Skip to content

Raise execute-code materialization limits with skip logs - #968

Draft
Y1fe1Zh0u wants to merge 1 commit into
dataelement:v1.11.4from
Y1fe1Zh0u:fix/execute-code-materialize-limits
Draft

Raise execute-code materialization limits with skip logs#968
Y1fe1Zh0u wants to merge 1 commit into
dataelement:v1.11.4from
Y1fe1Zh0u:fix/execute-code-materialize-limits

Conversation

@Y1fe1Zh0u

Copy link
Copy Markdown
Collaborator

What changed

  • Raise the execute-code temporary-workspace per-file materialization limit from 10 MiB to 50 MiB.
  • Raise the aggregate materialization limit from 100 MiB to 500 MiB.
  • Emit warning logs when a file is skipped by either limit, including path, size, limit, and reason.
  • Add regression coverage for the new boundaries, both skip reasons, and files above the previous 10 MiB cap.

Why

Binary task inputs larger than 10 MiB were silently omitted from the temporary sandbox. Storage APIs could list the file while execute_code could not see it, which made the failure look like an overlay or filesystem inconsistency.

Impact

Documents up to 50 MiB can now be materialized for local tool execution, subject to a 500 MiB aggregate budget. Files outside either budget remain skipped, but the backend now records an actionable warning.

Validation

  • 18 passed in tests/test_agent_tools_storage_workspace.py
  • Ruff passed for the changed test module
  • git diff --check passed

Known gap

Live deployment performance with a fully eligible 500 MiB workspace has not been measured.

Raise execute-code workspace materialization limits so ordinary binary documents remain available inside the sandbox, and log files skipped at either budget boundary.

Constraint: Per-file materialization is capped at 50 MiB and aggregate materialization at 500 MiB.

Rejected: Silently retain the 10 MiB limit | task inputs disappeared without actionable evidence.

Confidence: high

Scope-risk: narrow

Directive: Keep skip logging when changing materialization budgets.

Tested: 18 scoped storage-workspace tests; scoped test Ruff; git diff --check.

Not-tested: Live deployment performance with a 500 MiB eligible workspace.
@Y1fe1Zh0u
Y1fe1Zh0u force-pushed the fix/execute-code-materialize-limits branch from 5fdf244 to c88958a Compare August 14, 2026 12:12
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant