improve informer test

lib/kubeclient/informer.rb

@@ -8,6 +8,7 @@ def initialize(client, resource_name, reconcile_timeout: 15 * 60, logger: nil)
       @logger = logger
       @cache = nil
       @started = nil
+      @stopped = false
       @watching = []
     end
 
@@ -21,22 +22,34 @@ def watch(&block)
 
     # not implicit so users know they have to `stop`
     def start_worker
+      @stopped = false
       @worker = Thread.new do
         loop do
-          fill_cache
-          watch_to_update_cache
-        rescue StandardError => e
-          # need to keep retrying since we work in the background
-          @logger&.error("ignoring error during background work #{e}")
-        ensure
-          sleep(1) # do not overwhelm the api-server if we are somehow broken
+          begin
+            fill_cache
+            watch_to_update_cache
+          rescue StandardError => e
+            # need to keep retrying since we work in the background
+            @logger&.error("ignoring error during background work #{e}")
+          ensure
+            sleep(1) # do not overwhelm the api-server if we are somehow broken
+          end
+          break if @stopped
         end
       end
-      sleep(0.01) until @cache
+      sleep(0.01) until @cache || @stopped
     end
 
     def stop_worker
-      @worker&.kill # TODO: be nicer ?
+      @stopped = true
+      [@waiter, @worker].compact.each do |thread|
+        begin
+          thread.run # cancel sleep so either the loop sleep or the timeout sleep are interrupted
+        rescue ThreadError
+          # thread was already dead
+        end
+        thread.join
+      end
     end
 
     private
@@ -69,7 +82,7 @@ def watch_to_update_cache
       stop_reason = 'disconnect'
 
       # stop watcher without using timeout
-      Thread.new do
+      @waiter = Thread.new do
         sleep(@reconcile_timeout)
         stop_reason = 'reconcile'
         @watcher.finish
@@ -88,6 +101,14 @@ def watch_to_update_cache
         @watching.each { |q| q << notice }
       end
       @logger&.info("watch restarted: #{stop_reason}")
+
+      # wake the waiter unless it's dead so it does not hang around
+      begin
+        Thread.pass # make sure we get into the sleep state of the waiter
+        @waiter.run
+      rescue ThreadError # rubocop:disable Lint/SuppressedException
+      end
+      @waiter.join
     end
   end
 end

test/test_informer.rb

@@ -4,8 +4,7 @@
 require 'stringio'
 require 'logger'
 
-# tests with_retries in kubeclient.rb
-class RetryTest < MiniTest::Test
+class TestInformer < MiniTest::Test
   def setup
     super
     skip if RUBY_ENGINE == 'truffleruby' # TODO: race condition in truffle-ruby fails random tests
@@ -87,14 +86,14 @@ def test_restarts_on_error
       status: 200
     )
     slept = []
-    informer.stubs(:sleep).with { |x| slept << x; sleep(0.01) }
+    informer.stubs(:sleep).with { |x| slept << x; sleep(0.02) }
 
     with_worker do
       assert_equal(['a'], informer.list.map { |p| p.metadata.name })
-      sleep(0.05)
+      sleep(0.2) # should give us 5+ restarts (each timeout is 1 sleep and 1 sleep before restart)
     end
 
-    assert slept.size >= 2, slept
+    assert slept.size >= 4, slept
     assert_requested(list, at_least_times: 2)
     assert_requested(watch, at_least_times: 2)
   end
@@ -131,10 +130,14 @@ def test_can_watch_watches
   def test_timeout
     timeout = 0.1
     informer.instance_variable_set(:@reconcile_timeout, timeout)
-    stub_list
+    list = stub_list
     Kubeclient::Common::WatchStream.any_instance.expects(:finish)
-    stub_request(:get, %r{/v1/watch/pods})
+    watch = stub_request(:get, %r{/v1/watch/pods})
+
     with_worker { sleep(timeout * 1.9) }
+
+    assert_requested(list)
+    assert_requested(watch)
   end
 
   private