Bug #70961 resistency when one of tmpdir become unavailable
Submitted: 20 Nov 2013 10:39
Reporter: martin fuxa Email Updates:
Status: Open Impact on me:
Category:MySQL Server Severity:S4 (Feature request)
Version: OS:Any
Assigned to: CPU Architecture:Any

[20 Nov 2013 10:39] martin fuxa
As DOC says "The multiple-directory feature can be used to spread the load between several physical disks",
as in some situation is reasonable have more disks as tmpdir. For performance RAID 1 is useless, RAID 0 fail after one disk fail. Then more single disk can be used.

But single disk is SPoF and shit happens. When disk fail, MySQL server still use it as tmpdir. And therefore query fails. And I remember situation (innodb workload only; 5.6.13 or 14) MySQL server crashed!
In case when is configured more disk (directories) as tmpdir (for example tmpdir = /tmp:/tmp2:/tmp3:/tmp4; which all dirs are individual single physical disks), would be nice after few errors server internaly
- marks this unavailable dir as failed
- logs something like: tmpdir /tmp2 unavailable, disabling.
- henceforward do not use it

This situation can become when filesystem is re-mounted after error by system (ext filesystem with errors=remount-ro) too.

Imho it's another feature than dynamic tmpdir (bugs #53119, #60036).

How to repeat:
see Description