On Mon, 2 Mar 2026 10:51:39 GMT, Daniel Jeliński <[email protected]> wrote:

> Raise the QUIC connection timeout in the H3MultipleConnectionsToSameHost test 
> above the jtreg timeout.
> 
> The test failed with a QUIC connection timeout, but the reason for the 
> timeout was not found in the logs. This kind of failure is usually caused by 
> one of:
> - packet loss; when that happens, the congestion controller slows down the 
> sending rate until the connection times out,
> - threading issues like races / deadlocks
> - overloaded machine; the threads might not get enough CPU time to respond 
> before the timeout expires.
> 
> Increasing the QUIC connection timeout should mitigate the packet loss and 
> CPU issues, and will provide additional information like thread / core dumps 
> to help with diagnosing threading issues.

Thanks. The `no error` line is normal, reported whenever a connection is 
closed. The interesting line is:

INFO: QUIC: QuicServerConnection(8) entering closing state, code 0x1 - 
java.io.IOException: QuicServerConnection(8): Too many probe time outs 
(APPLICATION: backoff 1024, duration PT1M44.32S, 
QuicRttEstimatorState[latestRttMicros=21059, minRttMicros=296, 
smoothedRttMicros=27687, rttVarMicros=18547, rttSampleCount=1945])

So, this particular failure was caused by packet loss. Please file a JBS issue. 
Thanks!

-------------

PR Comment: https://git.openjdk.org/jdk/pull/29993#issuecomment-3990106088

Reply via email to