Percona Server with XtraDB

fsp_reserve_free_extents fails to reserve space if FSP_SIZE < FSP_FREE_LIMIT

Reported by George Ormond Lorch III on 2012-11-27
6
This bug affects 1 person
Affects Status Importance Assigned to Milestone
Percona Server
Medium
George Ormond Lorch III
5.5
Medium
George Ormond Lorch III

Bug Description

An unsigned math error in fsp_reserve_free_extents causes function to believe that there are already many free extents available when FSP_SIZE < FSP_FREE_LIMIT, causing function to do nothing and possibly allow out of space errors to go undetected until attempted use of non-existent space occurs.

Related branches

Stewart Smith (stewart) on 2012-11-29
Changed in percona-server:
milestone: 5.5.28-29.2 → 5.5.29-29.4

This is an upstream bug. Was there any reason not report it there?

tags: added: innodb

The fix for this is correct in that it that handles size < free_limit situation correctly.

But do we arrive at size < free_limit in the first place, and if yes, how? That seems to require some sort of bug in fsp_fill_free_list to write the value at FSP_FREE_LIST that is larger than the value at FSP_SIZE.

I have just encountered this on 5.6, and if this situation has been analyzed on 5.5 already, it would be a great help.

George Ormond Lorch III (gl-az) wrote :

Laurynas, this was found while looking into handling file system quotas/enospc better without crashing for a customer. It didn't completely solve their problem but at least prevented the server from crashing as a result of the condition which was all they requested. IIRC there were already several related upstream bug reports.

George - I remember the context of this bug, but please note that I'm interested in some particular details which the context does not provide.
- Have we ever actually end up in FSP_SIZE < FSP_FREE_LIMIT situation or is it just defensive programming here?
- If yes, how?
- I searched and I was not able to find an upstream bug "fsp_reserve_free_extents fails to reserve space if FSP_SIZE < FSP_FREE_LIMIT", which would be a different bug from all the "MySQL fails if ENOSPACE" bugs.

To post a comment you must log in.
This report contains Public information  Edit
Everyone can see this information.

Other bug subscribers