On 08/06/2023 09:57, Rémy Maucherat wrote:
On Thu, Jun 8, 2023 at 6:07 AM Rainer Jung <rainer.j...@kippdata.de> wrote:

Am 08.06.23 um 05:50 schrieb Rainer Jung:
The new TestKeyHeader fails for me very frequently for TC 9 and TC 8.5,
but not for 10.1 or 11.

For TC 9 it fails roughly 30-50% of the times I run it. It fails for
jsse and for tcnative and for a wide range of JDKs and RHEL/SLES Linux
versions.

The failure happens only for NIO2, not for NIO. When it fails, the test
case takes about 60 seconds longer than when it is OK. The failing test
case is often testValid - the first test case that runs - and sometime
the second one testTooLong01.

It seems the pattern for TC 8.5 is the same but the tests are still
running.

You probably already fixed it for 10.1 and 11, but for 9.0 and 8.5 the
backport of 44e7282b54 seems to be missing.

I'm not running into the issue but I added the commit to the two branches.

Thanks.

Arguably, reading and ignoring the response was fixing the symptom not the cause. The root cause being that:
- Tomcat was stopped while processing an HTTP upgrade to WebSocket
- the upgrade did not complete
- the error was swallowed so the connection wasn't cleaned-up

https://github.com/apache/tomcat/commit/a2c34d6569f2b959413b06852d7a957ff9e9c39d

(back-ported to all versions) should have addressed the root cause.

Mark

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org
For additional commands, e-mail: dev-h...@tomcat.apache.org

Reply via email to