Slow query log improvements

[Keyur]

We recently encountered an issue with a particular type of query was not using the right index and resulted in doing a filesort-on-disk. The query itself was fast (taking < 100ms) but there were many processes were running these with different parameters. Soon the bottleneck became write IO and the server stalled. The server had long_query_time as 1s.

We realized there's currently no way to gain visibility into queries doing tmp/filesort based disk IO.

The attached patch helps solve that by adding a new variable "log_slow_override" with similar values as "log_slow_filter". It will log statements to the slow query log that match certain execution plans irrespective of what long_query_time is. It also guesstimate the size of the filesort disk-file and adds it to the log file.

The patch also fixes a couple of other bugs:
1) The enum enum_log_slow_filter (SLOG_F_QC_NO, etc.) is unnecessary and checking against them in sql_parse.cc is wrong. Sys_var_set parses the values of the variable "log_slow_filter" into powers of 2 automatically and one should check against QPLAN_* to see if a particular flag is enabled.
2) The char* array log_slow_filter_name is off by one when compared to the values in QPLAN_*. It was missing "qc".

[Keyur (keyurdg)]

The patch is based off of Percona-Server-5.5.27-rel28.1

[Laurynas Biveinis (laurynas-biveinis)]

Keyur -

Thanks for your contribution. For the couple of other bugs that you mention, please follow bug 1030783, we are fixing them there.

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PS-2396