Merge branch 'puma-init-restart' into 'master'

Fix cleanup of prometheus files on server start

Closes #62862

See merge request gitlab-org/gitlab-ce!29788

4 files changed, 42 insertions, 8 deletions

diff --git a/config/initializers/7_prometheus_metrics.rb b/config/initializers/7_prometheus_metrics.rb
index 68f8487d377..54cdefc2a10 100644
--- a/config/initializers/7_prometheus_metrics.rb
+++ b/config/initializers/7_prometheus_metrics.rb
@@ -43,14 +43,21 @@ if !Rails.env.test? && Gitlab::Metrics.prometheus_metrics_enabled?
   end
 end
 
-Gitlab::Cluster::LifecycleEvents.on_master_restart do
+def cleanup_prometheus_multiproc_dir
   # The following is necessary to ensure stale Prometheus metrics don't
   # accumulate over time. It needs to be done in this hook as opposed to
   # inside an init script to ensure metrics files aren't deleted after new
   # unicorn workers start after a SIGUSR2 is received.
-  prometheus_multiproc_dir = ENV['prometheus_multiproc_dir']
-  if prometheus_multiproc_dir
-    old_metrics = Dir[File.join(prometheus_multiproc_dir, '*.db')]
+  if dir = ::Prometheus::Client.configuration.multiprocess_files_dir
+    old_metrics = Dir[File.join(dir, '*.db')]
     FileUtils.rm_rf(old_metrics)
   end
 end
+
+Gitlab::Cluster::LifecycleEvents.on_master_start do
+  cleanup_prometheus_multiproc_dir
+end
+
+Gitlab::Cluster::LifecycleEvents.on_master_restart do
+  cleanup_prometheus_multiproc_dir
+end
diff --git a/config/unicorn.rb.example b/config/unicorn.rb.example
index 4637eb8bc6e..581fde84c95 100644
--- a/config/unicorn.rb.example
+++ b/config/unicorn.rb.example
@@ -88,9 +88,21 @@ before_exec do |server|
   Gitlab::Cluster::LifecycleEvents.do_master_restart
 end
 
+run_once = true
+
 before_fork do |server, worker|
-  # Signal application hooks that we're about to fork
-  Gitlab::Cluster::LifecycleEvents.do_before_fork
+  if run_once
+    # There is a difference between Puma and Unicorn:
+    # - Puma calls before_fork once when booting up master process
+    # - Unicorn runs before_fork whenever new work is spawned
+    # To unify this behavior we call before_fork only once (we use
+    # this callback for deleting Prometheus files so for our purposes
+    # it makes sense to align behavior with Puma)
+    run_once = false
+
+    # Signal application hooks that we're about to fork
+    Gitlab::Cluster::LifecycleEvents.do_before_fork
+  end
 
   # The following is only recommended for memory/DB-constrained
   # installations.  It is not needed if your system can house
diff --git a/config/unicorn.rb.example.development b/config/unicorn.rb.example.development
index ae3dc2e37e1..9a02d5f1007 100644
--- a/config/unicorn.rb.example.development
+++ b/config/unicorn.rb.example.development
@@ -21,9 +21,21 @@ before_exec do |server|
   Gitlab::Cluster::LifecycleEvents.do_master_restart
 end
 
+run_once = true
+
 before_fork do |server, worker|
-  # Signal application hooks that we're about to fork
-  Gitlab::Cluster::LifecycleEvents.do_before_fork
+  if run_once
+    # There is a difference between Puma and Unicorn:
+    # - Puma calls before_fork once when booting up master process
+    # - Unicorn runs before_fork whenever new work is spawned
+    # To unify this behavior we call before_fork only once (we use
+    # this callback for deleting Prometheus files so for our purposes
+    # it makes sense to align behavior with Puma)
+    run_once = false
+
+    # Signal application hooks that we're about to fork
+    Gitlab::Cluster::LifecycleEvents.do_before_fork
+  end
 
   # The following is only recommended for memory/DB-constrained
   # installations.  It is not needed if your system can house
diff --git a/lib/gitlab/cluster/lifecycle_events.rb b/lib/gitlab/cluster/lifecycle_events.rb
index e0f9eb59924..8f796748199 100644
--- a/lib/gitlab/cluster/lifecycle_events.rb
+++ b/lib/gitlab/cluster/lifecycle_events.rb
@@ -11,6 +11,9 @@ module Gitlab
     # We have three lifecycle events.
     #
     # - before_fork (only in forking processes)
+    #     In forking processes (Unicorn and Puma in multiprocess mode) this
+    #     will be called exactly once, on startup, before the workers are
+    #     forked. This will be called in the parent process.
     # - worker_start
     # - before_master_restart (only in forking processes)
     #