fix(executor): retry transient connection failures when binding a tool call's delegation

[waleedlatif1]

Summary

• Every tool call a workflow run delegates (Function, MCP, files, knowledge) first binds its executor delegation by loading the canonical workflow/run context — a plain read with no retry anywhere above it
• A single reset pooled connection (ECONNRESET) on that read failed the whole block, and the tool layer masks DB errors, so users saw "An internal error occurred while executing the tool. Please try again." with no retry actually happening
• Wrap the canonical loads in bindInternalExecutorDelegation with the existing withDatabaseReadRetry (at most 3 attempts, transient connection/serialization codes only); not_found and non-transient errors still propagate unchanged on the first attempt

Type of Change

• Bug fix

Testing

• Added tests for a reset connection on the run load and on the current-workflow load; confirmed both fail without the fix and pass with it (18/18)
• Existing infrastructure-failure test now also asserts a non-transient error is not retried
• bun run type-check, bun run lint, check:audits, block-registry, and docs-manifest:check pass

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.

1 Skipped Deployment

Project	Deployment	Actions	Updated
docs	Skipped		Sep 22, 2026 10:29pm UTC

[cubic-dev-ai]

No issues found across 2 files

Confidence score: 5/5

• Automated review surfaced no issues in the provided summaries.
• No files require special attention.

_{Re-trigger cubic}

[greptile-apps]

Confidence Score: 5/5

The PR appears safe to merge; the new retries are limited to idempotent canonical reads and preserve existing delegation validation and failure semantics.

Summary

This PR makes executor-delegation binding resilient to transient database read failures without changing its authorization or error-classification behavior.

• Wraps root workflow, run, execution, and current-workflow canonical context reads with the existing bounded database-read retry helper.
• Preserves immediate propagation for missing resources and non-transient infrastructure failures.
• Adds coverage for transient connection resets on run and deployment-version loads and verifies that permanent failures are attempted only once.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  Claims[Verified executor delegation] --> Root[Load canonical root context]
  Root --> RetryRoot{Transient database error?}
  RetryRoot -->|Yes, attempts remain| Root
  RetryRoot -->|No| Validate[Validate workflow and execution binding]
  Validate --> Current{Current workflow present?}
  Current -->|Yes| Child[Load current workflow context]
  Child --> RetryChild{Transient database error?}
  RetryChild -->|Yes, attempts remain| Child
  RetryChild -->|No| Scope[Verify workspace consistency]
  Current -->|No| Principal[Create scoped executor principal]
  Scope --> Principal

Loading

_{Reviews (1) · Last reviewed commit: "fix(executor): retry transient connectio..."}