Identify the materialized view or table whose output you want to deliver downstream.
Run CREATE SINK <sink_name> FROM <view_name> WITH (connector='kafka', ...) FORMAT DEBEZIUM ENCODE JSON (or the appropriate format for the target).
For non-Kafka sinks (e.g., Postgres, Iceberg, S3, ClickHouse), adjust the connector type and required WITH properties accordingly.
For Iceberg sinks, specify the catalog type (REST, Glue, HMS) and table path in the WITH clause.
Monitor sink health via the rw_sinks system table and check for any delivery lag or error states.
Drop and recreate the sink if the upstream view schema changes in a backward-incompatible way.
Known gotchas
Sink semantics depend on the chosen format: DEBEZIUM sinks emit diff rows, UPSERT sinks require a key, and APPEND-ONLY sinks do not support deletions — choose based on what the downstream system supports.
Some sink connectors (e.g., Iceberg) require additional infrastructure (a catalog service) to be running and reachable from RisingWave.
Connector property names and required fields differ across sink types and RisingWave versions; consult the version-matched documentation.
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp