Fix high cpu sasl loop after disconnect #2032
Conversation
…L response Currently disconnects went unnoticed and forced the client into a tight loop while waiting for an answer on a closed socket. This resulted in high CPU usage in the client and the inability to reconnect.
stefanseufert
force-pushed
the
FixHighCpuSaslLoopAfterDisconnect
branch
from
308b560
to
11ba647
Compare
|
Rebased and squashed fixes into original commit as per contributing guidelines |
|
@edenhill anything else you want me to do here? According to your comments in various threads, 11.6 seems to be close and I really hope to see this included. |
|
0.11.6 would be a stretch, it is way past its deadline and needs to get out the door. |
|
This is a really nasty bug for everyone using SASL auth. A restart of any broker node will immediately make the clients go 100% CPU per thread. I've seen boxes logging CPU loads in excess of 350 due to this issue, making them completely inaccessible. Since this also affects producers it will result in message loss (on the consumer side it "only" prevents the affected threads from receiving messages). There are no known workarounds and there is no way to detect the problem in the calling code. As far as I can see all platforms are affected. Not sure how much worse a bug could be? The only reason, why this hasn't caught much more attention seems to be the fact that auth isn't used that commonly. The fix on the other hand seems trivial ans will only affect those people using auth. ... but it certainly is your call. |
|
@stefanseufert Valid points, I'll get back to you on this tomorrow. |
|
Thanks alot for this! |
…L response (#2032) Currently disconnects went unnoticed and forced the client into a tight loop while waiting for an answer on a closed socket. This resulted in high CPU usage in the client and the inability to reconnect.
Should fix #1937