log(db): Add db failure cause log message

[TheodoreSpeaks]

Summary

• Log the underlying Postgres/driver error cause (code, severity, detail, routine, errno, syscall) when workflow execution fails
• Drizzle wraps the real driver error as error.cause, which the logger's Error serializer dropped — leaving only an opaque Failed query: ... message with no pg error code
• Walk the cause chain (depth-bounded to 10) and surface the first error carrying a code; this distinguishes a connection drop (08006), rejected connection (53300), and statement timeout (57014)
• Wrapped in try/catch so the diagnostic can never throw or hang inside the error handler — worst case is no cause field, never a masked failure

Type of Change

• Improvement (logging/observability)

Testing

Tested manually; bun run lint clean, bun run check:api-validation:strict passed, tsc --noEmit clean. Motivated by intermittent schedule-execution failures where the workspace/personal env reads in preprocessing fail with an opaque "Failed query" and no driver error code.

Checklist

• Code follows project style guidelines
• Self-reviewed my changes
• Tests added/updated and passing
• No new warnings introduced
• I confirm that I have read and agree to the terms outlined in the Contributor License Agreement (CLA)

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
docs	Ready	Preview, Comment	May 26, 2026 11:43pm

[cursor]

PR Summary

Medium Risk
Migration CI path and statement_timeout changes affect how schema deploys run; execution logging is observability-only with a guarded helper.

Overview
Improves workflow execution failure logs by walking Drizzle’s wrapped error.cause chain and attaching driver/Postgres fields (code, severity, detail, etc.) so intermittent DB errors are easier to classify than a bare “Failed query” message.

CI database migrations now run via bun run ./scripts/migrate.ts instead of drizzle-kit migrate, and that script SET statement_timeout = 0 before applying migrations so long-running DDL is not cut off by a server timeout.

^{Reviewed by Cursor Bugbot for commit 63e3927. Bugbot is set up for automated code reviews on this repo. Configure here.}

[greptile-apps]

Greptile Summary

This PR improves observability for workflow execution failures by surfacing the underlying Postgres/driver error that Drizzle wraps in error.cause, and switches the CI migration runner to a custom script with richer diagnostics and a session-level SET statement_timeout = 0 guard.

• describeErrorCause walks the cause chain (depth-bounded to 10, wrapped in try/catch) and extracts the first error carrying a code, then passes it as a structured cause field to logger.error — enabling CloudWatch Log Insights to distinguish connection drops, pool exhaustion, and statement timeouts behind the previously opaque "Failed query" message.
• packages/db/scripts/migrate.ts is now the CI migration runner via .github/workflows/migrations.yml; it disables statement_timeout for the session and prints comprehensive Postgres diagnostic fields on failure via printMigrationError.

Confidence Score: 4/5

Safe to merge; changes are purely additive diagnostics with no behavioral impact on the execution path.

The core execution path is unchanged — describeErrorCause can never throw or mask the original error thanks to its try/catch. The migration script improvement is well-guarded. The only rough edge is the unbounded recursion in printMigrationError, which is harmless in the real failure modes this code targets but worth hardening before it becomes a pattern.

packages/db/scripts/migrate.ts — the recursive printMigrationError helper has no depth guard; worth adding one consistent with the approach used in describeErrorCause.

Important Files Changed

Filename	Overview
apps/sim/lib/workflows/executor/execution-core.ts	Adds describeErrorCause helper and passes its result as a third arg to logger.error; logic is depth-bounded, wrapped in try/catch, and integrates correctly with the structured-JSON logger path.
packages/db/scripts/migrate.ts	Prepends SET statement_timeout = 0 to prevent long DDL from timing out; printMigrationError recurses through the cause chain without a depth guard, which could infinite-loop on a pathological self-referential cause but is safe in practice.
.github/workflows/migrations.yml	Switches migration runner from bunx drizzle-kit migrate to the custom bun run ./scripts/migrate.ts, enabling richer error output and the new statement_timeout = 0 guard.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[executeWorkflowCore throws] --> B[catch block]
    B --> C[logger.error with error + describeErrorCause]
    C --> D{describeErrorCause}
    D --> E{error instanceof Error?}
    E -- No --> F[return undefined]
    E -- Yes --> G[Walk cause chain up to depth 10]
    G --> H{candidate.code defined?}
    H -- Yes --> I[driver = candidate, break]
    H -- No --> J[advance to candidate.cause]
    J --> G
    G -- no Error found --> K[driver = first Error or undefined]
    I --> L[filterUndefined on driver fields]
    K --> L
    L --> M{driver exists?}
    M -- No --> F
    M -- Yes --> N[Return: name/message/code/severity/detail/routine/errno/syscall]
    N --> O[logger merges as cause into structured JSON log entry]
    F --> P[cause field omitted from log entry]

Loading

Comments Outside Diff (1)

1. packages/db/scripts/migrate.ts, line 81-84 (link)

   Unbounded recursion in printMigrationError

   printMigrationError calls itself on error.cause with no depth limit. While real Postgres/Drizzle cause chains are shallow, a contrived or unexpected cycle (e.g. e.cause = e) would overflow the call stack and crash the migration process before it logs anything useful. Adding a simple depth counter (like the depth-bounded walk in describeErrorCause) would make this safe without sacrificing the diagnostic output.

_{Reviews (1): Last reviewed commit: "log(db): Add db failure cause log messag..." | Re-trigger Greptile}

[TheodoreSpeaks]

Superseded by #4749, which isolates the logging change cleanly off staging. The two DB/CI commits on this branch (fix(db): disable statement_timeout, fix(ci): route migration workflow) aren't in staging/main yet and will get their own PR.