Robinhood feed resume

Resume the Robinhood sequencer feed after a disconnect: the from_sequence cursor, the resume control frame, what bufferComplete means and how to reconcile a gap it reports.

The feed keeps a bounded in-memory replay window, so a client that reconnects quickly can continue where it stopped instead of starting blind. This page covers the cursor protocol; see Sequencer feed for the connection, frame shape and delivery model.

Reconnect with a cursor

wss://triport.io/ws/robinhood-feed?from_sequence=56798024

Pass the last fully processed sequence number — replay starts strictly after it. The value must be an unsigned decimal integer; anything else is rejected before the WebSocket upgrade:

HTTP 400  from_sequence must be an unsigned decimal integer

Persist the cursor before you acknowledge downstream work. If the cursor is lost, connect without the parameter and backfill the missing range through JSON-RPC.

The resume control frame

The first frame after a resume request is a control frame, not data:

{"version":1,
 "resume":{"requestedAfterSequence":56798024,"earliestAvailableSequence":56797001,"latestAvailableSequence":56798130,"bufferComplete":true},
 "reconciliation":{"status":"confirmed","latestConfirmedSequence":56798120,"frameLastSequence":56798130,"confirmationLag":10,"sourceGapsObserved":0}}

Replayed data frames carry replay: true and contain only sequence numbers greater than the cursor. The reconciliation object is described in Reconciliation.

When bufferComplete is false

bufferComplete is true only if the window is non-empty, your cursor is at or below latestAvailableSequence, and every sequence from requestedAfterSequence + 1 up to the newest one is present. It is therefore false when:

• the cursor is too old — the oldest retained sequence is already past it;
• the cursor is ahead of the window (for example after a restart on your side that stored a future value);
• the window is empty, which is the normal state right after a service restart.

In that case consume the available tail and reconcile the missing range yourself: read the corresponding blocks over JSON-RPC (eth_getBlockByNumber, eth_getBlockReceipts, eth_getLogs), or replay from your own durable store. sequenceNumber is the L2 block number, which makes that mapping direct.

Window limits

At most 8,192 frames, 64 MiB or five minutes — whichever limit is reached first. The window lives in process memory and is lost on process restart, so the five minutes are a ceiling, not a promise. It is an operational buffer for short interruptions, not a historical API: there is no replay of an arbitrary past day.

If your client cannot keep up

A client whose outbound queue cannot drain is disconnected with an explicit reason:

client buffer is full; reconnect with from_sequence to resume buffered data

Treat it as a signal to consume faster (or to do less work inline), then reconnect with your stored cursor. Stream concurrency per key is plan policy — 2 streams on Pro, 4 on Business and Enterprise; see Limits.