CortexDB 0.12.8 — what went wrong for each tenant, and what each one stores
0.12.8 is 0.12.7 plus outcome counters per tenant and an opt-in storage figure in the operator's usage listing, for the hosted admin. Upgrading from 0.12.7 is a drop-in replacement: the new counters are new fields in the existing usage rows (old rows read back as zeros), and nothing about requests or responses changes.
New
- Outcomes per tenant. The IR12 activity counters (
GET /v1/usageactivity andGET /v1/admin/usage/subtrees) now count how each request charged to a tenant ended:errors_5xx,errors_4xx(the tenant's own bad requests, never 401 or 403),refused_plan(402PLAN_LIMIT_REACHED) andrefused_rate(429SUBTREE_RATE_LIMITED), each with a_monthtwin, pluslast_error_atandlast_error_code. A request reaches its tenant at its plan check or its subtree rate limit (the memory routes); one refused before that counts for none. Each field is left out while 0. - Storage per subtree on request.
GET /v1/admin/usage/subtrees?include_storage=1addsstorage_events,storage_bytesandstorage_cappedto each subtree (thestoragefigure of/v1/usage). It walks each subtree's rows (at most 1,000,000 each) on the DB lane, so such a page holds at most 100 subtrees.