Metrics

Sequin exposes a metrics endpoint, using Prometheus text format, which can be integrated into your existing observability workflows. By default, this endpoint is at /metrics on port 8376.

Metrics are designed to answer:

• Is Sequin keeping up with the database? (Ingestion metrics)
• Is Sequin processing messages quickly? (Internal processing metrics)
• Are sinks keeping up with Sequin? (Delivery metrics)
• Is the Sequin service experiencing CPU or memory pressure? (VM metrics)

​

Metrics reference

​

Ingestion metrics

• sequin_ingestion_latency_us (Gauge)

  Measures latency between transaction commit and when Sequin first receives it. Calculated by instrumenting Sequin’s own heartbeat messages.

• sequin_messages_ingested_count (Counter)

  Number of messages Sequin has received from a given replication slot.

​

Internal processing metrics

• sequin_internal_latency_us (Histogram)

  Measures time between Sequin ingesting a message and beginning to process it.

​

Delivery metrics

• sequin_delivery_latency_us (Histogram)

  Measures time between Sequin beginning to deliver a message and a given sink’s successful acknowledgement. Some sinks are able to batch messages, in which case this metric measures the batch latency, rather than the per-message latency.

• sequin_message_deliver_attempt_count (Counter)

  Incremented for each message Sequin tries to deliver, before receiving a success or failure response.

• sequin_message_deliver_success_count (Counter)

  Incremented for each message that Sequin successfully delivers.

• sequin_message_deliver_failure_count (Counter)

  Incremented for each message that Sequin fails to deliver to a sink. This is because the sink was either unreachable, timed out, or returned an error.

​

Total latency

​

VM metrics

Sequin is implemented with Elixir. Elixir’s VM (the BEAM) exposes many metrics. The following are the ones we find most useful for monitoring Sequin:

• erlang_vm_statistics_reductions_total (Counter)

  Measures how many VM instructions are executed across all processes. Shows intensity of CPU-bound work. Recommend using with a rate query rather than directly.

• erlang_vm_memory_bytes_total (Gauge)

  Memory usage of the entire VM.

​

Grafana dashboard

Sequin provides a Grafana dashboard template for monitoring Sequin. This can be imported directly into your Grafana instance for use, or referenced as a guide for building your own.