Serverless: Connections Do Not Pool

Serverless: Connections Do Not Pool A workflow diagram generated by Archify. 01 / Request and managed compute 02 / Data tier EX / Connecting directly Arrival Contention Resolution Traffic Spike · the launch went well · Request and managed compute › Arrival Traffic Spike the launch went well Functions · hundreds at once · Request and managed compute › Arrival Functions hundreds at once Relational DB · sized for a few apps · Data tier › Contention Relational DB sized for a few apps Data Proxy · or an HTTP data API · Data tier › Resolution Data Proxy or an HTTP data API Multiplexed · a few connections · Data tier › Resolution · survives Multiplexed a few connections survives Exhausted · connection limit hit · Connecting directly › Resolution Exhausted connection limit hit one connection each through a proxy connecting directly Legend Policy Context / trace Cloud service External system

Why it bites in month three

  • • This is the most common serverless production incident
  • • It appears exactly when a launch goes well
  • • Execution limits are architectural constraints, not configuration details

What you actually bought

  • • No patching, no capacity planning, no node pool
  • • Scale to zero, and up
  • • A default security posture you did not have to build

What you traded it for

  • • IAM becomes the architecture, and misconfiguration is the dominant risk
  • • Cost becomes a per-request property of your code
  • • Lock-in is at the seams — the composition is the application