Monitor Data Connectivity Proxy¶

DCP provides observability through SQL interfaces and a Prometheus metrics endpoint on the agent.

Describe a DCP client¶

Use DESCRIBE DATA CONNECTIVITY PROXY to view the current state of a DCP client. The command returns one row. You need USAGE on the object, which is the same privilege DESCRIBE requires. If the client name needs quoting, quote it in the statement:

DESCRIBE DATA CONNECTIVITY PROXY my_dcp_client;

DESCRIBE DATA CONNECTIVITY PROXY "quoted_client_name";

AGENT_HEALTH is a derived summary of the reported statuses. It isn’t a stored field of its own. Values are HEALTHY, DEGRADED, or DOWN:

• DOWN: AGENT_STATUS isn’t DCP_AGENT_LIFECYCLE_CONNECTED, including a client that has never reported a status.
• HEALTHY: the client is connected, bootstrap succeeded, certificate rotation isn’t failed, and the data path isn’t failed or in progress.
• DEGRADED: the client is connected, but bootstrap isn’t confirmed success, or certificate rotation or the data path isn’t well.

An unreported data path doesn’t count as unhealthy. A certificate rotation in progress doesn’t demote health, because traffic still uses the current certificate. LAST_HEARTBEAT_AT is the last heartbeat timestamp; heartbeat is already folded into AGENT_HEALTH.

For current state, use DESCRIBE. Event history records when a status was entered, not what is true now.

The output includes the following columns:

View event history¶

INFORMATION_SCHEMA.DCP_CLIENT_EVENT_HISTORY returns one row per recorded status transition for a DCP client. AGENT_ID is required and is the DCP client name (the same identifier you use in DESCRIBE). You need USAGE on that client. If you don’t have it, Snowflake treats the object as not existing.

A row means the client entered that STATUS at OCCURRED_AT. It isn’t the current state. Use DESCRIBE for current state. Authentication failures appear on DESCRIBE (LAST_AUTH_FAILURE_REASON and LAST_AUTH_FAILURE_AT); they aren’t a published event type.

SELECT *
FROM TABLE(INFORMATION_SCHEMA.DCP_CLIENT_EVENT_HISTORY(
  AGENT_ID => 'my_dcp_client'
))
ORDER BY occurred_at DESC;

Columns: AGENT_ID, OCCURRED_AT, EVENT_TYPE, STATUS, REASON, RECORDED_AT.

EVENT_TYPE values:

STATUS values: SUCCESS, FAILURE, PENDING, IN_PROGRESS. PENDING and IN_PROGRESS mean the operation hadn’t finished when the row was recorded. A later row carries the outcome.

REASON is a DCP error code on failure (for example, DCP_BOOTSTRAP_TOKEN_INVALID or DCP_BOOTSTRAP_NETWORK_POLICY_BLOCKED), or NULL.

Check route reachability¶

INFORMATION_SCHEMA.DCP_CLIENT_ROUTE_CHECK is a live diagnostic. It returns one row per stage of the route from Snowflake through the named DCP client to a destination. Both arguments are required constants. AGENT_ID is the DCP client name. DESTINATION is host:port. You need USAGE on the client.

SELECT *
FROM TABLE(INFORMATION_SCHEMA.DCP_CLIENT_ROUTE_CHECK(
  AGENT_ID => 'my_dcp_client',
  DESTINATION => 'my-private-db.internal:5432'
))
ORDER BY stage_order;

Columns: AGENT_ID, STAGE_ORDER, STAGE, STATUS, DETAIL, REASON, CHECKED_AT.

Stages, in order: EGRESS_CONFIG, POLICY, BOOTSTRAP, CONTROL_PLANE, CERTIFICATES, DATA_PATH, SNOWFLAKE_INTERNAL, DESTINATION.

STATUS values: SUCCESS, FAILURE, IN_PROGRESS, SKIPPED, UNKNOWN.

• SKIPPED: an earlier stage didn’t resolve, so this stage wasn’t attempted. DETAIL names the gate to fix first.
• IN_PROGRESS: a caller action can still produce an answer. DETAIL names that action.
• UNKNOWN: nothing you change produces an answer for this stage.

Some stages use client telemetry. Those rows are empty or UNKNOWN until telemetry is reachable for the account.

Use this when a connector can’t reach a private source.

View connection history¶

INFORMATION_SCHEMA.DCP_CLIENT_CONNECTION_HISTORY returns one row per closed connection for a DCP client. Rows come from agent telemetry spans (dcp.connection) in the account event table. You need USAGE on the client. AGENT_ID is required (the DCP client name). TIME_RANGE_HOURS is optional: default 24, minimum 1, maximum 2160 (90 days). The window is on recorded time, not start time, so STARTED_AT can fall before the window.

SELECT *
FROM TABLE(INFORMATION_SCHEMA.DCP_CLIENT_CONNECTION_HISTORY(
  AGENT_ID => 'my_dcp_client',
  TIME_RANGE_HOURS => 24
))
ORDER BY recorded_at DESC;

Columns: AGENT_ID, CONNECTION_ID, WORKLOAD_ID, DESTINATION_HOST, DESTINATION_PORT, STATUS, REASON, STARTED_AT, ENDED_AT, RECORDED_AT, SETUP_DURATION_MS, BYTES_TO_DESTINATION, BYTES_FROM_DESTINATION.

STATUS values: SUCCESS, FAILURE, UNKNOWN. UNKNOWN covers a missing reason and a draining agent (shutting_down): neither settled the connection.

WORKLOAD_ID, destination host and port, and SETUP_DURATION_MS can be NULL when the connection was rejected before those values exist. Byte counts are always present (0 if the tunnel never carried traffic). REASON is NULL on a clean close.

Prometheus metrics (agent-side)¶

The agent exposes a Prometheus metrics endpoint at 127.0.0.1:9092 by default. Scrape this endpoint with any Prometheus-compatible monitoring stack, such as Prometheus, Datadog, or Grafana Agent.

Key metrics:

Change the metrics endpoint address¶

To listen on a different address or port:

docker run ... snowflake/dcp-agent:latest \
  ... \
  --metrics-addr 0.0.0.0:9092

Health check¶

Run the agent health check to validate connectivity without starting the data-plane workers:

docker run --rm \
  -v /etc/dcp/credentials:/credentials:ro \
  snowflake/dcp-agent:latest \
  --health-check \
  --dcp-client-name my_dcp_client \
  --credentials-file /credentials \
  --snowflake-account <your-account-identifier>

The health check runs the full bootstrap sequence (verifying that the control plane is reachable, the token is valid, the portal is reachable, and the proxy is reachable) and prints a structured pass/fail result for each step.

A local /ready HTTP endpoint is also available on the agent at 127.0.0.1:8086/ready for container orchestration health probes.