Downloaded 11 times
![[OSC2016] マイクロサービスを支える MQ を考える](https://cdn.slidesharecdn.com/ss_thumbnails/event-161105054759-thumbnail.jpg?width=640&height=640&fit=bounds)
Picking a message queue
- 1. Picking a MessageQueue. Vladislav Kirshtein Software engineer/consultant [email protected] 1
- 2.
- 3. About me Senior consultantat CodeValue since 2015 10 years of experience in large scale software development Particularly interested in highload projects, distributed systems, and cloud computing 3
- 4. Agenda Message oriented protocols AMQP- RabbitMQ ZeroMQ Cloud NATS 4
- 5. Protocols Stomp –Simple (or Streaming) Text Oriented Message Protocol MQTT – Message Queue Telemetry Transport AMQT – Advanced Message Queuing Protocol Vendor specific (Kafka,NATS,ZeroMQ etc)
- 6. STOMP – SimpleText Oriented Message Protocol • Simple interoperable protocol for asynchronous messaging • coming from HTTP school of design • Text Based • Communication between client and server is through a 'frame' • Client frames : SEND,SUBSCRIBE/ UNSUBSCRIBE, ACK/NACK… • Server frames: MESSAGE,RECEIPT,ERROR STOMP SERVER Client Client Send Subscribe Message Receipt ACK/NACK
- 7. MQTT – MessageQueue Telemetry Transport • Created by IBM for embedded devices telemetry • Ideal for constrained networks (low bandwidth, high latency, data limits, and fragile connections) • Binary format (2-byte header) • Pub/Sub • Reliable – QoS for reliability on unreliable networks • Security • Authentication • TLS/SSL • Simple – 43 page spec. • Connect/Disconnect + Publish + Subscribe/Unsubscribe 7
- 8. AMQP – AdvancedMessage Queue Protocol Conceived by JP Morgan in 2006 Answer to current Middleware Hell 100’s of applications 10,000’s of links every connection different massive waste of effort Goal to make interoperable Message Oriented Middleware 8
- 9. AMQP 0-10 vs1.0
- 10.
- 11.
- 12. Topic Exchange 12 Producer Exchange C Routingkey C C C C order.bissli
- 13. RabbitMQ Open source Support multipleprotocols AMQP,MQTT,STOMP,HTTP Routing capabilities Reliability Scalability through clustering High availability Management and monitoring Security Traceability and debugging Pluggable architecture 13
- 14. RabbitMQ 14 • Clustering • Actas single logical unit • Consumer seen cluster as a single node • Automatic metadata replication • Mirrored queues • Federation/Shovel • Consumes and republishers • Exchange to exchange • Works well across WAN • Two main difference • Shovel is low level • Shovel is more flexible
- 15. RabbitMQ 15 • Know yourrabbit (DevOps!) • Erlang • Configuration are Erlang tuples • You will need to deal with Erlang's message passing framework • Reading source and extend the system will be difficult • Optimized for discrete message handling • No content filtering • It's not too fast • Optimized for discrete message handling • Fan-out ~35000 mps • Direct ~35000 mps • Topic ~10000 mps • Consistent-Hash(plugin) ~40000 mps
- 16. RabbitMQ • Community support •https://groups.google.com/forum/#!forum/rabbitmq-users • Slack - https://rabbitmq-slack.herokuapp.com/ • Official site - http://www.rabbitmq.com/ • Books • RabbitMQ in Action • RabbitMQ in Depth • RabbitMQ Essentials • RabbitMQ best practices – CloudAMQP • https://www.cloudamqp.com/blog/2017-12-29-part1-rabbitmq-best-practice.html • Video 16
- 17. RabbitMQ - Demo 17 ClientServer Request queue Response queue https://github.com/vgrokk/RabbitMQDemo
- 18. RabbitMQ - Demo 18 ProducerExchange C C C C C order.bissli https://github.com/vgrokk/RabbitMQDemo
- 19. No more brokers 19 •Brokers are single point of failure • Not horizontally scalable • Reduce reliability with addition of nodes • HA provides minimal benefit, and adds complexity
- 20.
- 21. ZeroMQ What Ø comefor? The original Ø meant: • Zero broker • Zero latency (as close as possible) With the time • Zero administration • Zero cost • Zero waste 21
- 22.
- 23. ZeroMQ Creators - iMatixCEO Pieter Hintjens and Martin Sustrik Super socket library Language agnostic (40+ languages) Multiple network protocols Multiple connection patterns High throughput/low latency (~5m mps, 30usec latency) Messages Batching Atomic Small (~25k lines of code) Open source. Large supported community. Multiple forks (nanomessage) 23
- 24. ZeroMQ - Cons 24 Durability Staticrouting No out of the box solution Service discovery Management Delivery deadlines QoS Low level Internal queue management
- 25. ZeroMQ - Sockets •Unicast – inproc, ipc (except on Windows), tcp • Multicast – pgm, epgm • Asynchronous • Designed for particular messaging patterns • Connect to multiple endpoints • Automatically reconnect • Queue messages when under heavy load 25
- 26. ZeroMQ - Sockets 26 •PUB - Outgoing broadcast • SUB - Broadcast listener • REQ - Synchronous (1 msg at a time) request • REP - Synchronous reply • DEALER - Async request • ROUTER - Async reply • PUSH - One way outgoing to PULL(s) • PULL - One way incoming from PUSH(es) • PAIR - One-to-one 2-way with another PAIR PUB and SUB REQ and REP REQ and ROUTER DEALER and REP DEALER and ROUTER DEALER and DEALER ROUTER and ROUTER PUSH and PULL PAIR and PAIR
- 27.
- 28.
- 29.
- 30.
- 31. Cloud You host yourservices on cloud You don’t have a dedicated budget or skills to maintain your own queueing system You aren’t necessarily interested in lowest latency You prefer easy solutions over powerful solutions when queuing is considered You don’t plan on processing large amounts of messages 31
- 32. Self - hosted 32 Yoursystem is distributed or hosted outside of public cloud You want an open solution that you can reuse You’re worried about latency You want to configure exchanges or use some other non-trivial methods of message distribution You want your queuing system to be extendable and configurable, and You want maximum performance and have dedicated resources to invest in your messaging solution.
- 33.
- 34. Storage queue PROS • Canstore over 80Gb in queue • Low cost (100x cheaper than Azure Service Bus) • Batch receive (up to 32 messages) • Only pay for storage and operations • Redundancy(LRS, ZRS, GRS, RA-GRS) • Rest api CONS • Maximum size of a message is 64KB • Maximum mps is 2000 • Maximum TTL for each message is 7 days • No order guaranty • Only polling (for receiver) • No batch support for sender 34
- 35.
- 36. Service Bus 36 Scheduled delivery QoS Poisonmessage handling Defer Auto Forward Sessions Batching Ordering Auto-delete on idle OnMessage (.NET only) Duplicate detection Content filters Transactions Dead lettering
- 37. Amazon 37 AWS SQS Amazon SNS Amazon MQ Amazon SQS Amazon Kinesis AWSIOT Managed queue Managed pub/sub Managed Apache ActiveMQ Managed event streaming
- 38. Amazon SQS • Unlimitedqueues and messages • Payload Size: up to 256KB • Batches • Long polling • Retain messages in queues for up to 14 days 38 • Send and read messages simultaneously • Message locking • Queue sharing • Server-side encryption (SSE) • Dead Letter Queues (DLQ) • Amazon CloudWatch metrics + alerts
- 39. Amazon SNS 39 • Fan-outpattern • Reliability • Flexible • Nearly unlimited throughput • Secure • Payloads up to 256 KB • Amazon CloudWatch
- 40. Amazon MQ 40 Amazon MQis a managed message broker service for Apache ActiveMQ that makes it easy to set up and operate message brokers in the cloud.
- 41. Amazon MQ 41 ActiveMQ Features •Transient & Persistent • Queues & topics (FIFO) • Consistency • Local & Distributed transactions • Message filtering • Ordering • Scheduled messages • Unlimited message size • Unlimited retention AmazonMQ Features • Provisioning • Updates • Security • Maintenance • Monitoring • Troubleshooting • High availability
- 42. 42 • Was createdby Derek Colission in 2010 • Open source, lightweight, high-performance cloud native messaging system • Capable of sending 11-12 million messages per second • Written in Go • Available in two interoperable modules • NATS Server • NATS Streaming Server
- 43.
- 44.
- 45. NATS Server Highly performant,extremely lightweight Clustered servers/clustered aware clients Built in resilience and high availability Auto discovery for topology Text based protocol (payload as array of bytes) Monitorable on a dedicated port TLS support Auto Pruning of Clients Pure Pub/Sub Request/Reply ,Queueing pattern Smart routing with wildcards 45 Durability Transactions QoS Flow control (rate matching/limitng)
- 46. NATS Streaming Layer ontop of NATS Enhanced message protocol - NATS Streaming implements its own enhanced message format using Google Protocol Buffers. Message/event persistence QoS Flow control (rate matching/limiting) Message replay by subject All available messages Last published value All messages since specific sequence number All messages since a specific date/time Durable subscribers 46
- 47.
- 48.
- 49. NATS cluster 49 • Forhigh availability, a full mash of NATS servers can be setup • Clients can connect to any of nodes to communicate with other clients, the NATS cluster would then route the messages • Routing will be done if clients showed interests in subject • Whenever a new NATS servers joins then members already in cluster connected as well to form the full mash
- 50.
- 51.