. It deliberately uses a conservative global revision, so frequent changes to unrelated routes can reduce the benefit; an undersized cache can also churn. Both limitations will be documented in the PR. The full namesrv suite currently passes on JDK 8: 133 tests, no failures/errors/skips. The local JMH read-path comparison shows a substantial allocation reduction on cache hits; I will include the reproducible setup, registration results and scope limitations in the PR rather than extrapolate to production throughput. -- 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]
