feat(health): expose failure-log via /api/health?history=1 by koala73 · Pull Request #3641 · koala73/worldmonitor

koala73 · 2026-05-10T04:20:19Z

Summary

Surfaces the health:last-failure + health:failure-log Redis keys via a query-param path on /api/health, so diagnosing UptimeRobot flips no longer requires Upstash credentials.

Why

The /api/health classifier already writes incident snapshots to Redis on every non-OK probe:

health:last-failure — most recent unhealthy snapshot (24h TTL)
health:failure-log — last 50 deduped incidents (7-day TTL)

But there was no read endpoint. WM 2026-05-10 had a night of UptimeRobot DOWN/UP flips on correlationCards; the only way to diagnose was hand-pulling health:failure-log via curl + Upstash token. This PR adds the missing read surface.

Behavior

GET /api/health?history=1
{
  "lastFailure": { "at": "...", "status": "WARNING", "critCount": 0, "warnCount": 1, "problems": ["correlationCards:STALE_SEED(19min)"] },
  "failureLog": [
    { "at": "2026-05-10T04:04:29Z", "status": "WARNING", "problems": ["correlationCards:STALE_SEED(19min)"] },
    { "at": "2026-05-10T02:21:49Z", "status": "WARNING", "problems": ["correlationCards:STALE_SEED(17min)"] },
    ...
  ],
  "checkedAt": "2026-05-10T04:30:00Z"
}

GET /api/health             ← unchanged (full classification probe)
GET /api/health?compact=1   ← unchanged (compact problem-list shape)

The new path is early-return: it reads two Redis keys and returns. Skips the full bootstrap-keys probe (which is expensive). Cheap to call from a browser tab.

Discriminator

Strict === '1' match on the history query param. Any other value (0, true, yes, 01) falls through to the classification path. Tests cover all three cases.

Graceful degradation

When Upstash is unreachable (creds missing, network failure, timeout), redisPipeline returns null and the handler responds with { lastFailure: null, failureLog: [] }. Never throws, never 5xx.

Test plan

node --test tests/health-history-endpoint.test.mjs — 3/3 pass
- Returns history shape on ?history=1, no exceptions when Redis unconfigured
- Default /api/health?compact=1 does NOT include history fields (no leak into the classification response)
- Non-1 values for ?history (0, true, yes, 01) all fall through to the classification path
node --test tests/health-content-age.test.mjs — 18/18 pass (existing classifier tests unchanged)
node --check api/health.js — syntax OK
npx esbuild --bundle --format=esm --platform=browser — bundles cleanly for the edge runtime
Post-merge: curl https://api.worldmonitor.app/api/health?history=1 returns the live failure log

Companion

Pairs with #3640 (correlationCards budget bump). Together they (a) reduce false-positive flips and (b) make residual flips easier to diagnose.

Adds an early-return query-param path that surfaces the existing `health:last-failure` + `health:failure-log` Redis state without running the full freshness probe. Background: every non-OK /api/health response writes a snapshot `{at, status, critCount, warnCount, problems}` to Redis: health:last-failure — most recent unhealthy snapshot (24h TTL) health:failure-log — last 50 deduped incidents (7-day TTL) Pre-fix this data was operator-only — diagnosing UptimeRobot flips required Upstash credentials. WM 2026-05-10 hit this directly: a night of correlationCards STALE_SEED flaps could only be diagnosed by hand-pulling the failure-log via curl + token. The history endpoint now answers "what's been bringing health down?" from a browser. Behavior: GET /api/health?history=1 → { lastFailure, failureLog, checkedAt } Reads only 2 Redis keys; skips the full bootstrap-keys probe. GET /api/health (default) → unchanged: full classification probe + summary + problems. The history path uses a strict `=== '1'` match on the query param. Any other value (including 0, true, yes) falls through to the classification path. Tests cover all three cases. Graceful degradation: when Upstash is unreachable, redisPipeline returns null and the handler returns `{ lastFailure: null, failureLog: [] }` — never throws, never 5xx. 3/3 new tests pass; 18/18 existing health tests unchanged.

vercel · 2026-05-10T04:20:24Z

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
worldmonitor	Ready	Preview, Comment	May 10, 2026 4:37am

greptile-apps · 2026-05-10T04:24:46Z

Greptile Summary

This PR adds an early-return path on GET /api/health?history=1 that reads two Redis keys (health:last-failure, health:failure-log) and returns the incident history without executing the full bootstrap-key freshness probe. The motivation is diagnosing UptimeRobot flips without needing direct Upstash credentials.

New handler branch (api/health.js): Strict === '1' match on ?history exits before the expensive STRLEN/GET pipeline; gracefully degrades to { lastFailure: null, failureLog: [] } when Redis is unreachable.
New test file (tests/health-history-endpoint.test.mjs): Three tests verify the history shape, graceful degradation on no-Redis, and that non-"1" values don't trigger the early-return.

Confidence Score: 4/5

Safe to merge — the new branch is well-isolated, never throws, and reuses the existing response headers without touching the full probe path.

The early-return logic is correct end-to-end: optional chaining handles a null pipeline result, parseJson catches malformed JSON, and the LRANGE range covers the full bounded list. The only findings are cosmetic redundancies that carry no runtime risk.

No files require special attention; both changed files are straightforward additions with no interaction with the existing classification path.

Important Files Changed

Filename	Overview
api/health.js	Adds a 38-line early-return block for `?history=1`; logic is correct with proper optional chaining, graceful null fallback, and reuse of existing `headers`; minor: the outer `typeof lastFailureRaw === 'string'` guard is redundant with `parseJson`'s own type check, and `new Date(Date.now())` can be written as `new Date()`.
tests/health-history-endpoint.test.mjs	Three focused tests covering the happy path (no-Redis degradation), the non-trigger case, and the discriminator logic; uses the ESM cache-busting query-param trick (`?second-import`, `?third-import`) which is valid but non-obvious.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[GET /api/health] --> B{OPTIONS request?}
    B -- Yes --> C[204 No Content]
    B -- No --> D{history param equals 1?}
    D -- Yes --> E[redisPipeline: GET health:last-failure\nLRANGE health:failure-log 0 -1\ntimeout 4s]
    E --> F{results null?}
    F -- Yes --> G[lastFailure null\nfailureLog empty array]
    F -- No --> H[parseJson on each result\nfilter nulls from log array]
    H --> I[lastFailure + failureLog + checkedAt]
    G --> J[200 JSON early return]
    I --> J
    D -- No --> K[Full freshness probe\nSTRLEN all bootstrap keys\nGET all seed-meta keys\ntimeout 8s]
    K --> L{Redis reachable?}
    L -- No --> M[REDIS_DOWN 200 response]
    L -- Yes --> N[Classify all keys\ncompute overall status]
    N --> O{overall not HEALTHY?}
    O -- Yes --> P[Write health:last-failure\nAppend to health:failure-log in Redis]
    P --> Q[Return classification JSON]
    O -- No --> Q

_{Reviews (1): Last reviewed commit: "feat(health): expose failure-log via /ap..." | Re-trigger Greptile}

greptile-apps · 2026-05-10T04:24:50Z

+      const lastFailureRaw = results?.[0]?.result;
+      const failureLogRaw = results?.[1]?.result;
+      const body = {
+        lastFailure: typeof lastFailureRaw === 'string' ? parseJson(lastFailureRaw) : null,


The outer typeof lastFailureRaw === 'string' guard is redundant — parseJson already returns null for any non-string input via its own typeof raw !== 'string' check. Both branches produce identical results, so the ternary can be collapsed to a direct parseJson call.

Suggested change

const lastFailureRaw = results?.[0]?.result;

const failureLogRaw = results?.[1]?.result;

const body = {

lastFailure: typeof lastFailureRaw === 'string' ? parseJson(lastFailureRaw) : null,

const lastFailureRaw = results?.[0]?.result;

const failureLogRaw = results?.[1]?.result;

const body = {

lastFailure: parseJson(lastFailureRaw),

greptile-apps · 2026-05-10T04:24:51Z

+        failureLog: Array.isArray(failureLogRaw)
+          ? failureLogRaw.map(parseJson).filter((e) => e !== null)
+          : [],
+        checkedAt: new Date(Date.now()).toISOString(),


new Date(Date.now()) constructs a Date from the current timestamp, which is identical to new Date() with no arguments. The Date.now() call is redundant here.

Suggested change

checkedAt: new Date(Date.now()).toISOString(),

checkedAt: new Date().toISOString()

Greptile PR #3641 review: P2 — `typeof lastFailureRaw === 'string'` ternary is redundant because the parseJson helper already returns null for any non-string input via its own `typeof raw !== 'string'` check. Both branches produced identical results — collapsed to a direct parseJson call. P2 — `new Date(Date.now())` is identical to `new Date()` with no args. Dropped the redundant Date.now() wrapper. Tests still 3/3.

koala73 · 2026-05-10T04:36:00Z

Both P2 nits addressed in a17295f7a:

P2 — redundant typeof === 'string' guard: collapsed to a direct parseJson(lastFailureRaw) call. The helper already returns null for any non-string input.

P2 — redundant Date.now() wrapper: changed new Date(Date.now()) → new Date().

3/3 tests still pass. Ready for merge.

vercel Bot deployed to Preview May 10, 2026 04:21 View deployment

greptile-apps Bot reviewed May 10, 2026

View reviewed changes

vercel Bot deployed to Preview May 10, 2026 04:37 View deployment

koala73 merged commit 047bc3c into main May 10, 2026
11 checks passed

koala73 deleted the feat/health-history-endpoint branch May 10, 2026 04:44

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(health): expose failure-log via /api/health?history=1#3641

feat(health): expose failure-log via /api/health?history=1#3641
koala73 merged 2 commits into
mainfrom
feat/health-history-endpoint

koala73 commented May 10, 2026

Uh oh!

vercel Bot commented May 10, 2026 •

edited

Loading

Uh oh!

greptile-apps Bot commented May 10, 2026

Uh oh!

greptile-apps Bot May 10, 2026

Uh oh!

greptile-apps Bot May 10, 2026

Uh oh!

koala73 commented May 10, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

	checkedAt: new Date(Date.now()).toISOString(),
	checkedAt: new Date().toISOString()

Conversation

koala73 commented May 10, 2026

Summary

Why

Behavior

Discriminator

Graceful degradation

Test plan

Companion

Uh oh!

vercel Bot commented May 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

greptile-apps Bot commented May 10, 2026

Greptile Summary

Confidence Score: 4/5

Important Files Changed

Flowchart

Uh oh!

greptile-apps Bot May 10, 2026

Choose a reason for hiding this comment

Uh oh!

greptile-apps Bot May 10, 2026

Choose a reason for hiding this comment

Uh oh!

koala73 commented May 10, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

vercel Bot commented May 10, 2026 •

edited

Loading