21.6.15.4 The ndbinfo cluster_locks Table

The cluster_locks table provides information about current lock requests holding and waiting for locks on NDB tables in an NDB Cluster, and is intended as a companion table to cluster_operations. Information obtain from the cluster_locks table may be useful in investigating stalls and deadlocks.

The cluster_locks table contains the following columns:

• node_id

  ID of reporting node

• block_instance

  ID of reporting LDM instance

• tableid

  ID of table containing this row

• fragmentid

  ID of fragment containing locked row

• rowid

  ID of locked row

• transid

  Transaction ID

• mode

  Lock request mode

• state

  Lock state

• detail

  Whether this is first holding lock in row lock queue

• op

  Operation type

• duration_millis

  Milliseconds spent waiting or holding lock

• lock_num

  ID of lock object

• waiting_for

  Waiting for lock with this ID

Notes

The table ID (tableid column) is assigned internally, and is the same as that used in other ndbinfo tables. It is also shown in the output of ndb_show_tables.

The transaction ID (transid column) is the identifier generated by the NDB API for the transaction requesting or holding the current lock.

The mode column shows the lock mode; this is always one of S (indicating a shared lock) or X (an exclusive lock). If a transaction holds an exclusive lock on a given row, all other locks on that row have the same transaction ID.

The state column shows the lock state. Its value is always one of H (holding) or W (waiting). A waiting lock request waits for a lock held by a different transaction.

When the detail column contains a * (asterisk character), this means that this lock is the first holding lock in the affected row's lock queue; otherwise, this column is empty. This information can be used to help identify the unique entries in a list of lock requests.

The op column shows the type of operation requesting the lock. This is always one of the values READ, INSERT, UPDATE, DELETE, SCAN, or REFRESH.

The duration_millis column shows the number of milliseconds for which this lock request has been waiting or holding the lock. This is reset to 0 when a lock is granted for a waiting request.

The lock ID (lockid column) is unique to this node and block instance.

The lock state is shown in the lock_state column; if this is W, the lock is waiting to be granted, and the waiting_for column shows the lock ID of the lock object this request is waiting for. Otherwise, the waiting_for column is empty. waiting_for can refer only to locks on the same row, as identified by node_id, block_instance, tableid, fragmentid, and rowid.

The cluster_locks table was added in NDB 7.5.3.