Errors and Observability
Failure policies, backpressure metrics, and the dashboard.
Process errors
Use Process when a transform can fail.
stream.Process(fn,
operator.WithProcessFailurePolicy(operator.ProcFailureDLQ),
operator.WithProcessDLQ(dlqSink),
)Use Fail policies for data that must not be skipped. Use DLQ policies when
bad records should be retained for later inspection.
Sink errors
Kafka and Postgres sinks support retry and failure policy options. After retries are exhausted, the configured policy decides whether the record is dropped, sent to DLQ, or fails the pipeline.
Backpressure
Weibo uses bounded channels between execution stages. If a sink is slow, its input edge fills. The previous stage blocks while sending, and that pressure propagates back to the source. This bounds memory without dropping records.
Tune the edge capacity with:
env := weibo.NewEnv().WithBufferSize(2048)Metrics
Weibo records Prometheus metrics for pipeline, stages, edges, operators, and workers.
Important metrics:
| Metric | Meaning |
|---|---|
weibo_edge_queue_size | Current buffered records on an edge. |
weibo_edge_queue_capacity | Edge capacity. |
weibo_stage_records_in_total | Records entering a stage. |
weibo_stage_records_out_total | Records leaving a stage. |
weibo_stage_send_block_seconds_total | Time blocked on downstream backpressure. |
weibo_stage_workers | Active worker count for a stage. |
weibo_stage_errors_total | Stage errors. |
Dashboard
The dashboard package exposes pipeline structure and metrics for live inspection. Labels passed to SDK operators are used in dashboard metadata:
stream.Filter(isCompleted, "completed-orders")