22.5.10.29 The ndbinfo operations_per_fragment Table

The operations_per_fragment table provides information about the operations performed on individual fragments and fragment replicas, as well as about some of the results from these operations.

The following table provides information about the columns in the operations_per_fragment table. For each column, the table shows the name, data type, and a brief description. Additional information can be found in the notes following the table.

The fq_name contains the fully qualified name of the schema object to which this fragment replica belongs. This currently has the following formats:

The $unique suffix shown for unique indexes is added by mysqld; for an index created by a different NDB API client application, this may differ, or not be present.

The syntax just shown for fully qualified object names is an internal interface which is subject to change in future releases.

Consider a table t1 created and modified by the following SQL statements:

If t1 is assigned table ID 11, this yields the fq_name values shown here:

For indexes or BLOB tables, the parent_fq_name column contains the fq_name of the corresponding base table. For base tables, this column is always NULL.

The type column shows the schema object type used for this fragment, which can take any one of the values System table, User table, Unique hash index, or Ordered index. BLOB tables are shown as User table.

The table_id column value is unique at any given time, but can be reused if the corresponding object has been deleted. The same ID can be seen using the ndb_show_tables utility.

The block_instance column shows which LDM instance this fragment replica belongs to. You can use this to obtain information about specific threads from the threadblocks table. The first such instance is always numbered 0.

Since there are typically two replicas, and assuming that this is so, each fragment_num value should appear twice in the table, on two different data nodes from the same node group.

Since NDB does not use single-key access for ordered indexes, the counts for tot_key_reads, tot_key_inserts, tot_key_updates, tot_key_writes, and tot_key_deletes are not incremented by ordered index operations.

The tot_key_refs column shows the number of key operations refused by the LDM. Generally, such a refusal is due to duplicate keys (inserts), Key not found errors (updates, deletes, and reads), or the operation was rejected by an interpreted program used as a predicate on the row matching the key.

The attrinfo and keyinfo attributes counted by the tot_key_attrinfo_bytes and tot_key_keyinfo_bytes columns are attributes of an LQHKEYREQ signal (see The NDB Communication Protocol) used to initiate a key operation by the LDM. An attrinfo typically contains tuple field values (inserts and updates) or projection specifications (for reads); keyinfo contains the primary or unique key needed to locate a given tuple in this schema object.

The value shown by tot_frag_scans includes both full scans (that examine every row) and scans of subsets. Unique indexes and BLOB tables are never scanned, so this value, like other scan-related counts, is 0 for fragment replicas of these.

tot_scan_rows_examined may display less than the total number of rows in a given fragment replica, since ordered index scans can limited by bounds. In addition, a client may choose to end a scan before all potentially matching rows have been examined; this occurs when using an SQL statement containing a LIMIT or EXISTS clause, for example. tot_scan_rows_returned is always less than or equal to tot_scan_rows_examined.

tot_scan_bytes_returned includes, in the case of pushed joins, projections returned to the DBSPJ block in the NDB kernel.

tot_qd_frag_scans can be effected by the setting for the MaxParallelScansPerFragment data node configuration parameter, which limits the number of scans that may execute concurrently on a single fragment replica.