Slow log parse performance improvement for splitByColon #54538
Description
Enhancement
Do cpu profile for query like:
SELECT
Digest,
Query,
Conn_ID,
(UNIX_TIMESTAMP(Time) + 0E0) AS timestamp,
Query_time,
Mem_max,
Process_keys
FROM
`INFORMATION_SCHEMA`.`CLUSTER_SLOW_QUERY`
WHERE
Time BETWEEN FROM_UNIXTIME(?)
AND FROM_UNIXTIME(?)
ORDER BY
Query_time DESC
LIMIT 100;
Find the splitByColon takes about 80% cpu times:
The implementation uses regexp to parse key: value strings:
tidb/pkg/executor/slow_query.go
Lines 526 to 531 in f158c65
However, in most cases, we can just go through the string, and take string before colon as key, string after colon as value, which seems can improve performance significantly.