innodb

The innodb domain includes InnoDB metrics from INFORMATION_SCHEMA.INNODB_METRICS.

For example:

mysql> SELECT * FROM innodb_metrics WHERE name='trx_rseg_history_len' LIMIT 1\G
*************************** 1. row ***************************
           NAME: trx_rseg_history_len
      SUBSYSTEM: transaction
          COUNT: 0
      MAX_COUNT: 0
      MIN_COUNT: 0
      AVG_COUNT: NULL
    COUNT_RESET: 0
MAX_COUNT_RESET: 0
MIN_COUNT_RESET: 0
AVG_COUNT_RESET: NULL
   TIME_ENABLED: 2021-08-17 08:24:14
  TIME_DISABLED: NULL
   TIME_ELAPSED: 1905927
     TIME_RESET: NULL
         STATUS: enabled
           TYPE: value
        COMMENT: Length of the TRX_RSEG_HISTORY list

The exact NAME value is used for the Blip metric name. In the example above, the Blip metric name is trx_rseg_history_len, even though this metric means history list length (HLL). Metric names are unique by SUBSYSTEM.

The status.global domain includes many of the same metrics because, historically, only SHOW GLOBAL STATUS existed. It’s a best practice to collect InnoDB metrics with this domain and exclude them from status.global.

As a starting point, these are good InnoDB metrics to collect:

plan:
  collect:
    innodb:
      metrics:
        # Transactions
        - "trx_active_transactions"
        
        # Row locking
        - "lock_timeouts"
        - "lock_row_lock_current_waits"
        - "lock_row_lock_waits"
        - "lock_row_lock_time"
        
        # Page flushing
        - "buffer_flush_adaptive_total_pages"   #  adaptive flushing
        - "buffer_LRU_batch_flush_total_pages"  #  LRU flushing
        - "buffer_flush_background_total_pages" #  legacy flushing
        
        # Transaction log utilization (%)
        - "log_lsn_checkpoint_age"     # checkpoint age
        - "log_max_modified_age_async" # async flush point
        
        # Transaction log -> storage waits
        - "innodb_os_log_pending_writes"
        - "innodb_log_waits"
        
        # History List Length (HLL)
        - "trx_rseg_history_len"
        
        # Deadlocks
        - "lock_deadlocks"

None.

None.

Technically InnoDB metric names are unique by subsystem, but currently they’re unique for the whole table. It’s probably safe to graph them by metric name alone, but if there’s a name collision in the future, one metrics will be lost when reporting.

None.

See 17.15.6 InnoDB INFORMATION_SCHEMA Metrics Table.