decentraland / comms-message-sfu / 25503714461

90%

push

github

fix: guard publishData against non-connected Room (#113)

* fix: guard publishData against non-connected Room

A recent dev SIGSEGV happened inside @livekit/rtc-node native code
during the SDK's internal reconnect. Independently, there is a
latent race in routeMessage where an in-flight routing call can
outlive a concurrent disconnect and end up invoking publishData on
a Room whose native state is being torn down — which can crash the
process.

Add a ConnectionState check before each publishData so we bail
cleanly (logged + failed metric) instead of touching torn-down
native handles.

Also bump @livekit/rtc-node ^0.13.14 -> ^0.13.27, which brings in
upstream fixes around FFI event ordering, waiting for the disconnect
response, and native handle disposal — all in the same crash class.

Switch @dcl/protocol to the experimental dist-tag so we keep
tracking the experimental branch where Chat.forwarded_from lives;
main dropped that field.

* chore: bail before DB lookups when Room is not connected; dedupe state tests

Address review comments:

- Add an early ConnectionState check before db.belongsToCommunity /
  db.getCommunityMembers so a routing call against an already-disconnected
  Room avoids two unnecessary queries before failing.
- Collapse the CONN_DISCONNECTED / CONN_RECONNECTING test pair into a
  single it.each, matching the convention in disconnected-handler.spec.ts.
- Add a test that exercises the per-batch guard specifically — connection
  state flips during the members lookup, so the early check passes but
  the publish-time guard still bails before publishData.

65 of 78 branches covered (83.33%)

Branch coverage included in aggregate %.

5 of 5 new or added lines in 1 file covered. (100.0%)

261 of 286 relevant lines covered (91.26%)

8.99 hits per line