bob-barrett commented on pull request #9902: URL: https://github.com/apache/kafka/pull/9902#issuecomment-771823264
> For instance, we could try all the resolved IPs before applying the backoff. However, we would still have to wait until the connection timeouts for each IP. @dajac That's right, and that's my concern with exhausting the currently-resolved IPs. The default connection setup timeout is 10 seconds. Even without applying the backoff, 10 seconds of delay per additional IP address is pretty noticeable. > Also, I suppose that we assume that closing connection is rather infrequent in normal circumstances so the cost of the DNS resolution is low. On the other hand, we relies on the DNS cache if the resolution happens really frequently. Am I getting this right? I think this is the right way to think about it, yes. There's already a DNS cache, which can be configured as needed. We're effectively adding our own additional cache (with a very long TTL) on top of that by preserving the resolved IPs. If a client has been connected to a server for hours or days and then disconnects, it makes complete sense to me to re-resolve DNS. ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
