From 3ccd51253dbe32b1bb5d85aa5cdfa59a1b9a70c2 Mon Sep 17 00:00:00 2001
From: Philipp Thun <philipp.thun@sap.com>
Date: Thu, 27 Aug 2026 14:09:46 +0200
Subject: [PATCH] Make drain plugin drain spec deterministic

The 'works off all remaining jobs before exiting' example raced the
worker loop against TERM signal delivery. On slow/contended runners the
drain flag (exit_on_complete) flipped late, so work_off ran extra times
and the count inflated past 3.

Block the worker after its first work_off call until the test confirms
the drain TERM handler has set exit_on_complete, then release it. This
removes the timing dependency and makes the call count deterministic.
---
 .../local_worker_drain_plugin_spec.rb         | 28 +++++++++++++++----
 1 file changed, 23 insertions(+), 5 deletions(-)

diff --git a/spec/unit/lib/delayed_job/local_worker_drain_plugin_spec.rb b/spec/unit/lib/delayed_job/local_worker_drain_plugin_spec.rb
index 5fb09db1cac..6a873c8a0cf 100644
--- a/spec/unit/lib/delayed_job/local_worker_drain_plugin_spec.rb
+++ b/spec/unit/lib/delayed_job/local_worker_drain_plugin_spec.rb
@@ -23,19 +23,37 @@
     after { Delayed::Worker.sleep_delay = Delayed::Worker::DEFAULT_SLEEP_DELAY }
 
     it 'works off all remaining jobs in the queue before exiting' do
-      work_off_calls = Queue.new
+      work_off_count = 0
+      worker_started = Queue.new
+      resume_worker = Queue.new
       allow(worker).to receive(:work_off) do
-        work_off_calls.push(:called)
-        work_off_calls.size < 3 ? [1, 0] : [0, 0]
+        work_off_count += 1
+        if work_off_count == 1
+          worker_started.push(:yes)
+          # Block after the first call until the test has confirmed the TERM handler ran,
+          # so the number of work_off calls does not depend on signal-delivery timing.
+          resume_worker.pop
+        end
+        work_off_count < 3 ? [1, 0] : [0, 0]
       end
 
       worker_thread = Thread.new { worker.start }
-      work_off_calls.pop # wait until worker has started
+      worker_started.pop # wait until worker has started
+
       Process.kill('TERM', Process.pid)
+      # Wait (max. 1s) until the drain TERM handler has actually flipped the flag.
+      deadline = Process.clock_gettime(Process::CLOCK_MONOTONIC) + 1
+      until Delayed::Worker.exit_on_complete
+        raise 'Timed out waiting for Delayed::Worker.exit_on_complete' if Process.clock_gettime(Process::CLOCK_MONOTONIC) >= deadline
+
+        sleep 0.01
+      end
+      resume_worker.push(:now)
+
       worker_thread.join(5)
 
       expect(worker_thread.alive?).to be(false)
-      expect(work_off_calls.size).to eq(3)
+      expect(work_off_count).to eq(3)
     end
   end
 end