zhuqi-lucas commented on code in PR #533: URL: https://github.com/apache/yunikorn-core/pull/533#discussion_r1411766477
########## pkg/events/event_streaming.go: ########## @@ -0,0 +1,180 @@ +/* + Licensed to the Apache Software Foundation (ASF) under one + or more contributor license agreements. See the NOTICE file + distributed with this work for additional information + regarding copyright ownership. The ASF licenses this file + to you under the Apache License, Version 2.0 (the + "License"); you may not use this file except in compliance + with the License. You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License. +*/ + +package events + +import ( + "sync" + "time" + + "go.uber.org/zap" + + "github.com/apache/yunikorn-core/pkg/log" + "github.com/apache/yunikorn-scheduler-interface/lib/go/si" +) + +const defaultChannelBufSize = 1000 + +// EventStreaming implements the event streaming logic. +// New events are immediately forwarded to all active consumers. +type EventStreaming struct { + buffer *eventRingBuffer + stopCh chan struct{} + eventStreams map[*EventStream]eventConsumerDetails + sync.Mutex +} + +type eventConsumerDetails struct { + local chan *si.EventRecord + consumer chan<- *si.EventRecord + stopCh chan struct{} + name string + createdAt time.Time +} + +// EventStream handle type returned to the client that wants to capture the stream of events. +type EventStream struct { + Events <-chan *si.EventRecord +} + +// PublishEvent publishes an event to all event stream consumers. +// +// The streaming logic uses bridging to ensure proper ordering of existing and new events. +// Events are sent to the "local" channel from where it is forwarded to the "consumer" channel. +// +// If "local" is full, it means that the consumer side has not processed the events at an appropriate pace. +// Such a consumer is removed and the related channels are closed. +func (e *EventStreaming) PublishEvent(event *si.EventRecord) { + e.Lock() + defer e.Unlock() + + for consumer, details := range e.eventStreams { + if len(details.local) == defaultChannelBufSize { + log.Log(log.Events).Warn("Listener buffer full due to potentially slow consumer, removing it") + e.removeEventStream(consumer) + continue + } + + details.local <- event + } +} + +// CreateEventStream sets up event streaming for a consumer. The returned EventStream object +// contains a channel that can be used for reading. +// +// When a consumer is finished, it must call RemoveEventStream to free up resources. +// +// Consumers have an arbitrary name for logging purposes. The "count" parameter defines the number +// of maximum historical events from the ring buffer. +func (e *EventStreaming) CreateEventStream(name string, count uint64) *EventStream { + consumer := make(chan *si.EventRecord, defaultChannelBufSize) + stream := &EventStream{ + Events: consumer, + } + local := make(chan *si.EventRecord, defaultChannelBufSize) + stop := make(chan struct{}) + history := e.createEventStreamInternal(stream, local, consumer, stop, name, count) + + go func(consumer chan<- *si.EventRecord, local <-chan *si.EventRecord, stop <-chan struct{}) { + // Store the refs of historical events; it's possible that some events are added to the + // ring buffer and also to "local" channel. + // It is because we use two separate locks, so event updates are not atomic. + // Example: an event has been just added to the ring buffer (before createEventStreamInternal()), + // and execution is about to enter PublishEvent(); at this point we have an updated "eventStreams" + // map, so "local" will also contain the new event. + seen := make(map[*si.EventRecord]bool) + for _, event := range history { + consumer <- event + seen[event] = true + } + for { + select { + case <-e.stopCh: + close(consumer) + return + case <-stop: + close(consumer) + return + case event := <-local: + if seen[event] { + continue + } + // since events are processed in a single goroutine, doubling is no longer + // possible at this point + seen = make(map[*si.EventRecord]bool) + consumer <- event + } + } + }(consumer, local, stop) + + log.Log(log.Events).Info("Created event stream", zap.String("consumer name", name)) + return stream +} + +func (e *EventStreaming) createEventStreamInternal(stream *EventStream, + local chan *si.EventRecord, + consumer chan *si.EventRecord, + stop chan struct{}, + name string, + count uint64) []*si.EventRecord { + // stuff that needs locking + e.Lock() + defer e.Unlock() + + e.eventStreams[stream] = eventConsumerDetails{ + local: local, + consumer: consumer, + stopCh: stop, + name: name, + createdAt: time.Now(), + } + + return e.buffer.GetRecentEvents(count) +} + +// RemoveEventStream stops the streaming for a given consumer. Must be called to avoid resource leaks. +func (e *EventStreaming) RemoveEventStream(consumer *EventStream) { + e.Lock() + defer e.Unlock() + + e.removeEventStream(consumer) +} + +func (e *EventStreaming) removeEventStream(consumer *EventStream) { + if details, ok := e.eventStreams[consumer]; ok { + log.Log(log.Events).Info("Removing event stream consumer", zap.String("name", details.name), Review Comment: Nit Is it possible, we can logging the remove reason, such as: "Connection closed for event stream client" "Cannot set write deadline" "Marshalling error" -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
