etcd 注解版

Note: 仅用于学习

etcd is a distributed reliable key-value store for the most critical data of a distributed system, with a focus on being:

• Simple: well-defined, user-facing API (gRPC)
• Secure: automatic TLS with optional client cert authentication
• Fast: benchmarked 10,000 writes/sec
• Reliable: properly distributed using Raft

etcd is written in Go and uses the Raft consensus algorithm to manage a highly-available replicated log.

参考链接

• etcd 官方文档
• raft dissertation
• The Raft Consensus Algorithm
• raft-zh_cn
• 深入浅出 etcd/raft
• etcd raft 处理流程图系列
• Raft 在 etcd 中的实现
• raft 笔记系列
• Etcd Raft 架构设计和源码剖析
• 分布式键值存储 etcd 原理与实现
• Implementing Raft: Part 0 - Introduction
• Diving into ETCD's linearizable reads
• Notes about ETCD
• Raft Consensus Protocol Made Simpler
• raft 算法中的集群成员变更问题
• 万级 K8s 集群背后 etcd 稳定性及性能优化实践
• Understanding performance aspects of etcd and Raft - Hitoshi Mitake, NTT Laboratories
• Understanding Distributed Consensus with Raft

Important issues and PRs

issues

• etcd-issues
• Data inconsistency in etcd version 3.5.0
• Small number of high latency requests after a leader election
• raft: liveness problems during apply-time configuration change
• Random performance issue on etcd 3.4
• Etcd cluster downgrade
• etcd move the boltdb freelist hashmap feature out of experimental
• Fully concurrent reads design proposal
• Task list: support raft learner in etcd
• Corrupted WAL and snapshot restoring process
• Leases may be auto-renewed indefinitely due to leader elections
• Prevent disruptive rejoining node
• Support non-voting members in etcd server
• raft: deadlock during PreVote migration process
• etcdserver: use PreVote so majority partition's leader isn't forced to step down
• raft: improve the availability related to member change
• ReadIndex may read stale value
• what is different about Revision, ModRevision and Version?

PRs

• server: set multiple concurrentReadTx instances share one txReadBuffer
• server: make applier use ReadTx() in Txn() instead of ConcurrentReadTx()
• raft: postpone MsgReadIndex until first commit in the term
• etcdserver: resend ReadIndex request on empty apply request
• raft: implement fast log rejection
• Lease Checkpoints fix
• mvcc: fully concurrent read
• add flags to setup backend related config
• lease: Persist remainingTTL to prevent indefinite auto-renewal of long lived leases
• mvcc: Clone for batch index compaction and shorten lock
• lease: Add a heap to optimize lease expiration checks
• configure Raft Pre-Vote to reduce disruptive rejoining servers
• raft: Avoid scanning raft log in becomeLeader
• implemented leader lease when quorum check is on
• rewrite snapshot merging and sending for v3
• Only send snapshot when member is online
• Implement etcd QoS POC