Hi all,
We’re integrating RTMS with the Video SDK for transcript-only consumption, running our NodeJS backend as multiple pods behind an ALB. We have a requirement to never lose transcript data across reconnects or deploys, and ran into several gaps in the docs that would help a lot to get clarified. Grouped into three areas:
- Reconnection logic, what the SDK handles vs. what we need to implement
The failover and reconnection doc walks through 4 scenarios, but the sample code is all raw WebSocket handling. It’s not clear how much of this the@zoom/rtmsSDK (Client()/join()/leave()) handles internally versus what we’re expected to implement ourselves.
-
For each of the 4 scenarios, does the SDK do any of this automatically, or is it all application responsibility?
-
For Scenario 3 (media-only interruption), with signaling still alive: is re-issuing
client.join()with the same payload the correct recovery? Could callingjoin()again while already joined cause a conflicting handshake? Is there a more targeted method intended for reconnecting just the media leg? -
Is there an official SDK-based reconnection example planned? We noticed issue #116 is tracking exactly this gap for
examples/node.md/python.md. Would appreciate a timeline if there is one. -
Does any buffering/replay exist for data generated during a mid-session interruption (as distinct from the documented
buffer_datawindow before the first connection is established)? Specifically for transcript data, or is anything generated during that gap unrecoverable by design?
- What each SDK callback actually means relative to the WS lifecycle
-
onJoinConfirm: does this fire only once both signaling and media sockets are established and ready to carry data, or on signaling success alone while media is still connecting? -
onLeave: does this fire only on signaling-connection loss (full session teardown), or can it also fire for a media-only drop? If both, how should it be distinguished fromonMediaConnectionInterrupted? -
Can
onLeavefire beforeonJoinConfirmever fired (e.g. a join that failed outright), and if so does it mean something different there vs. a mid-session leave? -
Is there a documented state diagram anywhere mapping join, signaling up, media up, data flowing, interrupted, reconnecting, left to the specific callbacks fired at each transition? This would resolve most of the ambiguity above in one shot if it exists or could be published.
- Running multiple pods behind an ALB
-
Is there an official recommended architecture for a horizontally-scaled backend where the webhook endpoint is a single ALB target group across many pods, and a webhook for an existing stream can land on a different pod than the one holding the original connection?
-
For Scenario 1 (a repeat rtms_started for an already-active stream), our plan is to have the receiving pod open both new WS connections without first asking the original pod to disconnect, on the assumption that Zoom has already closed that original connection on its end. Is that the recommended approach, or should we be explicitly tearing down the old connection first?
-
For Scenario 2 (rtms_interrupted), our plan is the same: the receiving pod opens both WS connections directly, without coordinating with whichever pod held the original connection. Is that correct, or is there a reason we should sequence this differently?
Thanks!