Monitoring the Connection Manager

Monitoring the Connection Manager#

You can view the status of the Connection Manager and its connections through SQL queries and HTTP endpoints.

Available SQL tables and views#

The Connection Manager provides a number of tables and views that can be used to monitor the status of the Connection Manager and its connections. These include:

  • bdr.stat_activity - which is information from pg_stat_activity enhanced with addition columns regarding the connection_manager_client_addr and connection_manager_client_port is the connection has come through the connection manager, and session_read_only if it has connected through the read-only port.

  • bdr.stat_connection_manager - which is a view that provides statistics about the Connection Manager’s status.

  • bdr.stat_connection_manager_connections - which is a view that provides statistics about the Connection Manager’s connections.

  • bdr.stat_connection_manager_node_stats - which is a view that provides statistics about the Connection Manager on each of the data nodes.

  • bdr.stat_connection_manager_hba_file_rules - which is a view that shows which HBA file rules for the connection manager are being used on this node.

Available HTTP/HTTPS endpoints#

The Connection Manager can be monitored through the HTTP API.

Endpoints returning true/false will also return a 200 status code for true and a 503 status code for false.

The following endpoints are available:

Endpoint

Description

/connection/is-live

Is the connection manager live (listening), always returns “true”, if the manager is not running, the client will simply fail to open the connection/url.

/connection/is-ready

Is the connection manager ready to accept connections, returns true(200)/false(503).

/connection/is-ready-ro

Is the connection manager ready to accept read-only connections, returns true(200)/false(503).

/connection/is-ready-rw

Is the connection manager ready to accept read-write connections, returns true(200)/false(503).

/node/is-read-write

Is this PGD node, not the connection manager but the PGD node itself, a read-write node (is it write leader), returns true(200)/false(503).

/node/is-read-only

Is this PGD node, not the connection manager but the PGD node itself, a read-only node (not the write leader), returns true(200)/false(503).

/group/read-write-info

Returns information about the read-write pool on this instance of connection manager - a list of nodes in the pool in JSON format with node id, node name, node host, node port and node dbname. For the read-write pool, the pool only contains one entry.

/group/read-only-info

Returns information about the read-only pool on this instance of connection manager - a list of nodes in the pool in JSON format with node id, node name, node host, node port and node dbname.

Below is an example of a response body from the /group/read-write-info endpoint:

[
  {
    "id": 683485707,
    "name": "node-1",
    "host": "host-1",
    "port": 5432,
    "dbname": "pgddb"
  }
]

Logging#

All Connection Manager log messages are written to the PostgreSQL log.

The behavior of %r and %h escape sequences in log_line_prefix has been altered to log “proxy_address/client_address” and “proxy_port/client_port” respectively.

This is achieved by the proxy setting a GUC for the server connections it uses. As users can override this GUC, any security context derived from the client_address will need to be verified by referring to the full session logs.

Contents