• feat(rivetkit): allow configuring runtime mode via env #4866
• fix(gateway): fix protocol handling for safari #4865 👈 (View in Graphite)
• fix(rivetkit): standardize startEngine + runtime mode based on config #4851
• fix(engine): remove envoy sqlite lifecycle bookkeeping #4850
• test(rivetkit): assert force sleep websocket close reason #4849
• test(sqlite): expand db stress coverage #4846 : 1 other dependent PR (#4847 )
• test(raw-websocket): add serverless smoke coverage #4845
• feat(serverless): configure drain grace period #4844
• fix(publish): enable frontend build for engine target and use turbo build instead of turbo build:engine #4843
• feat(rivetkit): expose conn handle in onWebSocket context #4842
• fix(rivetkit): normalize waitUntil/keepAwake promises to null to avoid serde undefined error #4841
• feat(pegboard-envoy): track and close serverless sqlite actors on stop/shutdown #4840
• fix(rivetkit): remove requirement for token #4839
• fix(pegboard-envoy): allow sqlite requests across lifecycle states #4838
• fix(pegboard): route scoped actor key reads by runner dc #4837
• fix(sqlite-storage): let sqlite open take over stale owners #4834
• fix(engine-tests): parallelize multi-DC startup + enable 6 multi-DC tests #4833
• fix(api): honor cursor in actor_ids list path #4832
• fix(api): silently filter invalid IDs from actor_ids query instead of 400 #4831
• fix(api-peer): make actors_delete idempotent for already-destroyed actors #4830
• fix(pegboard): use Iterator streaming mode in list_names so limit param is honored #4829
• chore(engine-tests): convert envoy tests to Sleep crash policy + triage ignored #4828
• feat(engine-tests): wire envoy actor names through prepopulate_actor_names #4827
• main

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

PR Review: fix(gateway): fix protocol handling for safari

Summary

This PR fixes a Sec-WebSocket-Protocol negotiation bug that caused WebSocket connections to fail in Safari. The root cause was the proxy unconditionally echoing "rivet" as the selected protocol regardless of what the client offered. The fix introduces select_websocket_response_protocol() which properly reads the client's offered protocols and picks the most appropriate one, and relaxes path-based routing to not error on a missing protocol header.

Findings

1. Incomplete data-carrying protocol filtering in the fallback (Low severity)

select_websocket_response_protocol filters rivet_conn_params.* and rivet_token.* in the fallback branch, but the client also sends other data-carrying protocols:

• rivet_encoding.* (e.g. rivet_encoding.cbor)
• rivet_target.*
• rivet_actor.*
• rivet_test_ack_hook.*

If "rivet" is absent from the offered list (which is the scenario this fix targets), the fallback could echo back rivet_encoding.cbor or rivet_actor.<uuid> as the selected protocol, which would confuse clients that parse the response header.

In practice, the client always sends "rivet" first so the fallback likely never fires for normal Rivet clients. But for browsers that reorder protocols (precisely the Safari case), this edge case is live. A simpler and more robust filter would be:

// Exclude all data-carrying protocol tokens; only echo plain negotiation identifiers
!protocol.starts_with("rivet_")

2. Duplicate SEC_WEBSOCKET_PROTOCOL constant (Minor)

guard/src/routing/mod.rs already defines SEC_WEBSOCKET_PROTOCOL. The PR adds a separate private definition in guard-core/src/proxy_service.rs. These are in different crates so there's no conflict, but it's worth considering whether guard-core should re-export or reference a shared location if these crates are expected to share protocol constants.

3. No tests added (Minor)

Given that the bug is browser-specific and involves protocol ordering behavior, a unit test for select_websocket_response_protocol would be valuable for regressions:

• Safari-style ordering (data-carrying protocols appear before "rivet")
• Missing "rivet" in the offered list
• Empty protocol header → None
• All protocols are data-carrying → None

What's correct

• The pegboard_gateway change is correct: path-based routing extracts the token from the URL path, so a missing Sec-WebSocket-Protocol header is legitimately optional there. Returning Ok(None) rather than an error is the right call.
• The fix is applied symmetrically to both WebSocket response paths in proxy_service.rs (the error-proxy path and the normal path).
• Excluding rivet_token.* and rivet_conn_params.* from the echoed response header is correct — those values should never be reflected back to the client.
• Using named constants instead of raw string literals is good practice.

Summary

The fix is sound and directly addresses the Safari compatibility issue. The main concern worth a follow-up is the incomplete filtering of data-carrying protocol prefixes (rivet_encoding.*, rivet_target.*, rivet_actor.*) in the fallback branch — extending the filter to !protocol.starts_with("rivet_") would close that gap more completely.

Preview packages published to npm

Install with:

npm install rivetkit@pr-4865

All packages published as 0.0.0-pr.4865.b488635 with tag pr-4865.

Engine binary is shipped via @rivetkit/engine-cli on linux-x64-musl, linux-arm64-musl, darwin-x64, and darwin-arm64. Windows users should use the release installer or set RIVET_ENGINE_BINARY.

Docker images:

docker pull rivetdev/engine:slim-b488635
docker pull rivetdev/engine:full-b488635

npm install rivetkit@pr-4865
npm install @rivetkit/react@pr-4865
npm install @rivetkit/rivetkit-napi@pr-4865
npm install @rivetkit/workflow-engine@pr-4865