puma memory configuration: per_worker_max_memory_mb or puma_ per_worker_max_memory_mb?

[xueshanf]

This repository puma.rb template uses 'puma_per_worker_max_memory_mb', however, all other gitlab documents use "per_worker_max_memory_mb" in puma.rb file. For example in puma

I have tried to allocate 2048 Mb memory using PUMA_PER_WORK_MAX_MEMORY_MB, but it doesn't seem to take effect:

{"severity":"WARN","time":"2025-08-16T23:27:00.441Z","memwd_max_strikes":5,"memwd_cur_strikes":6,"message":"rss memory limit exceeded","memwd_rss_bytes":1576566784,"memwd_max_rss_bytes":1572864000,"memwd_handler_class":"Gitlab::Memory::Watchdog::Handlers::PumaHandler","memwd_sleep_time_s":60,"pid":2555,"worker_id":"puma_3"}

This is the puma.rb file generated:

before_fork do
  # Signal to the puma killer
  Gitlab::Cluster::PumaWorkerKillerInitializer.start(@config.options, puma_per_worker_max_memory_mb: 2048, puma_master_max_memory_mb: 800) unless ENV['DISABLE_PUMA_WORKER_KILLER']

  # Signal application hooks that we're about to fork
  Gitlab::Cluster::LifecycleEvents.do_before_fork
end

Should the puma.rb configuration template uses per_worker_max_memory_mb, instead of max_per_worker_max_memory_mb?

Xueshan

[kkimurak]

Yes, the configuration file puma.rb was out of date and contained legacy variable names.
Also, puma-worker-killer was removed in v16.0 (see gitlab.com/gitlab-org/gitlab!118645), so the configuration parameter PUMA_PER_WORK_MAX_MEMORY_MB is no longer meaningful. Gitlab now uses watchdog by default, so we may want to check if we can take similar approach for it.
Ref: gitlab.com/gitlab-org/gitlab!103472