• chore(rivetkit): remove inline tests #4668
• fix(guard): serialize gateway actor keys correctly #4655
• fix(api): subscribe before namespace workflow dispatch #4654
• fix(rivetkit): restore hibernatable sockets and hydrate serverless starts #4658
• fix(rivetkit-native): expose full hibernation metadata to JS #4657
• fix(pegboard): persist and replay hibernating requests #4656 👈 (View in Graphite)
• fix(rivetkit): keep internal error exposure behavior consistent #4661
• test(rivetkit): stabilize actor db driver tests #4664
• fix(test): stabilize lifecycle, sleep, queue, and run edge cases #4660 : 1 other dependent PR (#4665 )
• test(rivetkit): re-enable gateway URL and direct-registry coverage #4659
• fix(rivetkit): add lifecycle error retry and gateway HTTP routing #4666
• chore(engine): publish engine bases in ci #4649 : 2 other dependent PRs (#4650 , #4652 )
• chore: lockfile lefthook #4647
• fix(rivetkit): surface native sqlite kv errors to callers #4645
• fix(rivetkit): prevent sleep races during disconnect and db work #4644
• fix(rivetkit): isolate engine envoys and propagate startup failures #4643
• test(rivetkit): remove wasm sqlite-only fixtures and coverage #4642
• chore(sqlite-vfs): remove wasm sqlite packages and workspace wiring #4641
• chore(rivetkit): remove wasm sqlite runtime #4614
• fix(sqlite-native): restore native startup kv preload #4639
• fix(sqlite-native): delete metadata before chunk range #4638
• fix(sqlite-native): keep truncate cache coherent #4637
• fix(sqlite-vfs): use delete range for truncate cleanup #4636
• fix(sqlite-native): restore kv error hook #4635
• perf(sqlite-native): avoid cloning cached read chunks #4634
• perf(sqlite-native): gate kv operation labels #4633
• perf(sqlite-native): reuse read cache for partial writes #4632
• perf(sqlite-native): remove delete file existence probe #4631
• fix(sqlite-native): gate native read cache behind env var #4630
• fix(pegboard): isolate runner config dc lookup failures #4625 : 1 other dependent PR (#4626 )
• docs(pegboard): note runner config upsert split write risk #4624
• fix(pegboard): unpack global runner config keys correctly #4623
• fix(pegboard-gateway): enforce tunnel message state #4622
• fix(pegboard-runner): clear terminal tunnel routes #4621
• fix(pegboard): skip protocol version keys in runner pool backfill #4620
• fix(pegboard): create missing normal runner configs #4619
• fix(epoxy): reuse cached optimistic value #4618
• main

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

PR Review: fix(pegboard): persist and replay hibernating requests

Note: This is a draft PR. The description/test plan are not filled in.

Summary

This PR fixes a bug where hibernating WebSocket request IDs were not being persisted and replayed when an actor restarts. The old approach relied on a gateway guard intercepting CommandStartActor and populating hibernating_requests before it reached the runner, but this was not working correctly. The fix moves the population logic into insert_and_send_commands activities and the keepalive task so requests are correctly carried across restarts.

Issues

Correctness

1. Duplicate UDB fetch in actor2/runtime.rs (double population)

In actor2/runtime.rs, send_outbound fetches hibernating requests and sets them on the CommandStartActor before calling InsertAndSendCommands::run. Then insert_and_send_commands also fetches and overwrites hibernating_requests for every CommandStartActor. The second fetch overwrites the first, making the fetch in send_outbound redundant. This adds unnecessary UDB load. One of the two sites should be removed.

2. mk1 runner path is not covered

actor/runtime.rs has mk1 (non-mk2) runner code paths (e.g. the ctx.signal(crate::workflows::runner::Command { ... }) calls) that still send hibernating_requests: Vec::new() and do not go through insert_and_send_commands. If mk1 runners support hibernation, this is a bug regression. If they don't (or are deprecated for this use case), a comment explaining why would help.

3. Non-deterministic read inside a Gasoline activity

insert_and_send_commands now performs a live UDB read (hibernating_request::list) inside the activity before writing. Gasoline activities can be retried, and the read result could differ between attempts (e.g., a keepalive expired between retries). This could cause non-deterministic replay behavior. Worth evaluating whether this read should be moved outside the activity boundary or made part of the activity input instead.

Code Quality

4. Stale/misleading comments

Multiple call sites in actor/runtime.rs and actor2/runtime.rs that construct CommandStartActor with hibernating_requests: Vec::new() still carry the old comment:

// Empty because request ids are ephemeral. This is intercepted by guard and populated before it reaches the runner

Since the interception now happens inside insert_and_send_commands (not the guard), these comments are misleading and should be updated to reflect the new behavior.

5. tracing::info! in high-frequency hot paths

The new log lines in upsert.rs, delete.rs, and list.rs use tracing::info!. The keepalive loop calls upsert on every ping interval tick, so on systems with many hibernating actors this will produce high-volume info-level output. tracing::debug! is more appropriate for the per-tick upsert and list operations (consistent with the existing tracing::debug! calls in the keepalive loop body). tracing::info! is fine for delete since that's less frequent.

Minor

6. Unrelated change in envoy-client/connection.rs

The addition of "rivet_target.envoy" as a WebSocket subprotocol appears unrelated to the hibernating request persistence fix. It should either be explained in the PR description or split into a separate commit.

7. No tests added

The test checklist item is unchecked. Given this is a distributed systems bug fix involving actor restart and request replay, integration tests would add significant confidence. Even a basic test that hibernates a WebSocket connection, restarts the actor, and verifies the request is replayed would be valuable.

Suggestions

• Remove the redundant hibernating_requests fetch from send_outbound in actor2/runtime.rs (issue 1).
• Update or remove the stale // intercepted by guard comments at all CommandStartActor construction sites (issue 4).
• Downgrade the new tracing::info! calls in upsert.rs and list.rs to tracing::debug! (issue 5).
• Add a comment on the mk1 paths explaining whether they need this fix or why they don't (issue 2).
• Consider the Gasoline replay semantics for the UDB read inside the activity (issue 3).