fix(webhooks): always emit _bucket so cold-start lambdas don't strand the card - #41
Merged
Conversation
… the card
Partner webhooks on Vercel Fluid Compute sometimes land on a lambda whose
in-memory snapshot hasn't seen prior traffic. The old patch-diff optimization
in translatePartnerWebhook skipped _bucket when the server's local snapshot
already sat in the target bucket — but the client, hydrating from a
different instance, may have the offer in a different bucket. The status-
only patch then silently stranded the card (deriveActivityStatus short-
circuits on bucket='in_progress' before checking the RETRY status), and
admins had to fire attribution_failed twice before Needs Attention updated.
Fix: always include the canonical _bucket for attribution_{verified,failed}
and reward_paid. No-op when server and client agree; self-heals when they
don't. HAR + investigation notes captured alongside the regression test.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
The 5MB HAR added noise to the PR diff. The key request/response rows are already captured in the investigation markdown, which is the load-bearing artifact for future readers. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
ProgressRing and AnimatedNumber capture `start = performance.now()` and then drive cubic easing from the `now` argument handed back by `requestAnimationFrame`. Under jsdom + GH Actions jitter, the two clocks sometimes disagree enough that the first frame fires with `now < start` — negative `t`, `(1-t)^3` blows up, and the UI briefly renders `-2370%` / `-6554%` before catching up. Locally the offsets stay in sync and tests pass 8/8, but CI failed the ProgressRing suite on PR #41 (unrelated to the webhook fix it was scoped to). Clamping `t` to [0,1] makes a negative delta render `animated = 0` on that frame; the animation converges on its target from zero regardless of the RAF/performance-clock origin. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…stuck Previous attempt clamped `t` to [0,1], which stopped the wild negative renders but left the CI ring/number stuck at 0 — subsequent CI log showed stroke-dashoffset=circumference (animated=0) for the full 1s waitFor window. Root cause: jsdom's requestAnimationFrame timestamp and the synchronously- captured `performance.now()` come from different clock origins. With the clamp, negative deltas pin t to 0; the delta only turns positive when RAF's `now` finally crosses `performance.now()`'s origin, which on GH Actions runners can take longer than the waitFor timeout. Before the clamp, the negative delta just rendered garbage (-2370%, -6554%). Fix: ignore the RAF-provided timestamp entirely and read `performance.now()` inside the tick. Both anchor and delta now come from the same monotonic clock, so the animation progresses on the first frame regardless of the host's RAF timestamp semantics. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
attribution_failedrequired two clicks before the card moved into Needs Attention. Root cause:translatePartnerWebhookskipped_bucketin the emittedoffer_updatedpatch whenever the server's local snapshot already sat in the target bucket.deriveActivityStatusshort-circuits onbucket === 'in_progress'before it ever looks atstatus === 'RETRY'._bucketin the three status-changing partner events (attribution_verified,attribution_failed,reward_paid). No-op when server and client agree; self-heals when they don't.Evidence
Captured in
docs/investigations/2026-04-23-partner-webhook-double-fire.mdwith the full HAR alongside. From the admin-reported session:_bucket?pending(moved)Webhook #1 hit a lambda whose seed snapshot was already
pending, so the old code elided_bucket; the client (on a different lambda) was inin_progressand stayed there.Test plan
pnpm test src/server/webhooks/translate.test.ts— 14/14 pass, including a new end-to-endcross-instance drift regressionsuite that applies the emitted events to a drifted client view and asserts the offer lands inpendingwithRETRY.pnpm test— full suite 330/330 green.pnpm typecheck— clean.attribution_failedonce from the Partner Simulator on a preview deploy after a cold start and confirm the card moves on the first click.🤖 Generated with Claude Code
Note
Medium Risk
Changes partner webhook translation for status-changing events to always patch
_bucket, which affects how offer state is reconciled across realtime clients and could move cards between buckets more aggressively if assumptions are wrong.Overview
Fixes a production issue where partner webhook updates could require multiple fires to move an offer card, by making
translatePartnerWebhookalways include the canonical_bucketforattribution_verified,attribution_failed, andreward_paid(instead of emitting it only when the local snapshot differed).Expands
translate.test.tsto assert unconditional_bucketemission and adds an end-to-end regression test simulating cross-instance drift by applying translated events to a drifted client view.Separately hardens
AnimatedNumberandProgressRinganimations by samplingperformance.now()inside each RAF tick (and clampingt) to avoid jsdom clock-origin issues in tests, and adds an investigation write-up documenting the webhook drift root cause and fix.Reviewed by Cursor Bugbot for commit 5859c1a. Bugbot is set up for automated code reviews on this repo. Configure here.