Fix APNs P8 HTTP2 error recovery (#734)

SixiS · web-flow · commit 6792e8c04cf6 · 2025-07-01T07:01:35.000-03:00
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -5,6 +5,7 @@
 * Support for proxying HTTP requests [\#728](https://github.com/rpush/rpush/pull/728) ([jaspreet-3911](https://github.com/jaspreet-3911))
 * Fix active record logger dependency for older rails versions that was breaking the build [\#731](https://github.com/rpush/rpush/pull/731) ([SixiS](https://github.com/sixis))
 * Specify ostruct dependency ([not bundled with ruby since 3.5](https://github.com/ruby/ruby/blob/4eaa245fccd3dd9a61fe1b5f114a6fb47907640a/lib/bundled_gems.rb#L18)) [\#740](https://github.com/rpush/rpush/pull/740) ([grekko-headacy](https://github.com/grekko-headacy))
+* Fix APNs P8 HTTP2 error recovery [\#734](https://github.com/rpush/rpush/pull/734) ([kjvarga](https://github.com/kjvarga)) ([SixiS](https://github.com/SixiS))
 
 [Full Changelog](https://github.com/rpush/rpush/compare/v9.2.0...HEAD)
 
diff --git a/lib/rpush/daemon/apnsp8/delivery.rb b/lib/rpush/daemon/apnsp8/delivery.rb
@@ -25,11 +25,17 @@ def perform
 
           # Send all preprocessed requests at once
           @client.join(timeout: CLIENT_JOIN_TIMEOUT)
+          error = @client.check_for_error
+          raise error if error
         rescue NetHttp2::AsyncRequestTimeout => error
           mark_batch_retryable(Time.now + 10.seconds, error)
           @client.close
           raise
-        rescue Errno::ECONNREFUSED, SocketError, HTTP2::Error::StreamLimitExceeded => error
+        rescue OpenSSL::SSL::SSLError,
+               Errno::ECONNRESET,
+               Errno::ECONNREFUSED,
+               SocketError,
+               HTTP2::Error::StreamLimitExceeded => error
           # TODO restart connection when StreamLimitExceeded
           mark_batch_retryable(Time.now + 10.seconds, error)
           raise
diff --git a/lib/rpush/daemon/dispatcher/apnsp8_http2.rb b/lib/rpush/daemon/dispatcher/apnsp8_http2.rb
@@ -35,14 +35,34 @@ def cleanup
         def create_http2_client(app)
           url = URLS[app.environment.to_sym]
           options = { connect_timeout: DEFAULT_TIMEOUT }
-
           configure_proxy(options)
-
           client = NetHttp2::Client.new(url, options)
+
+          client.instance_eval do
+            @error_mutex = Mutex.new
+            @error = nil
+
+            def record_error(error)
+              @error_mutex.synchronize { @error = error }
+            end
+
+            def check_for_error
+              @error_mutex.synchronize do
+                return unless @error
+
+                error = @error.dup
+                @error = nil
+                error
+              end
+            end
+          end
+
           client.on(:error) do |error|
+            client.record_error(error)
             log_error(error)
             reflect(:error, error)
           end
+
           client
         end
       end
diff --git a/spec/unit/daemon/apnsp8/delivery_spec.rb b/spec/unit/daemon/apnsp8/delivery_spec.rb
@@ -49,5 +49,16 @@
         end
       end
     end
+
+    context 'with an HTTP2 client that raises Errno::ECONNRESET' do
+      before do
+        allow(http2_client).to receive(:call_async).and_raise(Errno::ECONNRESET)
+      end
+
+      it 'marks the batch as retryable' do
+        expect(batch).to receive(:mark_all_retryable).with(anything, anything)
+        expect { delivery.perform }.to raise_error(Errno::ECONNRESET)
+      end
+    end
   end
 end
diff --git a/spec/unit/daemon/dispatcher/apnsp8_http2_spec.rb b/spec/unit/daemon/dispatcher/apnsp8_http2_spec.rb
@@ -0,0 +1,54 @@
+require 'unit_spec_helper'
+
+describe Rpush::Daemon::Dispatcher::Apnsp8Http2 do
+  let(:app) { double(environment: "sandbox", apn_key: "my_key") }
+  let(:delivery_class) { double }
+  let(:notification) { double }
+  let(:batch) { double(mark_delivered: nil, all_processed: nil) }
+  let(:http2) { double(on: true) }
+  let!(:token_provider) { double }
+  let(:queue_payload) { Rpush::Daemon::QueuePayload.new(batch, notification) }
+  let(:dispatcher) { described_class.new(app, delivery_class) }
+
+  it 'constructs a new persistent connection' do
+    expect(NetHttp2::Client).to receive(:new).and_call_original
+    described_class.new(app, delivery_class)
+  end
+
+  describe 'dispatch' do
+    before do
+      allow(NetHttp2::Client).to receive_messages(new: http2)
+      allow(Rpush::Daemon::Apnsp8::Token).to receive_messages(new: token_provider)
+    end
+
+    it 'delivers the notification' do
+      delivery = double
+      expect(delivery_class).to receive(:new).with(app, http2, token_provider, batch).and_return(delivery)
+      expect(delivery).to receive(:perform)
+      dispatcher.dispatch(queue_payload)
+    end
+  end
+
+  describe 'error catching' do
+    let(:dispatcher) { described_class.new(app, Rpush::Daemon::Apnsp8::Delivery) }
+    let(:client) { dispatcher.instance_variable_get("@client") }
+    let(:notification1) { double('Notification 1', data: {}, as_json: {}).as_null_object }
+
+    before do
+      allow(batch).to receive(:each_notification) do |&blk|
+        [notification1].each(&blk)
+      end
+      allow_any_instance_of(Rpush::Daemon::Apnsp8::Delivery).to receive(:prepare_headers).and_return({})
+      allow(client).to receive(:socket_loop).and_raise(Errno::ECONNRESET)
+    end
+
+    it 'records and raises errors' do
+      expect(batch).to receive(:mark_all_retryable)
+      expect(client).to receive(:record_error).with(Errno::ECONNRESET).and_call_original
+      payload = Rpush::Daemon::QueuePayload.new(batch, notification)
+      expect do
+        dispatcher.dispatch(payload)
+      end.to raise_error(Errno::ECONNRESET)
+    end
+  end
+end
