Thank you, Matt. I'll get back to you after I've written some unit tests.
Tim On Mon, Jan 25, 2021 at 10:37 AM Matt Sicker <[email protected]> wrote: > Go to https://github.com/apache/logging-log4j2/compare and click the > "compare across forks" link at the top to make a PR from your forked > repo. > > On Mon, 25 Jan 2021 at 12:15, Tim Perry <[email protected]> wrote: > > > > Matt, > > > > Thanks for clarifying. > > > > I'd be happy to write some tests and submit a PR. How should I submit a > > pull request? I don't think I can do it from the github repo I linked to. > > > > Thanks, > > Tim > > > > > > On Mon, Jan 25, 2021 at 9:59 AM Matt Sicker <[email protected]> wrote: > > > > > I'm just saying that I don't think any of the developers here would > > > object to functional changes you'd like to introduce here, especially > > > if you think this change makes sense for users other than yourself. > > > > > > If you submit your changes as a PR (and preferably add automated tests > > > if possible), we'd be happy to merge! > > > > > > On Mon, 25 Jan 2021 at 11:51, Tim Perry <[email protected]> wrote: > > > > > > > > Matt, et al., > > > > > > > > I agree the deployment patterns you mention are more common and I > > > wouldn't > > > > start a new project embedding log4j in each WAR. However, I'm trying > to > > > > upgrade some old spring apps and my hands are tied on the deployment > > > > pattern. > > > > > > > > As mentioned in my comment on LOG4J2- 2624, the changes I proposed > don't > > > > fundamentally change the lifecycle hooks for web modules and each > class > > > > loader will still have its own independent log4j config. The changes > just > > > > provide the ability to stop log4j a little later. To me, this is a > low > > > risk > > > > change since the default behaviour is unchanged. If my approach of > > > passing > > > > the Log4jWebLifeCycle around in the ServletContext is unacceptable, > I'm > > > > happy to revisit the code and come up with another solution. Here > are the > > > > proposed changes: > > > > > > > > https://github.com/perry2of5/logging-log4j2/commit/56455af53920d69ff7a49a63c5bbf38773069e8d > > > > > > > > I'd really like to fix these bugs. If you are telling me there are > more > > > > important things for the log4j team to work on and that there is no > > > > interest from the log4j committers to make these changes, I can > accept > > > > that. However, I think these changes would be welcomed by some log4j > > > users > > > > and I hope one of the log4j committers will work with me on solving > these > > > > issues. > > > > > > > > Thanks, > > > > Tim > > > > > > > > > > > > > > > > > > > > On Sun, Jan 24, 2021 at 6:29 PM Matt Sicker <[email protected]> > wrote: > > > > > > > > > I'm not sure how much any of the devs here use the log4j-web module > > > > > anymore (seems more common to use fat jars or one app per servlet > > > > > container instance at least), so it's hard to say about any > > > > > idiosyncrasies. The main purpose of the lifecycle hooks for web > > > > > modules is to allow each class loader to have its own independent > > > > > log4j config, though I'm not sure how common that deployment > pattern > > > > > is anymore. There are alternative strategies such as hooking into > the > > > > > server code itself so that logging can shutdown with the server > rather > > > > > than the individual applications, but that's a different use case. > > > > > > > > > > As for design ideas, I think I had initially wanted to refactor the > > > > > web context API to mimic how Spring Framework registers itself in > the > > > > > ServletContext, though I never got around to doing that, and now I > > > > > typically use JVM-global logging configurations instead, so I never > > > > > revisited that. > > > > > > > > > > On Fri, 22 Jan 2021 at 11:53, Tim Perry <[email protected]> > wrote: > > > > > > > > > > > > Hello, > > > > > > > > > > > > I'd like to help fix LOG4J2-2624 and LOG4J2-1606. How can I help? > > > > > > > > > > > > To me, the challenge is to ensure log4j is initialized the very > first > > > > > time > > > > > > the ServletContext is provided to any object during application > > > loading > > > > > and > > > > > > startup and to stop log4j during the very last event or execution > > > hook a > > > > > > servlet 3.0 container exposes. Right now using the servlet 3.0 > > > > > > auto-configuration stops log4j too soon in some cases and using > the > > > > > servlet > > > > > > 2.5 configuration starts log4j too late in some cases. > > > > > > > > > > > > FWIW, I have posted a proposed fix in > > > > > > https://issues.apache.org/jira/browse/LOG4J2-1606. I'm not sure > if > > > it is > > > > > > the correct way to go. For one thing, it puts the > Log4jWebLifeCycle > > > > > > initializer into the ServletContext so that another object can > grab > > > it > > > > > and > > > > > > use it during log4j shutdown. Somewhere in the log4j dev > archives I > > > saw a > > > > > > note about moving data out of the ServletContext so that it > can't be > > > > > > overwritten. I'm not sure if my solution would need to be > modified or > > > > > > abandoned in light of this. > > > > > > > > > > > > The code changes I posted are based on a custom log4j-web > artifact I > > > > > > created for a client. It works for them on their Tomcat 8.x > servers. > > > > > > However, I'm not sure if I'm relying on any idiosyncratic > behaviour > > > of > > > > > > Tomcat or if there are earlier or later servlet container events > / > > > hooks > > > > > > that can be used to trigger configuration to happen earlier on > > > startup or > > > > > > stop log4j later when an application is stopped. > > > > > > > > > > > > If I can be of any help fixing these issues, I'd like to help. > > > > > > > > > > > > I've gotten a lot of good use out of log4j over the years. Thank > you > > > for > > > > > > maintaining it. > > > > > > > > > > > > Tim > > > > > > > > >
