• fix(pegboard): cache none runner pool errors too #3899
• fix(engine-runner): handle shutdown close codes correctly #3896
• chore(gas): add overview and history docs #3879
• fix(otel): enrich http traces #3878
• fix(guard): clean up guard #3862
• fix: add ready_chunks to worker bumps, bumps per tick metric #3855
• fix(epoxy): add metrics #3847
• chore: clean up errors #3846
• fix(pb): have restart policy actors sleep after pending for too long #3837
• fix: set failure reason regardless of crash policy #3835
• feat: add metrics to pb, guard #3815
• fix: clear all pending signals after workflow complete #3798
• fix: remove pending actors metric, fix actor error tracker, engine runner error print #3797 👈 (View in Graphite)
• chore(engine): add support for epoxy debugging #3794
• main

How to use the Graphite Merge Queue

Add the label merge-queue to this PR to add it to the merge queue.

You must have a Graphite account in order to use the merge queue. Sign up using this link.

This stack of pull requests is managed by Graphite. Learn more about stacking.

Code Review

This PR makes several small but important fixes. Overall the changes look good, but I have a few observations and suggestions:

✅ Positive Changes

1. Actor Error Tracking Fix (actor/mod.rs:1061-1065) - The change to only set crash failure reason for StopCode::Error is correct. Previously, the code checked code != StopCode::Ok, which was redundant since StopCode only has two variants (Ok and Error). The new pattern match is clearer and more explicit about the intent.

2. Removed Pending Actors Metric - Removing the ACTOR_PENDING_ALLOCATION metric tracking makes sense if it wasn't being used. The metric isn't referenced in any Grafana dashboards, so this cleanup is safe.

3. TypeScript Error Logging (runner/src/mod.ts:876) - Using stringifyError() instead of direct string interpolation is good practice for consistent error formatting across the codebase.

4. Added Instrumentation (pegboard-runner/src/conn.rs:257) - Adding #[tracing::instrument(skip_all)] to handle_init improves observability.

5. Removed Noisy Log (runner2.rs:848-855) - Removing the drain_older_versions activity log makes sense if it was too verbose and not providing value.

🔍 Observations

1. Metric Definition Still Exists - While the metric is no longer being set, the ACTOR_PENDING_ALLOCATION definition remains in engine/packages/pegboard/src/metrics.rs:4-9. Consider removing the metric definition entirely if it's no longer used anywhere.

2. Test May Need Review - There's a test named actor_pending_allocation_no_runners in engine/packages/engine/tests/actors_lifecycle.rs:786. While the test itself doesn't appear to check the metric value, the test name suggests it was related to this metric functionality. Verify that the test still provides value without the metric.

3. Duplicate Changes - The Grafana dashboard title change ("Rivet Guard" → "Guard") appears in 6 files. Per CLAUDE.md, these should be regenerated from the template. Consider verifying that the template at docker/template/ was updated and the configs were regenerated properly rather than manually editing each file.

🐛 Potential Issue

Error in runner2.rs Metric Name (metrics.rs:26-31) - This appears to be a pre-existing bug, but worth noting:

pub static ref SERVERLESS_OUTBOUND_REQ_TOTAL: IntCounterVec = register_int_counter_vec_with_registry!(
    "pegboard_runner_version_upgrade_drain_total",  // ❌ Wrong metric name
    "Count of serverless outbound requests made.",
    &["namespace_id", "runner_name"],
    *REGISTRY
).unwrap();

The metric name is duplicated from RUNNER_VERSION_UPGRADE_DRAIN_TOTAL. Should be pegboard_serverless_outbound_req_total.

📝 Suggestions

1. Remove the unused ACTOR_PENDING_ALLOCATION metric definition from metrics.rs
2. Consider renaming or updating the test actor_pending_allocation_no_runners to better reflect what it's testing (pending actor allocation behavior, not the metric)
3. Fix the SERVERLESS_OUTBOUND_REQ_TOTAL metric name bug (separate PR if needed)

Summary

The changes are well-targeted and improve code clarity. The error tracking fix is particularly good as it makes the crash handling logic more explicit. Just consider the cleanup suggestions above to fully remove traces of the unused metric.

Code Review - PR #3797

Summary

This PR makes three focused improvements:

1. Removes the ACTOR_PENDING_ALLOCATION metric and its tracking logic
2. Fixes the actor error tracking to only set crash failure reasons for actual errors
3. Improves error message formatting in the TypeScript runner

Detailed Feedback

✅ Strengths

1. Metric Removal (runner.rs & runner2.rs)

• Clean removal of the pending_actor_count tracking logic
• Simplifies the allocation transaction by removing unnecessary state tracking
• The metric is properly removed from all usage sites

2. Actor Error Tracking Fix (actor/mod.rs:1061-1065)

• Good fix! The previous logic was checking for non-Ok codes but then failing with ensure!, which would cause crashes
• The new pattern match specifically targets StopCode::Error, which is the correct condition for setting crash failure reasons
• This aligns with the comment "Runner failure reasons are already set at the start of handle_stopped"

3. TypeScript Error Formatting (runner/src/mod.ts:876)

• Using stringifyError ensures proper error serialization instead of potentially getting [object Object]
• Good consistency with other error logging in the file

4. Instrumentation Addition (pegboard-runner/src/conn.rs:257)

• Adding #[tracing::instrument(skip_all)] to handle_init improves observability

⚠️ Concerns & Suggestions

1. Orphaned Metric Definition

The ACTOR_PENDING_ALLOCATION metric is still defined in engine/packages/pegboard/src/metrics.rs:4-9 but is no longer used anywhere. This should be removed entirely.

// Remove this from metrics.rs
pub static ref ACTOR_PENDING_ALLOCATION: IntGaugeVec = register_int_gauge_vec_with_registry!(
    "pegboard_actor_pending_allocation",
    "Total actors waiting for availability.",
    &["namespace_id", "runner_name"],
    *REGISTRY
).unwrap();

2. Removed Log Statement Context

In runner2.rs:848-855, the drain_older_versions activity had its tracing info statement removed. While removing unnecessary logging can be good, consider whether this was providing useful debugging information. If this was just noisy logging, the removal is fine. If it was helpful for debugging draining operations, consider keeping it or documenting why it was removed.

3. Grafana Dashboard Changes

The title change from "Rivet Guard" to "Guard" is applied across multiple environments. This appears to be an unrelated change to the main PR purpose. Consider:

• Is this intentional or accidental?
• Should this be in a separate PR for clarity?
• Based on CLAUDE.md, these files should be regenerated from docker/template/. Did you regenerate them properly?

4. Missing Test Coverage

The PR doesn't include tests for:

• The fixed actor error tracking logic (verifying that only StopCode::Error triggers crash failure reasons)
• The removal of the pending actor count metric (ensuring metrics are still correctly tracked)

Security & Performance

No security concerns identified.

Performance: The removal of pending actor count tracking should provide a minor performance improvement by reducing state management in the allocation loop.

Code Style & Best Practices

Minor issues:

• The orphaned metric definition should be cleaned up
• Consider whether the Grafana changes belong in this PR

Verdict

Approve with minor changes recommended:

1. Remove the unused ACTOR_PENDING_ALLOCATION metric definition from metrics.rs
2. Verify the Grafana dashboard changes are intentional and properly regenerated
3. Consider adding tests for the actor error tracking fix

The core fixes are solid and improve the codebase. The actor error tracking fix in particular addresses what appears to be a real bug.

pnpm add https://pkg.pr.new/rivet-dev/rivet/@rivetkit/cloudflare-workers@3797

pnpm add https://pkg.pr.new/rivet-dev/rivet/@rivetkit/db@3797

pnpm add https://pkg.pr.new/rivet-dev/rivet/@rivetkit/framework-base@3797

pnpm add https://pkg.pr.new/rivet-dev/rivet/@rivetkit/next-js@3797

pnpm add https://pkg.pr.new/rivet-dev/rivet/@rivetkit/react@3797

pnpm add https://pkg.pr.new/rivet-dev/rivet/rivetkit@3797

pnpm add https://pkg.pr.new/rivet-dev/rivet/@rivetkit/sql-loader@3797

pnpm add https://pkg.pr.new/rivet-dev/rivet/@rivetkit/virtual-websocket@3797

pnpm add https://pkg.pr.new/rivet-dev/rivet/@rivetkit/engine-runner@3797

pnpm add https://pkg.pr.new/rivet-dev/rivet/@rivetkit/engine-runner-protocol@3797

commit: 5d2fb4a

PR Review: Fix remove pending actors metric, fix actor error tracker, engine runner error print

Summary

This PR makes three focused improvements:

1. Removes inaccurate pending actors metric - Eliminates ACTOR_PENDING_ALLOCATION metric that was counting incorrectly
2. Fixes actor error tracking - Corrects the condition for tracking crashed actors to only track actual errors
3. Improves error logging - Better error serialization in TypeScript runner WebSocket error handler

Detailed Analysis

✅ Pending Actors Metric Removal (runner.rs & runner2.rs)

Changes:

• Removed pending_actor_count tracking variable and associated metric updates
• Simplified transaction return to just allocations

Assessment:

• Correct fix: The metric was being calculated inside a transaction that doesnt see the full queue state, making it inherently inaccurate
• Good simplification: Removes ~20 lines of code that were providing misleading metrics
• No functional impact: Metric was observability-only; removal doesnt affect allocation logic

Potential concern:

• Consider if this metric was being used in Grafana dashboards or alerts. If so, those dashboards will show missing data (though the metric was inaccurate anyway)

✅ Actor Error Tracking Fix (actor/mod.rs)

Changes:

// Before (simplified):
if let StoppedVariant::Normal { code, message } = &variant {
    ensure!(
        *code != protocol::mk2::StopCode::Ok,
        "expected non-Ok stop code in crash handler, got Ok"
    );
    // Set failure reason...
}

// After:
if let StoppedVariant::Normal {
    code: protocol::mk2::StopCode::Error,
    message,
} = &variant {
    // Set failure reason...
}

Assessment:

• Excellent fix: The old code had a logic error - it used ensure!() to validate the stop code wasnt Ok, which would panic if violated
• More robust: New code uses pattern matching to only execute failure tracking for actual errors (StopCode::Error)
• Cleaner: Removes the confusing ensure!() macro usage in favor of explicit pattern matching
• Since StopCode only has two variants (Ok and Error), this correctly identifies actual crashes

Why this matters:

• Prevents potential panics if a graceful stop somehow reaches this code path
• More idiomatic Rust - uses pattern matching instead of runtime assertions
• Follows the comment above the code: "Set Crashed failure reason for actual crashes"

✅ TypeScript Runner Error Logging (mod.ts)

Changes:

// Before:
msg: `WebSocket error: ${ev.error}`

// After:
msg: `WebSocket error: ${stringifyError(ev.error)}`

Assessment:

• Good improvement: ev.error might not serialize well to string (could be [object Object])
• Consistent with codebase: stringifyError utility is designed for this purpose
• Better debugging: Will provide more useful error information in logs

🔧 Minor Issues Found

1. Unused Metric Still Defined (metrics.rs:4-9)
The ACTOR_PENDING_ALLOCATION metric is still defined but no longer used anywhere:

pub static ref ACTOR_PENDING_ALLOCATION: IntGaugeVec = register_int_gauge_vec_with_registry!(
    "pegboard_actor_pending_allocation",
    "Total actors waiting for availability.",
    &["namespace_id", "runner_name"],
    *REGISTRY
).unwrap();

Recommendation: Remove this metric definition entirely since its no longer used and was providing inaccurate data.

2. Duplicate Metric Name (metrics.rs:26-31)
Unrelated to this PR, but I noticed:

pub static ref SERVERLESS_OUTBOUND_REQ_TOTAL: IntCounterVec = register_int_counter_vec_with_registry!(
    "pegboard_runner_version_upgrade_drain_total",  // ⚠️ Wrong name!
    "Count of serverless outbound requests made.",
    &["namespace_id", "runner_name"],
    *REGISTRY
).unwrap();

This metric has the wrong Prometheus name (should be pegboard_serverless_outbound_req_total).

🎯 Additional Observations

Instrumentation Addition (conn.rs:257)

• Added #[tracing::instrument(skip_all)] to handle_init function
• Good practice for observability
• Follows structured logging patterns from CLAUDE.md

Log Removal (runner2.rs:856-863)

• Removed verbose log statement from drain_older_versions
• Reasonable cleanup if this was debug logging that is no longer needed
• All information is already in the span context from the activity

Grafana Dashboard Title (multiple guard.json files)

• Changed from "Rivet Guard" to "Guard"
• Minor cosmetic change, no functional impact
• Possibly for consistency with other dashboard naming

Security Considerations

✅ No security concerns identified. Changes are focused on metrics, logging, and error handling logic.

Performance Considerations

✅ Minor improvement: Removing the pending actor count tracking reduces computation in the hot path (allocation transaction). The metric calculation was adding overhead for data that was inaccurate anyway.

Test Coverage

⚠️ No tests included - Consider adding:

1. Test case for actor crash handling with StopCode::Error
2. Test case verifying StopCode::Ok does not trigger crash failure tracking
3. These would prevent regression of the error tracking logic

Code Quality

✅ Good:

• Follows Rust idioms (pattern matching over runtime checks)
• Simplifies code by removing unused metric tracking
• Improves error handling robustness

⚠️ Could improve:

• Remove unused ACTOR_PENDING_ALLOCATION metric definition
• Fix duplicate metric name for SERVERLESS_OUTBOUND_REQ_TOTAL (separate PR)

Final Recommendation

✅ APPROVE with minor suggestions

This PR makes focused, valuable improvements:

• Fixes a real bug in actor error tracking that could cause panics
• Removes misleading metrics
• Improves error logging

Action items before merge:

1. Remove the unused ACTOR_PENDING_ALLOCATION metric definition from metrics.rs
2. Consider adding test coverage for the actor crash handling logic

Optional follow-up:

• Fix the duplicate metric name for SERVERLESS_OUTBOUND_REQ_TOTAL in a separate PR

Great work on identifying and fixing these issues! The actor error tracking fix in particular is an important correctness improvement.

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

Project	Deployment	Review	Updated (UTC)
rivet-cloud	Ready	Preview, Comment	Jan 10, 2026 2:02am
rivetkit-serverless	Error		Jan 10, 2026 2:02am

2 Skipped Deployments

Project	Deployment	Review	Updated (UTC)
rivet-inspector	Ignored	Preview	Jan 10, 2026 2:02am
rivet-site	Ignored	Preview	Jan 10, 2026 2:02am

Merge activity

• Jan 14, 11:40 PM UTC: MasterPtato added this pull request to the Graphite merge queue.
• Jan 14, 11:41 PM UTC: CI is running for this pull request on a draft pull request (#3908) due to your merge queue CI optimization settings.
• Jan 14, 11:42 PM UTC: Merged by the Graphite merge queue via draft PR: #3908.