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

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
docs	Skipped		May 21, 2026 10:10pm

PR Summary

Medium Risk
Mostly performance-focused query/invalidations changes, but it touches multiple API routes and adds DB indexes/migration; mistakes could cause stale UI data or locking/plan regressions in production.

Overview
Reduces database and cache churn across several hot paths. Many API handlers now project only needed columns (rather than select()), narrow .returning() payloads, and skip unnecessary writes (e.g., no-op webhook PATCH; webhook success update only when failedCount is non-zero).

Cuts client polling/invalidation frequency. Logs list/dashboard live refresh moves from 3s to 10s (detail stays 3s only for running/pending), table SSE dispatch invalidations are debounced, subscription invalidations are narrowed to subscriptionKeys.users() and useSubscriptionData staleTime is increased to 5 minutes; useDeploymentInfo stops forcing refetchOnMount: 'always'.

Tightens and batches server-side writes. MCP lifecycle updates collapse select-then-update into single guarded updates (including server/tool scoping), workspace env credential membership sync is batched via ON CONFLICT upserts/insert-ignore, and migration 0211_breezy_cloak adds indexes on chat, webhook, and workflow_schedule to support the updated access patterns.

^{Reviewed by Cursor Bugbot for commit 3cf4f7d. Configure here.}

Greptile Summary

This PR applies a broad performance sweep across hot paths: narrowing query projections (JSONB columns and SELECT *), batching credential membership writes, reducing React Query cache churn, and adding four PostgreSQL indexes for common query patterns. It also closes a latent workspace-guard gap in MCP server UPDATE by folding the check into the single UPDATE…RETURNING statement.

• DB read reduction: 8+ SELECT * callsites narrowed to projected columns; 3 SELECT-then-UPDATE/DELETE patterns in MCP collapsed to single statements; markWebhookSuccess and webhook PATCH skip writes when no-op; credential membership writes batched with ON CONFLICT DO UPDATE/DO NOTHING
• Client-side throttling: React Query staleTime bumped from 30 s → 5 min for subscription data; invalidation narrowed from subscriptionKeys.all to subscriptionKeys.users(); logs-page polling 3 s → 10 s; SSE dispatch-invalidate debounced 250 ms
• Migration 0211_breezy_cloak: adds four indexes on chat, webhook (×2), and workflow_schedule; migration is safe to run against the current table sizes noted in the PR description

Confidence Score: 5/5

Safe to merge — all hot-path optimisations are logically equivalent to the code they replace, workspace guards are preserved or tightened, and the migration targets tables small enough that lock windows are negligible.

The batched credential writes and ON CONFLICT logic match the old per-row loop behaviour. The MCP UPDATE/DELETE collapses include the workspace predicate directly in the WHERE clause and, for the tool update, actually close an old gap where the bare UPDATE had no serverId check. The React Query staleTime bump is paired with explicit invalidation on every mutating operation. No data loss or correctness regression identified.

No files require special attention beyond the minor JSDoc discrepancy in subscription.ts.

Important Files Changed

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Workflow Run / Wand / UsageIndicator socket] -->|invalidateQueries| B[subscriptionKeys.users]
    B --> C[useSubscriptionData\nstaleTime=5min]
    C --> D[GET /api/billing user]

    E[SSE dispatch event] -->|scheduleDispatchInvalidate\n250ms debounce| F[tableKeys.activeDispatches]
    F --> G[refetch activeDispatches]

    H[Logs page live mode] -->|refetchInterval 10s| I[logsQuery + dashboardStatsQuery]

    J[PATCH /api/webhooks/:id] --> K{setClause empty?}
    K -->|yes| L[return 200 no-op]
    K -->|no| M[UPDATE webhook]

    N[markWebhookSuccess] --> O{failedCount already 0?}
    O -->|yes, skip| P[no UPDATE]
    O -->|no| Q[UPDATE failedCount=0]

    R[ensureWorkspaceCredentialMemberships] --> S[SELECT existing revoked]
    S --> T[filter out revoked users]
    T --> U[batch INSERT ON CONFLICT DO UPDATE\npreserve joinedAt via COALESCE]

    V[syncWorkspaceEnvCredentials /\ncreateWorkspaceEnvCredentials] --> W[batch INSERT credentials\nON CONFLICT DO NOTHING]
    W --> X[batch INSERT memberships\nON CONFLICT DO NOTHING]

_{Reviews (3): Last reviewed commit: "fix(webhooks): include NULL failedCount ..." | Re-trigger Greptile}

Thanks for the review. Addressed in 4baad33:

• MCP delete error ambiguity (workflow-mcp-lifecycle.ts:518-529) — restored the server SELECT before the DELETE so a workspace mismatch returns 'Server not found' and a genuinely missing tool returns 'Tool not found'. Matches the pattern used in performUpdateWorkflowMcpTool. The DELETE itself is unchanged (still scoped by id + serverId); the SELECT is a cheap PK lookup so the extra round-trip cost is negligible.
• chatDeploy projection (manage.ts:45-50) — narrowed to the explicit 9-column projection (id, identifier, title, description, authType, allowedEmails, outputConfigs, password, customizations) matching what's consumed downstream. Aligns with the JSONB sweep theme.

@greptile

@cursor review

@greptile

@cursor review