bug: GitHub App preview deployments intermittently fail with internal errors; check conclusion can contradict actual deployment status

[Octember]

Summary

Since around 2026-06-09/10, GitHub App–initiated preview deployments for our Trigger Cloud project proj_pnreuadgcfkartotfwml (preview environment, per-PR branches) have been failing intermittently — roughly half or more of pushes — with generic internal errors. Successes are interleaved with failures on identical code, so this does not look like a problem with our build.

Failure shapes observed

1. Fast internal failure (check completes in ~6–60s, too fast for a real build):

Status: Failed (Build failed)
Error: Something went wrong while progressing the deployment, please try again shortly. If you are seeing this error consistently, please reach out to us.

Example deployments: h6gmbayy, uqli4nzf, dp1p9v0m, siu1o8vl, vlnrnr1a, qqxgvf1q

2. ~3-minute failure pointing at build logs:

Status: Failed (Deployment failed)
Error: Deployment failed, please check the build logs for more details.

Example deployments: sy4f3jsh, 4d29op07, 0l2labd2

3. Failure with no version / no deployment link at all (the GitHub check run carries no deployment URL, and the deployment doesn't appear in the project's deployment list).

4. Status mismatch — deployment succeeded but the GitHub check reports failure:
Deployment g4rk7h7y (v20260610.2) shows as DEPLOYED in the deployments API/dashboard, but the GitHub check run for it concluded failure with "Deployment failed, please check the build logs for more details." (completed 2026-06-10T19:16:18Z, deployment finished 2026-06-10T19:16:17Z). A CLI-initiated deploy of the same commit (9jee6c6n, v20260610.3) succeeded concurrently.

Additional failed deployment IDs from earlier on 2026-06-10: cft5o8bb, lv3lt50n, wkoybpqp.

Expected

GitHub App preview deployments either build/deploy reliably, or surface a real, actionable error — and the GitHub check conclusion should match the deployment's actual final status.

Context

• Trigger Cloud (cloud.trigger.dev), project proj_pnreuadgcfkartotfwml, preview environment with per-PR branch deploys via the GitHub App integration.
• Monorepo (bun workspaces), private repo. CLI-driven deploys of the same commits (via trigger.dev deploy from CI) succeed reliably; only the GitHub App–initiated builds are flaky.
• We're aware we can disable the GitHub App preview-deploy toggle (and likely will), but reporting since the error text says to reach out if seen consistently — and the success-marked-as-failure case (g4rk7h7y) looks like a genuine bug in the integration's check reporting.

Environment

Cloud product issue (no local repro). Deploys initiated by the Trigger.dev GitHub App from GitHub pushes, not from a local environment.

[JerePlum99]

Adding a data point — we're hitting what looks like the same regression, but through the Vercel integration (atomic deployments) rather than
the GitHub App directly. Since both ride the same git-deploy + check-run plumbing, it seems worth noting here.

Timeline: Started ~«2026-06-XX», lining up with the 06-09/10 window in this issue. Up until this week our Vercel preview deployments
posted a "trigger.dev build" deployment check and the Trigger.dev deployment linked back to the Vercel deployment. Newer preview deployments no
longer get the check, and the deployment shows as unlinked.

Setup:

• Trigger.dev Cloud + Vercel integration, atomic deployments enabled «for prod / preview?», GitHub connected.
• Pinned to trigger.dev@4.4.6 — though this appears to be Cloud-side and SDK-version-independent (nothing Vercel-related in the 4.4.x / 4.5.0-rc
  changelogs).

What works:

• The task build/deploy itself succeeds — CLI completes and registers every task, e.g. Version 20260615.2 deployed with 28 detected tasks on
  preview branch «dev-850-...».
• Vercel → Trigger env-var sync works on preview.

What's broken (matches failure shape # 3 above — "no deployment link at all"):

• No "trigger.dev build" check is posted on the Vercel deployment.
• The Trigger.dev deployment is not linked to the Vercel deployment. In the dashboard the "Linked" column shows "–", i.e.
  vercelDeploymentUrl is null — which per the OSS presenters means no IntegrationDeployment(integrationName='vercel') row was written for this
  worker deployment.

Already ruled out on our side:

• Reinstalled/reset the Vercel integration — no change.
• Removed the deprecated syncVercelEnvVars build extension — no change.
• Worker deploys fine via the CLI; it's specifically the check + link writeback that doesn't happen (consistent with the CLI-deploy-works
  workaround noted in this thread).

[wyatt-halluminate]

seeing same issue

[matt-aitken]

We cannot reproduce this issue and there has been no further information since July 2026. Closing for now. If you still see this on the latest version, please reopen the issue or open a new one with a minimal reproduction. Please include the deployment ID and the time of the failure.