Add a NebulaGraph vertex sink connector
Implemented SeaTunnel's first NebulaGraph connector slice, enabling batch-oriented vertex writes to existing spaces and tags through the official Java SessionPool client.
apache/seatunnel · #11865
New graph-data connector
SeaTunnel can now write batched vertex inserts and updates to NebulaGraph with validated configuration, parameterized properties, and at-least-once delivery.
Problem
SeaTunnel had no supported path for writing row data into NebulaGraph. A production connector needed to construct safe nGQL, validate graph and schema inputs, batch records, flush at checkpoints, and coordinate a multi-service NebulaGraph environment reliably in E2E tests.
Approach
Added INSERT and UPDATE vertex modes, parameterized property values, escaped string vertex IDs, configuration and type validation, checkpoint-aware batching, registration and bilingual documentation. Reworked the E2E startup sequence so storage readiness is checked only after host registration removes the service dependency cycle.
Impact and scope
- Extends SeaTunnel into graph-data pipelines with an additive, production-oriented NebulaGraph sink.
- Protects generated statements through validated identifiers, parameterized values, and explicit handling of unsupported changelog and data types.
- Provides clear at-least-once behavior with batch and checkpoint flushing for both insert and update workloads.
Validation
- Added unit coverage for configuration, host parsing, statement construction, vertex ID escaping, batching, checkpoints, changelog handling, unsupported types, and failed-write cleanup.
- Added NebulaGraph 3.8 E2E coverage for insert and update paths; the final Apache Build completed successfully after the startup-order correction.
- All authored changes and the final merge commit are attributable to Goutam Adwant; the GitHub-verified merge received two upstream approvals.