Chaos day: four failure drills

Guard 19
RegionZoneZoneHTTP/2HTTP/2SQLSQLUsers1,200 req/sLoad balanc…1 vCPU · 1 GBNode API1 vCPU · 1 GBNode API1 vCPU · 1 GBPostgres1 vCPU · 2 GB
RegionZoneZoneSQLSQLUsersLoad ba…Node APINode APIPostgres

Keep every goal green in each wave.

Traffic
Browse1,080 req/s · reads onlyOrders120 req/s · writes only
Limit
Node API answers about 1,000 req/s
Bill
  • Load balancer (L7)4 USD/mo
  • Node API × 28 USD/mo
  • Postgres8 USD/mo
20 of 24 USD/mo

Goals

Not tested yet

The share of time when every target was met.99 of 100 requests finish faster than this.The share of requests that fail or get no answer.The share of requests that fail or get no answer.What the servers and services cost each month.

Four 50 s waves each inject one failure, in order: an API instance crashes, the database runs 12 times slower, zone A fails, an API instance runs 6 times slower. Keep browse p99 under 300 ms and errors under 2 percent for 95 percent of the run. Each wave adds budget.

Parts you can add: Node API, Load balancer (L7), Redis cache. At most 10 parts. Budget: 24 USD/mo.

Tap a part to add it, or drag it onto a trace.Click a part to add it, or drag it onto a trace.