autoresume state handling improvements

[matthewlouisbrockman]

auto-resume was not mirroring the state handling like in /resume and /connect. On catalog miss, it could return routing for sandboxes still in pausing/snapshotting transition, which let post-connect run traffic hit 502s.

Changes:

• orchestrator now has HandleExistingSandboxAutoResume in autoresume.go to handle the state transitions.

---

Note

Medium Risk
Touches auto-resume/resume routing and orchestrator state handling, which can change when requests fall back to DB-backed resumes or return 409s; misclassification could increase resume failures or transient errors.

Overview
Auto-resume now checks orchestrator state before attempting DB-backed resume: running sandboxes return routing immediately, pausing/snapshotting sandboxes are waited on and re-checked (bounded retries/time budget), killing/unknown states fail, and snapshot metadata is reloaded before fallback to avoid resuming from stale pre-pause data. A new shared SandboxStillTransitioningMessage is returned as FailedPrecondition and is propagated through client-proxy into a dedicated SandboxStillTransitioningError, which the HTTP proxy renders as a 409 (JSON or browser HTML) with added tests covering the new state-machine and error-template behavior.

^{Written by Cursor Bugbot for commit 2eb2308. This will update automatically on new commits. Configure here.}

[matthewlouisbrockman]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: aa2c4735a8

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[cursor]

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

Autofix Details

Bugbot Autofix prepared a fix for the issue found in the latest run.

• ✅ Fixed: Tight spin loop on failed snapshot transition
  • Added a retry backoff when sandbox state remains snapshotting after refresh so auto-resume no longer busy-spins on failed snapshot transitions.

Or push these changes by commenting:

@cursor push 8d53281ca5

Preview (8d53281ca5)

diff --git a/packages/api/internal/handlers/proxy_grpc.go b/packages/api/internal/handlers/proxy_grpc.go
--- a/packages/api/internal/handlers/proxy_grpc.go
+++ b/packages/api/internal/handlers/proxy_grpc.go
@@ -24,6 +24,8 @@
 	sharedutils "github.com/e2b-dev/infra/packages/shared/pkg/utils"
 )
 
+const snapshottingStateRetryDelay = 100 * time.Millisecond
+
 type SandboxService struct {
 	proxygrpc.UnimplementedSandboxServiceServer
 
@@ -134,6 +136,14 @@
 
 			updatedSandbox, getSandboxErr := getSandbox(ctx)
 			if getSandboxErr == nil {
+				if sbx.State == sandbox.StateSnapshotting && updatedSandbox.State == sandbox.StateSnapshotting {
+					select {
+					case <-time.After(snapshottingStateRetryDelay):
+					case <-ctx.Done():
+						return "", false, status.Error(codes.Internal, "error waiting for sandbox snapshot to finish")
+					}
+				}
+
 				sbx = updatedSandbox
 
 				continue

diff --git a/packages/api/internal/handlers/proxy_grpc_test.go b/packages/api/internal/handlers/proxy_grpc_test.go
--- a/packages/api/internal/handlers/proxy_grpc_test.go
+++ b/packages/api/internal/handlers/proxy_grpc_test.go
@@ -5,6 +5,7 @@
 	"errors"
 	"fmt"
 	"testing"
+	"time"
 
 	"github.com/google/uuid"
 	"github.com/stretchr/testify/assert"
@@ -404,6 +405,45 @@
 		assert.Equal(t, "error waiting for sandbox snapshot to finish", st.Message())
 	})
 
+	t.Run("snapshotting sandbox with no active transition does not spin", func(t *testing.T) {
+		t.Parallel()
+
+		ctx, cancel := context.WithTimeout(t.Context(), 20*time.Millisecond)
+		defer cancel()
+
+		waitCalls := 0
+		getCalls := 0
+
+		_, handled, err := handleExistingSandboxAutoResume(
+			ctx,
+			"test-sandbox",
+			testSandboxForAutoResume(sandbox.StateSnapshotting),
+			func(context.Context) error {
+				waitCalls++
+
+				return nil
+			},
+			func(context.Context) (sandbox.Sandbox, error) {
+				getCalls++
+
+				return testSandboxForAutoResume(sandbox.StateSnapshotting), nil
+			},
+			func(sandbox.Sandbox) (string, error) {
+				t.Fatal("getNodeIP should not be called while sandbox remains snapshotting")
+
+				return "", nil
+			},
+		)
+		require.Error(t, err)
+		assert.False(t, handled)
+		st, ok := status.FromError(err)
+		require.True(t, ok)
+		assert.Equal(t, codes.Internal, st.Code())
+		assert.Equal(t, "error waiting for sandbox snapshot to finish", st.Message())
+		assert.Equal(t, 1, waitCalls)
+		assert.Equal(t, 1, getCalls)
+	})
+
 	t.Run("pausing sandbox returns internal error when refreshed sandbox lookup fails unexpectedly", func(t *testing.T) {
 		t.Parallel()

_{This Bugbot Autofix run was free. To enable autofix for future PRs, go to the Cursor dashboard.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 964c63291e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 99c73584aa

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[jakubno]

just NITs

[cursor]

PR Summary

Medium Risk
Changes auto-resume control flow and error propagation across API, orchestrator, and proxies, which could alter client-visible behavior during sandbox transitions. Risk is moderated by added unit/integration tests for the new transition and error-handling paths.

Overview
Improves sandbox auto-resume to avoid returning routing information while a sandbox is still pausing/snapshotting by adding orchestrator-side state handling with bounded waiting/retries, reloading snapshot metadata before resume, and propagating a dedicated FailedPrecondition (sandbox is still transitioning) signal end-to-end so the client proxy and shared proxy can surface a 409 "still transitioning" response (JSON or browser HTML) instead of 502s.

^{Written by Cursor Bugbot for commit 724c237. This will update automatically on new commits. Configure here.}

[cursor]

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

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