On 09/11/2021 18:33, Mark Thomas wrote:
On 09/11/2021 10:28, Michael Osipov wrote:
Am 2021-11-09 um 10:40 schrieb Mark Thomas:
All,
I am seeing some fairly consistent crashes when running the APR tests
on Windows for 10.0.x. I'm going to spend some time trying to track
down the root cause before I tag the next 10.0.x release.
That's surprising, all of a sudden?
Yes, but it seems related to me switching to a new, clean built Windows
VM for pre-release testing and release building. The VM is on different
(slower) hardware so I suspect I have just uncovered a timing issue.
Can you share the dump/backtrace of the crash while you are analyzing
the issue?
I have various dump files but haven't got anything useful out of them (I
suspect my lack of knowledge is the primary factor here).
I have managed to make progress though. It looks like the Connector
fails to initialise, we don't handle that properly and when it tries to
continue with a null socket, APR gets a tad upset.
I'm hopeful I'll get to the bottom of this later today.
Fixed it. The short version is:
- default time wait delay and ephemeral port range on Windows meant unit
tests were running out of ports
- APR wasn't handling the init() failure and carrying on with an invalid
socket
- as soon as any code tried to use the invalid socket, the JVM crashed
I've fixed the init() failure handling so we now see test failures
rather than JVM crashes.
I've adjusted the OS settings for the VM where I run the tests so the
time wait delay is shorter so the test process doesn't run out of ports.
The 10.0.x tag should follow later today.
Mark
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org
For additional commands, e-mail: dev-h...@tomcat.apache.org
