Ruby 4.0.0 upgrade - unexpected graceful shutdown

[booleanbetrayal]

After migrating from ruby:3.4.7-alpine3.23 to ruby:4.0.0-alpine3.23, bundle exec prometheus_exporter --timeout 5 -b 0.0.0.0 -p 9394 produces a change in behavior where the exporter gracefully exits after a few seconds.

Has anyone observed this issue? This is currently blocking our Ruby 4.0.0 uptake.

I, [2026-01-07T23:41:24.398376 #1]  INFO -- : Starting prometheus exporter on 0.0.0.0:9394
[2026-01-07 23:41:24] INFO  WEBrick 1.9.2
[2026-01-07 23:41:24] INFO  ruby 4.0.0 (2025-12-25) [aarch64-linux-musl]
[2026-01-07 23:41:24] INFO  WEBrick::HTTPServer#start: pid=1 port=9394
10.20.10.102 - - [07/Jan/2026:23:41:51 UTC] "GET /ping HTTP/1.1" 200 4
- -> /ping
127.0.0.1 - - [07/Jan/2026:23:41:28 UTC] "POST /send-metrics HTTP/1.1" 200 2
- -> /send-metrics
127.0.0.1 - - [07/Jan/2026:23:41:32 UTC] "POST /send-metrics HTTP/1.1" 200 0
- -> /send-metrics
127.0.0.1 - - [07/Jan/2026:23:41:37 UTC] "POST /send-metrics HTTP/1.1" 200 0
- -> /send-metrics
127.0.0.1 - - [07/Jan/2026:23:41:57 UTC] "POST /send-metrics HTTP/1.1" 200 0
- -> /send-metrics
127.0.0.1 - - [07/Jan/2026:23:42:03 UTC] "POST /send-metrics HTTP/1.1" 200 0
- -> /send-metrics
127.0.0.1 - - [07/Jan/2026:23:42:02 UTC] "POST /send-metrics HTTP/1.1" 200 0
- -> /send-metrics
[2026-01-07 23:42:18] INFO  going to shutdown ...
10.20.10.79 - - [07/Jan/2026:23:42:18 UTC] "GET /metrics HTTP/1.1" 200 0
- -> /metrics
[2026-01-07 23:42:18] INFO  WEBrick::HTTPServer#start done.

Edit: Bundler was also updated from 4.0.1 to 4.0.3, in case that could be related.

[rlineweaver]

I think this is due to https://redmine.ruby-lang.org/issues/21812, and thus will hopefully be fixed in an upcoming version of Ruby 4.

prometheus_exporter starts the webrick server within a Thread and then calls sleep with no arguments. When handling a /metrics request, it uses backticks to shell out to getconf for determining memory consumption. This is the perfect recipe to trigger the ruby issue referenced above; when the getconf subprocess exits, the main sleep awakens, and the process exits.

[booleanbetrayal]

Thanks for this dive @rlineweaver !

[n-rodriguez]

@rlineweaver if the issue is triggered by shelling out to getconf, maybe we should move this call in the constructor (PAGESIZE won't change at runtime)?

[n-rodriguez]

Also it would be nice to add tests to exercise this part of the code. For now it's not covered by the test suite : https://github.com/discourse/prometheus_exporter/actions/runs/20598847761

[n-rodriguez]

@rlineweaver I confirm that 50e3aa9 fixes the bug 👍

Thank you for the diag! Thank to you I can now switch to Ruby 4.0 😃 (we won't have to wait for https://redmine.ruby-lang.org/issues/21812 to be fixed)