Standardizing on Envoy: How Slack Tamed Service-to-Service Traffic
Why Slack adopted Envoy as a common proxy layer for internal service traffic, replacing a patchwork of per-service networking logic.
Engineering at scale
9 articles · 36 min of reading
Why Slack adopted Envoy as a common proxy layer for internal service traffic, replacing a patchwork of per-service networking logic.
How Slack built the Flannel edge cache and a websocket gateway tier to hydrate huge workspaces instantly without hammering primary databases.
Inside the public postmortem for Slack's January 2021 outage, where a networking degradation turned into a self-reinforcing overload across the stack.
How Slack moved its large PHP codebase onto Hack and HHVM, gaining static typing and JIT performance without a rewrite in a different language.
How Slack structures incident severity, on-call response, and blameless postmortems to turn outages into durable operational improvements.
How Slack rebuilt its Electron desktop app's architecture to support switching between many workspaces without the memory and performance problems piling up.
How Slack's original one-workspace-per-shard data model buckled under Enterprise Grid and giant workspaces, and why finer-grained sharding replaced it.
Why Slack adopted Vitess to shard MySQL horizontally at the connection-pooling and proxy layer instead of migrating off relational storage entirely.
Why Slack outgrew a Redis-backed job queue as its job volume and job diversity exploded, and what moving the queue to Kafka bought in return.