Operate Etcd cluster

⇢ 

NOTE: This is my perspective aggregation. You can easily find these such of knowledges in the references.

Context

Etcd Version v3.4.0.

Requirements

Number of nodes

  • >= 3 nodes. A etcd cluster needs a majority of nodes, a quorum to agree on updates to the cluster state. For a cluster with n-members, quorum is (n/2)+1.

CPUs

  • Etcd doesn’t require a lot of CPU capacity.
  • Typical clusters need 2-4 cores to run smoothly.

Memory

  • Etcd performance depends on having enough memory (cache key-value data, tracking watchers…).
  • Typical 8GB is enough.

Disk

  • An etcd cluster is very sensitive to disk latencies. Since etcd must persist proposals to its log, disk activity from other processes may cause long fsync latencies. The upshot is etcd may miss heartbeats, causing request timeouts and temporary leader loss. An etcd server can sometimes stably run alongside these processes when given a high disk priority.
  • Check whether a disk is fast enough for etcd using fio. If the 99th percentile of fdatasync is <10ms, your storage is ok.
$ fio --rw=write --ioengine=sync --fdatasync=1 --directory=test-data \
    --size=22m --bs=2300 --name=mytest
  • SSD is recommended.

Network

  • Etcd cluster should be deployed in a fast and reliable network. Low latency ensures etcd members can communicate fast. High bandwidth can reduce the time to recover a failed etcd member.
  • 1GbE is sufficient for common etcd.
  • Note that the network isn’t the only source of latency. Each request and response may be impacted by slow disks on both the leader and followers.

Tuning

Time parameters

  • Heartbeat interval.
    • The frequency with which the leader will notify followers that it is still the leader.
    • Default: 100ms.
    • Best practice: Around 0.5-1.5 x round-trip time (RTT) between members. Measure RTT with ping.
    • Tradeoff: Too low -> etcd will send unnecessary messages -> increase the usage of CPU and network resources. Too high -> leads to high election timeout.
  • Election timeout.
    • How long a follower node will go without hearing a heartbeat before attempting to become leader itself.
    • Default: 1000ms.
    • Best practice: >= 10 x RTT and < 50s.
  • The heartbeat interval and election timeout value should be the same for all members in one cluster.
# Command line arguments:
$ etcd --heartbeat-interval=100 --election-timeout=500

# Environment variables:
$ ETCD_HEARTBEAT_INTERVAL=100 ETCD_ELECTION_TIMEOUT=500 etcd

Disk

  • An etcd server can sometimes stably run alongside these processes when given a high disk priority using ionice.
# best effort, highest priority
$ sudo ionice -c2 -n0 -p `pgrep etcd`

Snapshot

  • etcd appends all key changes to a log file -> huge log that grows forever ☝️
  • Solution: Make periodic snapshots (save the current and remove old logs).
  • Default: make snapshots after every 10 000 changes.
  • Tuning: Just in case that etcd’s memory and disk usage is too high, lower threshold.
# Command line arguments:
$ etcd --snapshot-count=5000

# Environment variables:
$ ETCD_SNAPSHOT_COUNT=5000 etcd

Maintenance

History compaction

  • Etcd keeps an exact history of its keyspace, the history should be periodically compacted to avoid performance degradation and eventual storage space exhaustion.
  • Etcd can be set to automatically compact the keyspace with the --auto-compaction-* option with a period of hours.
# keep one hour of history
$ etcd --auto-compaction-retention=1 --auto-compaction-mode=periodic
  • Compaction modes:
    • Revision-based: --auto-compaction-mode=revision --auto-compaction-retention=1000 automatically Compact on “latest revision” - 1000 every 5-minute (when latest revision is 30000, compact on revision 29000). Use this when having a large keyspace.
    • Periodic: --auto-compaction-mode=periodic --auto-compaction-retention=72h automatically Compact with 72-hour retention window every 1-hour. Use this when having a huge number of revisions for a key-value pair.

Defragmentation

  • Compacting old revisions internally fragments etcd by leaving gaps in backend database - internal fragmentation.
  • Internal fragmentation space is available for use by etcd but unavailable to the host filesystem.
  • Solution: Release this space back to the filesystem with defrag.
$ etcdctl defrag
  • It should be run rather infrequently, as there is always going to be an unavoidable pause.

References