Debug print loaded extensions if PUMA_DEBUG is set #3030
Conversation
Not sure if this is the best place for it, but trying to move it to a separate method seemed challenging as this appears to be called for multiple places. Closes puma#3020
This needs to be anchored to the end of the string.
MSP-Greg
force-pushed
the
debug-print-loaded-extensions-3020
branch
from
a905543
to
62c552a
Compare
|
Thank you for the PR.
That's the messy part. Placing the code in If one starts Puma with a control server using a socket connection, the control server calls If one is using a 'single' setup (with no workers), one can log from the main process, but if one is using workers (a clustered setup), it would probably be best to log from Worker index 0 (Worker0). As an example, when running clustered, I've got a patch for that, not sure about how often/when to log the Worker0 extensions, but it does log correctly when the server starts. Also, the code below is using file extensions for system libraries for macOS & Windows, which are different from the file extensions used by Ruby extensions ( `$LOADED_FEATURES.grep(/\.(so|dylib|dll)\z/i).each { |f| log_writer.debug(" #{f}") }`So, in the patch I've got I used: re_ext = /\.#{RbConfig::CONFIG['DLEXT']}\z/i
$LOADED_FEATURES.grep(re_ext).each { |f| @log_writer.debug(" #{f}") }Lastly, I have yet to try any of the code with non-MRI Rubies, not sure what they'll do. Additionally, two items, first, when and how to log Worker0 on restarts, and second, I can't recall if $LOADED_FEATURES resolves symlinks or not. People may prefer to see resolved (actual) paths. I can push the patch if you'd like... Again, thanks. |
Not sure if this is the best place for it, but trying to move it to a separate method seemed challenging as this appears to be called for multiple places. Closes puma#3020
This needs to be anchored to the end of the string.
|
@MSP-Greg Please do push your patch, it's definitely better than what I submitted. |
|
Pushed the commit, the CI error is one of those intermittent issues that we have. We normally squash PR commits, I always have 'merge' commits when rebasing PR branches. Must be doing something wrong... |
|
@jeremyevans I made a mess of the branch. Can't seem to force push to the branch. I think the last commit Merge branch 'debug-print-loaded-extensions-3020... and the first set of commits (Debug print loaded extensions if PUMA_DEBUG is set and Fix regexp used for determining loading extensions) need to be deleted? I hate it when this happens. Sorry. |
|
No big deal. I think the best approach here is to delete this PR, and for you to push a PR with your changes. Does that make sense? |
|
Again, sorry for the mess in your branch. I created PR #3036. |
Description
This implements the feature requested in #3020, printing the loaded extensions if PUMA_DEBUG is set.
Not sure if this is implemented in the best place, but trying to move it to a separate method seemed challenging as the existing method appears to be called for multiple places, and I wanted to avoid redundancy.
Your checklist for this pull request
I didn't do this because there are no existing tests for printing the listening socket.
[ci skip]to the title of the PR.#issue" to the PR description or my commit messages.I didn't do this as there is no existing documentation for PUMA_DEBUG, beyond a comment in the systemd doc.
CI should be able to check this. I didn't run the tests nor Rubocop locally, though all CI jobs that ran in my fork passed.