26.10 Performance Schema Statement Digests and Sampling

When the parser receives an SQL statement, it computes a statement digest if that digest is needed, which is true if any of the following conditions are true:

The parser is also used by the STATEMENT_DIGEST_TEXT() and STATEMENT_DIGEST() functions, which applications can call to compute a normalized statement digest and a digest hash value, respectively, from an SQL statement.

The max_digest_length system variable value determines the maximum number of bytes available per session for computation of normalized statement digests. Once that amount of space is used during digest computation, truncation occurs: no further tokens from a parsed statement are collected or figure into its digest value. Statements that differ only after that many bytes of parsed tokens produce the same normalized statement digest and are considered identical if compared or if aggregated for digest statistics.

After the normalized statement has been computed, a SHA-256 hash value is computed from it. In addition:

Statement normalization transforms the statement text to a more standardized digest string representation that preserves the general statement structure while removing information not essential to the structure:

Consider these statements:

To normalize these statements, the parser replaces data values by ? and adjusts whitespace. Both statements yield the same normalized form and thus are considered “the same”:

The normalized statement contains less information but is still representative of the original statement. Other similar statements that have different data values have the same normalized form.

Now consider these statements:

In this case, the normalized statements differ because the object identifiers differ:

If normalization produces a statement that exceeds the space available in the digest buffer (as determined by max_digest_length), truncation occurs and the text ends with “...”. Long normalized statements that differ only in the part that occurs following the “...” are considered the same. Consider these statements:

If the cutoff happens to be right after the AND, both statements have this normalized form:

In this case, the difference in the second column name is lost and both statements are considered the same.

In the Performance Schema, statement digesting involves these components:

Some Performance Tables have a column that stores original SQL statements from which digests are computed:

The maximum space available for statement display is 1024 bytes by default. To change this value, set the performance_schema_max_sql_text_length system variable at server startup. Changes affect the storage required for all the columns just named.

The performance_schema_max_digest_length system variable determines the maximum number of bytes available per statement for digest value storage in the Performance Schema. However, the display length of statement digests may be longer than the available buffer size due to internal encoding of statement components such as keywords and literal values. Consequently, values selected from the DIGEST_TEXT column of statement event tables may appear to exceed the performance_schema_max_digest_length value.

The events_statements_summary_by_digest summary table provides a profile of the statements executed by the server. It shows what kinds of statements an application is executing and how often. An application developer can use this information together with other information in the table to assess the application's performance characteristics. For example, table columns that show wait times, lock times, or index use may highlight types of queries that are inefficient. This gives the developer insight into which parts of the application need attention.

The events_statements_summary_by_digest summary table has a fixed size. By default the Performance Schema estimates the size to use at startup. To specify the table size explicitly, set the performance_schema_digests_size system variable at server startup. If the table becomes full, the Performance Schema groups statements that have SCHEMA_NAME and DIGEST values not matching existing values in the table in a special row with SCHEMA_NAME and DIGEST set to NULL. This permits all statements to be counted. However, if the special row accounts for a significant percentage of the statements executed, it might be desirable to increase the summary table size by increasing performance_schema_digests_size.

For applications that generate very long statements that differ only at the end, increasing max_digest_length enables computation of digests that distinguish statements that would otherwise aggregate to the same digest. Conversely, decreasing max_digest_length causes the server to devote less memory to digest storage but increases the likelihood of longer statements aggregating to the same digest. Administrators should keep in mind that larger values result in correspondingly increased memory requirements, particularly for workloads that involve large numbers of simultaneous sessions (the server allocates max_digest_length bytes per session).

As described previously, normalized statement digests as computed by the parser are constrained to a maximum of max_digest_length bytes, whereas normalized statement digests stored in the Performance Schema use performance_schema_max_digest_length bytes. The following memory-use considerations apply regarding the relative values of max_digest_length and performance_schema_max_digest_length:

Because the Performance Schema statement event tables might store many digests, setting performance_schema_max_digest_length smaller than max_digest_length enables administrators to balance these factors:

The performance_schema_max_digest_length setting is not per session, it is per statement, and a session can store multiple statements in the events_statements_history table. A typical number of statements in this table is 10 per session, so each session will consume 10 times the memory indicated by the performance_schema_max_digest_length value, for this table alone.

Also, there are many statements (and digests) collected globally, most notably in the events_statements_history_long table. Here, too, N statements stored will consume N times the memory indicated by the performance_schema_max_digest_length value.

To assess the amount of memory used for SQL statement storage and digest computation, use the SHOW ENGINE PERFORMANCE_SCHEMA STATUS statement, or monitor these instruments:

The Performance Schema uses statement sampling to collect representative statements that produce each digest value in the events_statements_summary_by_digest table. These columns store sample statement information: QUERY_SAMPLE_TEXT (the text of the statement), QUERY_SAMPLE_SEEN (when the statement was seen), and QUERY_SAMPLE_TIMER_WAIT (the statement wait or execution time). The Performance Schema updates all three columns each time it chooses a sample statement.

When a new table row is inserted, the statement that produced the row digest value is stored as the current sample statement associated with the digest. Thereafter, when the server sees other statements with the same digest value, it determines whether to use the new statement to replace the current sample statement (that is, whether to resample). Resampling policy is based on the comparative wait times of the current sample statement and new statement and, optionally, the age of the current sample statement:

By default, performance_schema_max_digest_sample_age is 60 seconds (1 minute). To change how quickly sample statements “expire” due to age, increase or decrease the value. To disable the age-based part of the resampling policy, set performance_schema_max_digest_sample_age to 0.