Skip to content

EN Observability

Cavanshirpro edited this page Sep 19, 2026 · 1 revision

Health, readiness, metrics, audit and errors

English · Türkçe · Languages / Diller

Scope: JSON API Forge 0.5.1 · Source review: 17 September 2026 · Language: English

Endpoints

Endpoint Meaning
/health Cheap process liveness
/ready Readiness based on dependencies/projects
Configured metrics path Prometheus output when enabled
<prefix>/meta Project metadata subject to authorization
<prefix>/_openapi.json The project's actual endpoint contract

/health=200 does not establish that every application is healthy. No active apps, an invalid app or dependency failures can produce /ready=503. Detailed readiness can identify application names and error types, without exposing secrets.

OPERATOR_TOKEN is a separate credential for detailed readiness/metrics. Send X-Forge-Operator-Token or the appropriate Bearer header. An application key does not grant process-level operator access, and an absent token does not authorize the caller.

HTTP errors

Status Typical cause First check
400 Invalid JSON/parameter or missing idempotency header Actual endpoint contract
401 Missing/invalid credential or consumed bootstrap Credential type and lifetime
403 Permission, host/IP/HTTPS or scope rejection Principal and server policy
404 Missing/invisible route or row Prefix, action, tenant/owner
409 Revision, rowcount, idempotency or business conflict Previous state and key
413 Body/file/result limit Both proxy and Forge limits
422 Schema/type/field violation JSON schema and field allowlists
429 Rate budget exceeded Retry-After and client-IP resolution
503 Saturation, dependency or readiness failure Readiness, DB, Redis, concurrency

These are typical categories, not a guarantee that each endpoint returns every listed status. Error bodies also depend on the endpoint/model.

Request tracing

Request IDs correlate client and server diagnostics. SDK ForgeResponse exposes request_id, cache_status and idempotent_replay. The SDK preserves a request ID through one retry chain; it is not the business idempotency key.

Monitor request counts, latency, cache behavior and audit queue/drop/write-failure signals together. Averages can hide queueing and tail latency. Avoid logging tokens, complete bodies and personal data by default.

Audit boundaries

Application audit and Editor security audit serve different purposes. Alert on dropped audit entries and write failures. Application-level append-only design is not cryptographic immutability against a database administrator or a regulatory certification. Protect operational access and backups separately.

Sources and verification

This page describes the behavior at the pinned commits below. Recheck these contracts when upgrading.


English home · All English pages · Version maintenance

Clone this wiki locally