Expose the sandbox infrastructure errors reported by the compute plane by devin-ai-integration[bot] · Pull Request #223 · blaxel-ai/sdk-python

devin-ai-integration · 2026-08-20T02:14:36Z

Fixes ENG-5027

Summary

Regenerates the controlplane client from main (make sdk-controlplane) so the new Sandbox.errors history is typed, and surfaces it on both the async and sync sandbox wrappers:

@property
def errors(self) -> list[SandboxInfrastructureError]:  # code, fatal, instance, message, time
    errors = self.sandbox.errors
    return list(errors) if not isinstance(errors, Unset) and errors else []

Entries are oldest first; fatal marks the failure that moved the sandbox to FAILED (a plain VM_EXITED restart is informational). errors is only returned when a single sandbox is read — list/find-all projections don't carry it. message is a normalized reason; raw VM/VMM log lines never leave the compute plane.

tests/manual/sandbox_errors.py prints the history for a sandbox (NAME=<sandbox> uv run python tests/manual/sandbox_errors.py, otherwise it creates and deletes a throwaway one).

The regeneration also dropped the public DriveSpec.size (the control plane no longer exposes it — drives have no size limit), which broke the drive wrapper and its tests. DriveCreateConfiguration.size stays accepted but is now deprecated and no longer serialized into create/update bodies, and DriveInstance.size / SyncDriveInstance.size read the legacy value off spec.additional_properties["size"]. The drive unit tests build their DriveSpec fixtures without size accordingly.

Link to Devin session: https://app.devin.ai/sessions/37e2b2b146c04e8daf2812d81fa0e2e4
Requested by: @drappier-charles

Note

Two new commits since the previous review: drops the now-removed DriveSpec.size from the drive ACL integration fixture, and adds an integration test covering the sandbox.errors accessor on both a single-get and a listing.

^{Written by Mendral for commit aab7db2.}

Note

^{Cursor Bugbot is generating a summary for commit 60aaf6f. Configure here.}

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

devin-ai-integration · 2026-08-20T02:14:40Z

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

Address comments on this PR that start with 'DevinAI' or '@devin'.
Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

Disable automatic comment, CI, and merge conflict monitoring

mendral-app · 2026-08-20T02:16:39Z

🔀 Component Interaction Diagram

Based on the changes in this PR, here's how the sandbox infrastructure error flow works end-to-end:

sequenceDiagram
    participant User as User Application
    participant Wrapper as SandboxInstance<br/>(async/sync wrapper)
    participant API as Generated API Layer<br/>(list_sandboxes, etc.)
    participant Client as HTTP Client (httpx)
    participant CP as Control Plane API
    participant Compute as Compute Plane

    Note over Compute,CP: Error originates in infrastructure
    Compute->>CP: Report infrastructure error<br/>(code, fatal, instance, message, time)
    CP->>CP: Append to sandbox.errors[]

    Note over User,CP: User retrieves sandbox
    User->>Wrapper: sandbox.errors
    Wrapper->>API: get_sandbox(name)
    API->>Client: GET /api/sandboxes/{name}<br/>+ optional ?status= filter
    Client->>CP: HTTP Request
    CP-->>Client: Sandbox JSON (with errors[])
    Client-->>API: Response
    API-->>Wrapper: Sandbox model<br/>(SandboxInfrastructureError[] deserialized)
    Wrapper-->>User: list[SandboxInfrastructureError]<br/>(always a list, never Unset)

    Note over User,CP: Listing (projection excludes errors)
    User->>Wrapper: SandboxInstance.list(status="RUNNING")
    Wrapper->>API: list_sandboxes(status=...)
    API->>Client: GET /api/sandboxes?status=RUNNING
    Client->>CP: HTTP Request
    CP-->>Client: Sandbox[] (errors omitted)
    Client-->>API: Response
    API-->>Wrapper: Sandbox models (errors=UNSET)
    Wrapper-->>User: sandbox.errors → [] (empty list)

Summary of Key Flows

Flow	Description
Error surfacing	New `SandboxInfrastructureError` model propagates compute plane failures through the generated client to user-facing `.errors` property
Status filtering	8 list endpoints gain a `status` query param for server-side filtering
Wrapper contract	`.errors` always returns `list` (never `Unset`/`None`) — listings return `[]`, single gets return actual errors
Drive deprecation	`size` param removed from creation; legacy reads via `additional_properties` (separate from error flow)

New model: SandboxInfrastructureError with fields: code, fatal, instance, message, time

Note

Posted by PR Sequence Diagram · Tag @mendral-app with feedback.

mendral-app · 2026-08-20T02:17:33Z

📋 Created Linear issue ENG-5027 — status: In Progress

Assignee: Charles Drappier (reviewer — bot PR)
Labels: SDK, Sandbox
Project: Sandbox MK3.1 Release
Estimate: L (391 additions, 58 deletions, 20 files)
PR linked: ✅ Issue will auto-close when this PR merges

Auto-created because no Linear reference was found in the PR title, description, or branch name.

Note

Posted by Linear Issue Enforcer · Tag @mendral-app with feedback.

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

mendral-app · 2026-08-20T03:46:09Z

🧪 Testing Guide

What this PR addresses

This PR does three things:

Exposes sandbox infrastructure errors — adds a new errors property to both async and sync sandbox wrappers, returning a list of SandboxInfrastructureError (code, fatal, instance, message, time). These are infrastructure failures reported by the compute plane (e.g., VM_EXITED).
Deprecates DriveSpec.size — the control plane no longer exposes a size limit on drives. DriveCreateConfiguration.size is still accepted but no longer serialized. DriveInstance.size reads a legacy value from spec.additional_properties["size"].
Adds status filter parameter to all list endpoints (agents, applications, sandboxes, drives, functions, jobs, models, volumes) via a controlplane client regeneration.

Fixes ENG-5027.

Steps to reproduce / exercise the new behavior

Sandbox errors:

Set up a Blaxel workspace with sandbox access.
Run the manual test: NAME=<existing-sandbox> uv run python tests/manual/sandbox_errors.py (or omit NAME= to have it create/delete a throwaway sandbox).
To see actual errors, provoke a sandbox failure (e.g., use a sandbox that has previously failed or trigger a VM exit condition).
Alternatively, run the integration test: pytest tests/integration/core/sandbox/test_sandbox_errors.py — this verifies .errors returns a list on both a single-get and a list-all response.

Drive size deprecation:

Run existing drive unit tests: pytest tests/core/test_drive.py
Run the drive ACL integration test: pytest tests/integration/core/sandbox/test_drive_acl.py
Verify drives can still be created without specifying a size.

What to verify (expected behavior)

sandbox.errors returns list[SandboxInfrastructureError] (possibly empty) when reading a single sandbox.
sandbox.errors returns [] when accessed from a list/find-all response (errors aren't projected on listings).
Each error entry has: code, fatal (bool), instance, message, time.
fatal=True marks the failure that moved the sandbox to FAILED state; non-fatal entries (e.g., VM_EXITED restarts) are informational.
Drive creation no longer sends size in the request body, even if DriveCreateConfiguration.size is set.
DriveInstance.size still returns the legacy value (from additional_properties) for old drives.
All list endpoints accept an optional status query parameter for filtering.
Existing unit tests pass: pytest tests/core/test_drive.py
No regressions in sandbox or drive functionality.

Note

Posted by PR Testing Guide · Tag @mendral-app with feedback.

devin-ai-integration

Devin Review found 4 potential issues.

devin-ai-integration · 2026-08-20T03:49:36Z

+    async def test_reads_empty_error_history_on_healthy_sandbox(self):
+        """A sandbox that never hit an infrastructure failure reports no error."""
+        name = unique_name("errors")
+        await SandboxInstance.create(
+            {
+                "name": name,
+                "image": default_image,
+                "region": default_region,
+                "labels": default_labels,
+            }
+        )
+
+        try:
+            sandbox = await SandboxInstance.get(name)
+
+            # The shape of an entry (code, fatal, instance, message, time) is
+            # not exercisable here: covering it would mean provoking a real
+            # infrastructure failure on the compute plane.
+            assert isinstance(sandbox.errors, list)
+            assert sandbox.errors == []
+        finally:
+            await SandboxInstance.delete(name)


🟡 New sandbox test cleans up outside the required class-level fixture

The new sandbox test deletes its sandbox inline in a try/finally block (tests/integration/core/sandbox/test_sandbox_errors.py:41) instead of the class-level cleanup fixture the repository conventions mandate, so a hard failure or interruption before the finally block leaves a sandbox behind.
Impact: Leftover test sandboxes can accumulate in the workspace.

Repository convention on integration test cleanup

AGENTS.md (Testing section) states: "Always clean up sandboxes in class-level cleanup fixtures". Other suites follow this, e.g. tests/integration/core/sandbox/test_drive_acl.py:118-130 defines an autouse class-scoped cleanup fixture tracking created sandboxes/drives. The new TestSandboxErrors class has no such fixture.

Prompt for agents

TestSandboxErrors in tests/integration/core/sandbox/test_sandbox_errors.py creates a sandbox and deletes it inline via try/finally. Per AGENTS.md, integration tests must clean up sandboxes in a class-level (autouse, class-scoped) cleanup fixture that tracks the created resource names, as done in tests/integration/core/sandbox/test_drive_acl.py. Refactor the test to register created sandbox names on the class and delete them in such a fixture.

Was this helpful? React with 👍 or 👎 to provide feedback.

Valid — fixed in #224 (this PR was already merged): an autouse class-scoped cleanup fixture now deletes every sandbox the class created, mirroring test_drive_acl.py.

devin-ai-integration · 2026-08-20T03:49:37Z

        display_name: str | None = None,
        labels: Dict[str, str] | None = None,
-        size: int | None = None,  # Size in GB
+        size: int | None = None,  # Deprecated: drives have no size limit, this is ignored


🔍 Deprecated drive size is now silently ignored

DriveCreateConfiguration(size=...) is still accepted and stored on the config object, but is no longer serialized into create or update bodies (src/blaxel/core/drive/drive.py:206-224, src/blaxel/core/drive/drive.py:566-582). Callers who previously relied on setting a size will get no error and no warning, and after an update round-trip the legacy size value read from spec.additional_properties will disappear from the wrapper's view (the merged spec never carries the old value forward). Consider emitting a DeprecationWarning when a non-None size is passed so silent behavior change is visible.

Was this helpful? React with 👍 or 👎 to provide feedback.

Deliberate: the control plane dropped the field from the public spec, so there is nothing to send and an update round-trip legitimately loses the legacy value. I left size accepted so existing callers keep compiling rather than breaking at the call site. Adding a DeprecationWarning would make it noisier for callers who can't do anything about it, so I'd rather not do it without @drappier-charles asking for it.

devin-ai-integration · 2026-08-20T03:49:38Z

+    async def test_listing_does_not_carry_the_error_history(self):
+        """Listings project the field out, so they must not be read for it."""
+        page = await SandboxInstance.list(limit=1)
+
+        assert len(page.data) > 0
+        assert page.data[0].errors == []


🔍 Listing assertion depends on pre-existing workspace state

assert len(page.data) > 0 requires that the workspace already contains at least one sandbox; the test creates none of its own. In a clean workspace (or one where the only sandbox is created/deleted by the sibling test) this will fail spuriously. Creating a labeled sandbox in this test (and cleaning it up) would make the assertion deterministic.

Was this helpful? React with 👍 or 👎 to provide feedback.

Valid too, fixed in the same PR (#224): the listing test now creates its own labeled sandbox before listing, so len(page.data) > 0 no longer depends on pre-existing workspace state.

devin-ai-integration · 2026-08-20T03:49:39Z

+        errors = []
+        _errors = d.pop("errors", UNSET)
+        for errors_item_data in _errors or []:
+            errors_item = SandboxInfrastructureError.from_dict(errors_item_data)
+
+            errors.append(errors_item)
+


🔍 errors defaults to empty list rather than UNSET after deserialization

Like the existing events handling, errors is initialized to [] and only filled when the key is present, so a payload without errors yields errors=[] instead of UNSET. Round-tripping such a Sandbox through to_dict() will now emit "errors": [], which the control plane may interpret as an explicit (read-only) value on update bodies. This mirrors pre-existing behavior for events, so it is likely tolerated server-side, but worth confirming that update endpoints ignore the field.

Was this helpful? React with 👍 or 👎 to provide feedback.

Generated by openapi-python-client, so not editable here, and it mirrors the pre-existing events handling. Server side it's harmless: errors is readOnly in the spec and the controlplane update path never reads it off the body (the array is only ever appended to by the workload-error workflow), so an incoming "errors": [] is ignored.

cursor

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 60aaf6f. Configure here.}

cursor · 2026-08-20T03:51:23Z

+        page = await SandboxInstance.list(limit=1)
+
+        assert len(page.data) > 0
+        assert page.data[0].errors == []


Listing errors test assumes existing sandboxes

Medium Severity

test_listing_does_not_carry_the_error_history lists one sandbox and requires page.data to be non-empty. The sibling test creates a sandbox and deletes it in finally, so a clean workspace or this file run alone leaves nothing to list and the assertion fails independently of the errors accessor.

^{Reviewed by Cursor Bugbot for commit 60aaf6f. Configure here.}

Same finding as Devin Review's, already fixed in #224: the listing test creates its own labeled sandbox, and cleanup moved to a class-scoped fixture so nothing is deleted mid-class.

Expose the sandbox infrastructure errors reported by the compute plane

d040aec

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

drappier-charles self-assigned this Aug 20, 2026

drappier-charles self-requested a review August 20, 2026 02:14

This comment was marked as outdated.

Sign in to view

drappier-charles and others added 2 commits August 20, 2026 02:23

Drop the removed DriveSpec size from the drive ACL integration fixtures

a1baff9

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

Cover the sandbox error history with an integration test

aab7db2

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

This comment was marked as outdated.

Sign in to view

Drop the unreachable entry-shape assertions from the errors test

60aaf6f

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

drappier-charles marked this pull request as ready for review August 20, 2026 03:48

drappier-charles merged commit 90210d6 into main Aug 20, 2026
19 of 21 checks passed

drappier-charles deleted the cdrappier/devin/sandbox-errors-sdk branch August 20, 2026 03:48

devin-ai-integration Bot commented Aug 20, 2026

View reviewed changes

devin-ai-integration Bot mentioned this pull request Aug 20, 2026

Clean up the sandbox errors test through a class fixture #224

Closed

cursor Bot reviewed Aug 20, 2026

View reviewed changes

Uh oh!

Conversation

devin-ai-integration Bot commented Aug 20, 2026 • edited by cursor Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Uh oh!

devin-ai-integration Bot commented Aug 20, 2026

🤖 Devin AI Engineer

Uh oh!

This comment was marked as outdated.

Uh oh!

mendral-app Bot commented Aug 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔀 Component Interaction Diagram

Summary of Key Flows

Uh oh!

mendral-app Bot commented Aug 20, 2026

Uh oh!

This comment was marked as outdated.

Uh oh!

mendral-app Bot commented Aug 20, 2026

🧪 Testing Guide

What this PR addresses

Steps to reproduce / exercise the new behavior

What to verify (expected behavior)

Uh oh!

Uh oh!

devin-ai-integration Bot left a comment

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot Aug 20, 2026

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot Aug 20, 2026

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot Aug 20, 2026

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot Aug 20, 2026

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot Aug 20, 2026

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot Aug 20, 2026

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot Aug 20, 2026

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot Aug 20, 2026

Choose a reason for hiding this comment

Uh oh!

cursor Bot left a comment

Choose a reason for hiding this comment

Uh oh!

cursor Bot Aug 20, 2026

Choose a reason for hiding this comment

Listing errors test assumes existing sandboxes

Uh oh!

devin-ai-integration Bot Aug 20, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

devin-ai-integration Bot commented Aug 20, 2026 •

edited by cursor Bot

Loading

mendral-app Bot commented Aug 20, 2026 •

edited

Loading