CTDB metrics

The following section lists all the CTDB metrics::

  • CTDB version: Version of the CTDB protocol used by the node.
  • Current time of statistics: Time when the statistics are generated. This is useful when collecting statistics output periodically for post-processing.
  • Statistics collected since: Time when CTDB was started or the last time statistics was reset. The output shows the duration and the timestamp.
  • num_clients: Number of processes currently connected to CTDB's UNIX socket. This includes recovery daemon, CTDB tool and SMB processes (smbd, winbindd).
  • frozen: 1 if the databases are currently frozen, 0 if otherwise.
  • recovering: 1 if recovery is active, 0 if otherwise.
  • num_recoveries: Number of recoveries since the start of CTDB or since the last statistics reset.
  • client_packets_sent: Number of packets sent to client processes via UNIX domain socket.
  • client_packets_recv: Number of packets received from client processes via UNIX domain socket.
  • node_packets_sent: Number of packets sent to the other nodes in the cluster via TCP.
  • node_packets_recv: Number of packets received from the other nodes in the cluster via TCP.
  • keepalive_packets_sent: Number of keepalive messages sent to other nodes. CTDB periodically sends keepalive messages to other nodes. For more information, see the KeepAliveInterval tunable in CTDB-tunables(7) on the CTDB documentation website.
  • keepalive_packets_recv: Number of keepalive messages received from other nodes.
  • node: This section lists various types of messages processed which originated from other nodes via TCP.
    • req_call: Number of REQ_CALL messages from the other nodes.
    • reply_call: Number of REPLY_CALL messages from the other nodes.
    • req_dmaster: Number of REQ_DMASTER messages from the other nodes.
    • reply_dmaster: Number of REPLY_DMASTER messages from the other nodes.
    • reply_error: Number of REPLY_ERROR messages from the other nodes.
    • req_message: Number of REQ_MESSAGE messages from the other nodes.
    • req_control: Number of REQ_CONTROL messages from the other nodes.
    • reply_control: Number of REPLY_CONTROL messages from the other nodes.
  • client: This section lists various types of messages processed which originated from clients via UNIX domain socket.
    • req_call: Number of REQ_CALL messages from the clients.
    • req_message: Number of REQ_MESSAGE messages from the clients.
    • req_control: Number of REQ_CONTROL messages from the clients.
  • timeouts: This section lists timeouts occurred when sending various messages.
    • call: Number of timeouts for REQ_CALL messages.
    • control: Number of timeouts for REQ_CONTROL messages.
    • traverse: Number of timeouts for database traverse operations.
  • locks: This section lists locking statistics.
    • num_calls: Number of completed lock calls. This includes database locks and record locks.
    • num_current: Number of scheduled lock calls. This includes database locks and record locks.
    • num_pending: Number of queued lock calls. This includes database locks and record locks.
    • num_failed: Number of failed lock calls. This includes database locks and record locks.
  • total_calls: Number of req_call messages processed from clients. This number should be same as client --> req_call.
  • pending_calls: Number of req_call messages which are currently being processed. This number indicates the number of record migrations in flight.
  • childwrite_calls: Number of record update calls. Record update calls are used to update a record under a transaction.
  • pending_childwrite_calls: Number of record update calls currently active.
  • memory_used: The amount of memory in bytes currently used by CTDB using talloc. This includes all the memory used for CTDB´s internal data structures. This does not include the memory mapped TDB databases.
  • max_hop_count: The maximum number of hops required for a record migration request to obtain the record. High numbers indicate record contention.
  • total_ro_delegations: Number of read-only delegations created.
  • total_ro_revokes: Number of read-only delegations that were revoked. The difference between total_ro_revokes and total_ro_delegations gives the number of currently active read-only delegations.
  • hop_count_buckets: Distribution of migration requests based on hop counts values.
  • lock_buckets: Distribution of record lock requests based on time required to obtain locks. Buckets are < 1ms, < 10ms, < 100ms, < 1s, < 2s, < 4s, < 8s, < 16s, < 32s, < 64s, > 64s.
  • locks_latency: The minimum, the average and the maximum time (in seconds) required to obtain record locks.
  • reclock_ctdbd: The minimum, the average and the maximum time (in seconds) required to check if recovery lock is still held by recovery daemon when recovery mode is changed. This check is done in ctdb daemon.
  • reclock_recd: The minimum, the average and the maximum time (in seconds) required to check if recovery lock is still held by recovery daemon during recovery. This check is done in recovery daemon.
  • call_latency: The minimum, the average and the maximum time (in seconds) required to process a REQ_CALL message from client. This includes the time required to migrate a record from remote node, if the record is not available on the local node.
  • childwrite_latency: The minimum, the average and the maximum time (in seconds) required to update records under a transaction.
Parent topic: Protocol metrics