lhotari commented on PR #17870: URL: https://github.com/apache/pulsar/pull/17870#issuecomment-1262048972
> If the internal client uses a lot of threads or the exception handling is not good (such as frequent exception thread). It will affect the throughput of the entire broker. please elaborate more on this. What do you mean with "internal client uses a lot of threads" or "exception handling is not good"? Please provide examples. I can see why this is needed since the Pulsar code is not non-blocking. Running synchronized methods on Netty IO threads is harmful since it will become a performance bottleneck. We should make these problems visible and eventually address them. To detect blocking code, there's https://github.com/reactor/BlockHound that could be used. [Netty comes with BlockHound integration](https://github.com/netty/netty/blob/4.1/common/src/main/resources/META-INF/services/reactor.blockhound.integration.BlockHoundIntegration). BlockHound could be used in unit tests and integration tests to ensure that blocking method calls aren't done on Netty IO threads. With Pulsar we'd have long ways to fullfil this requirement. This just shows that we need a better concurrency model to guide the future development of Pulsar in Pulsar 3.0 and beyond that. The current situation is more or less a mess from a concurrency model perspective. -- 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]
