Avoid more cases of client wakeup when sending from sender / same thread #1765
Conversation
|
Some are just due to the extra parameter on calls, but it looks like a lot of the producer tests are (ironically?) failing with Haven't had a chance to look into it, but I'm guessing at least one of the bypassed As noted above, I'm not familiar enough with the threading and networking stuff going on here to really know how to debug those failures... |
|
Oops, accidentally wrote my earlier comment from my work account, so deleted that. Yeah, the threading/networking stuff here is tricky. @dpkp is the one who best understands, by far. |
|
I think #1767 is a simpler fix for the root issue here. Want to give it a try? |
|
Thanks @dpkp, I'll take a look. |
|
Thanks for the PR! I've updated / simplified a bit and fixed tests in #1769 , so I'm going to close this one. |
Extra fixes for #1760.
Note that I don't pretend to fully understand the threading/socket setup here - just extending the fix from #1761 to cover more cases, as my code seems to be hitting the
_maybe_refresh_metadata()path, and I spotted a few others that looked similar.I've done a couple of long running tests with this fix, and haven't run into any more
KafkaTimeoutErrors onwakeup().This still only covers producer calls. Is this issue/fix relevant to the consumer as well, or any other users of
KafkaClient?This change is