Re: Anything left for 2.25.0?

[Gary Gregory]

Thank you for the update.

Do you think it's possible to add a test to avoid regressions?

Gary

On Mon, Feb 3, 2025, 13:47 Volkan Yazıcı  wrote:

> Yes, see the `InternalLoggerRegistry` deadlock reported in #3399
> . The fix is ready
> in
> #3418  and verified by
> users in #3399. I asked Piotr for one final review. He will be back from
> FOSDEM this week. I prefer waiting for another week so that #3418 can be
> shipped with `2.25.0`.
>
> On Mon, Feb 3, 2025 at 6:45 PM Matt Sicker  wrote:
>
> > Do we have anything remaining to complete before we can release 2.25.0? I
> > can cut a release candidate if we’re ready to go.
>

Re: Anything left for 2.25.0?

[Volkan Yazıcı]

Yes, see the `InternalLoggerRegistry` deadlock reported in #3399
. The fix is ready in
#3418  and verified by
users in #3399. I asked Piotr for one final review. He will be back from
FOSDEM this week. I prefer waiting for another week so that #3418 can be
shipped with `2.25.0`.

On Mon, Feb 3, 2025 at 6:45 PM Matt Sicker  wrote:

> Do we have anything remaining to complete before we can release 2.25.0? I
> can cut a release candidate if we’re ready to go.

Anything left for 2.25.0?

[Matt Sicker]

Do we have anything remaining to complete before we can release 2.25.0? I can 
cut a release candidate if we’re ready to go.

Re: Anything left for 2.25.0?

[Piotr P. Karwasz]

Hi Gary,

On 3.02.2025 19:51, Gary Gregory wrote:

Do you think it's possible to add a test to avoid regressions?


It is hard to reproduce the problem in a test. There are actually two 
problems:


1. Something happens during the creation of a new logger that stops the 
process in some environments until some other lock is acquired. Since 
there are many customizable Log4j elements like property sources, we 
never found out what it is, we just corrected the problem of one user 
and he confirmed it is solved.


2. Due to a bug in `InternalLoggerRegistry`, the registry cached only 
the first instance of a given `Logger`. If that logger was GC-ed, all 
requests for a new logger of that name created a new logger. We could 
add a benchmark for that, but we don't have a framework to test for 
performance regressions.


Piotr