Server C2Q 2 x 8G ram. RAID 5( 3 hdd ), the mysql 5.1.26-rc \\ Red Hat 4.1.2-14
When collecting(two years ago) were young and stupid, but the server generally fit into their parameters.
So we have a relatively high load on MySQL 601.90 requests per second, of them updates\\inserts — 2% and ~70% — stmp prepare\\execute\\close, to share pure select remains 34.84%
And a week ago the base has learned to die has created heaps of processes which worked for half an hour.
Oddity 1 — one hour all repaired ITSELF
In General, began raking the state of the server.
As one of the points in the engine's code was added to dump the runtime operations in the database back to the database.
This code worked for queries that took longer than 0.1 sec slow_log them does not see, but this is brakes...
In General, there went the strangeness — the most common query that's running it handles, performs 0.0001 will report to the database that it runs 0.5 or even TWO seconds...
Oddity number two is the brakes are going in small batches at 5-10 ping requests approximately every 11 seconds.
And at this point, usually only a few tables Toromont( ie I see a pack is essentially the same query in the log at this point)
Since 99 brake had queries on innoDB tables was performed some dances — enabled file_per_table and tables of total landfill(11Гб) extinct in their little files( final total size of 4Gb, fragmentation there was dibou )
LA servers, 0.9
disposal screw — 15-20%
Free memory is.
Idea where the brakes are and what to do — no
As Percona or MariaDB (5.1.6?)Bonus pack — when mysql hangs — hear from him is not off, the processes are not completed.
Nothing except kill -9....