Documentation Home
MySQL 8.0 Reference Manual
Related Documentation Download this Manual
PDF (US Ltr) - 43.2Mb
PDF (A4) - 43.3Mb
Man Pages (TGZ) - 296.0Kb
Man Pages (Zip) - 401.3Kb
Info (Gzip) - 4.3Mb
Info (Zip) - 4.3Mb
Excerpts from this Manual

29.12.20.5 Transaction Summary Tables

The Performance Schema maintains tables for collecting current and recent transaction events, and aggregates that information in summary tables. Section 29.12.7, “Performance Schema Transaction Tables” describes the events on which transaction summaries are based. See that discussion for information about the content of transaction events, the current and historical transaction event tables, and how to control transaction event collection, which is disabled by default.

Example transaction event summary information:

mysql> SELECT *
       FROM performance_schema.events_transactions_summary_global_by_event_name
       LIMIT 1\G
*************************** 1. row ***************************
          EVENT_NAME: transaction
          COUNT_STAR: 5
      SUM_TIMER_WAIT: 19550092000
      MIN_TIMER_WAIT: 2954148000
      AVG_TIMER_WAIT: 3910018000
      MAX_TIMER_WAIT: 5486275000
    COUNT_READ_WRITE: 5
SUM_TIMER_READ_WRITE: 19550092000
MIN_TIMER_READ_WRITE: 2954148000
AVG_TIMER_READ_WRITE: 3910018000
MAX_TIMER_READ_WRITE: 5486275000
     COUNT_READ_ONLY: 0
 SUM_TIMER_READ_ONLY: 0
 MIN_TIMER_READ_ONLY: 0
 AVG_TIMER_READ_ONLY: 0
 MAX_TIMER_READ_ONLY: 0

Each transaction summary table has one or more grouping columns to indicate how the table aggregates events. Event names refer to names of event instruments in the setup_instruments table:

Each transaction summary table has these summary columns containing aggregated values:

  • COUNT_STAR, SUM_TIMER_WAIT, MIN_TIMER_WAIT, AVG_TIMER_WAIT, MAX_TIMER_WAIT

    These columns are analogous to the columns of the same names in the wait event summary tables (see Section 29.12.20.1, “Wait Event Summary Tables”), except that the transaction summary tables aggregate events from events_transactions_current rather than events_waits_current. These columns summarize read-write and read-only transactions.

  • COUNT_READ_WRITE, SUM_TIMER_READ_WRITE, MIN_TIMER_READ_WRITE, AVG_TIMER_READ_WRITE, MAX_TIMER_READ_WRITE

    These are similar to the COUNT_STAR and xxx_TIMER_WAIT columns, but summarize read-write transactions only. The transaction access mode specifies whether transactions operate in read/write or read-only mode.

  • COUNT_READ_ONLY, SUM_TIMER_READ_ONLY, MIN_TIMER_READ_ONLY, AVG_TIMER_READ_ONLY, MAX_TIMER_READ_ONLY

    These are similar to the COUNT_STAR and xxx_TIMER_WAIT columns, but summarize read-only transactions only. The transaction access mode specifies whether transactions operate in read/write or read-only mode.

The transaction summary tables have these indexes:

TRUNCATE TABLE is permitted for transaction summary tables. It has these effects:

  • For summary tables not aggregated by account, host, or user, truncation resets the summary columns to zero rather than removing rows.

  • For summary tables aggregated by account, host, or user, truncation removes rows for accounts, hosts, or users with no connections, and resets the summary columns to zero for the remaining rows.

In addition, each transaction summary table that is aggregated by account, host, user, or thread is implicitly truncated by truncation of the connection table on which it depends, or truncation of events_transactions_summary_global_by_event_name. For details, see Section 29.12.8, “Performance Schema Connection Tables”.

Transaction Aggregation Rules

Transaction event collection occurs without regard to isolation level, access mode, or autocommit mode.

Transaction event collection occurs for all non-aborted transactions initiated by the server, including empty transactions.

Read-write transactions are generally more resource intensive than read-only transactions, therefore transaction summary tables include separate aggregate columns for read-write and read-only transactions.

Resource requirements may also vary with transaction isolation level. However, presuming that only one isolation level would be used per server, aggregation by isolation level is not provided.