xfs
[Top] [All Lists]

Re: Delaylog

To: Arkadiusz Miskiewicz <arekm@xxxxxxxx>
Subject: Re: Delaylog
From: Christoph Hellwig <hch@xxxxxxxxxxxxx>
Date: Tue, 14 Sep 2010 19:23:21 -0400
Cc: xfs@xxxxxxxxxxx
In-reply-to: <201009142106.24448.arekm@xxxxxxxx>
References: <AANLkTin6FSBK3tMXhSxDuLdjyb+R0LhNnaa2CKt37FcV@xxxxxxxxxxxxxx> <201009142106.24448.arekm@xxxxxxxx>
User-agent: Mutt/1.5.20 (2009-08-17)
On Tue, Sep 14, 2010 at 09:06:24PM +0200, Arkadiusz Miskiewicz wrote:
> On Tuesday 14 of September 2010, Fabricio Archanjo wrote:
> > Hey all,
> > 
> >    I just trying delaylog in my server that has a mysql database. When
> > i monted my /var/lib/mysql with delaylog option, it showed me:
> > "Enabling EXPERIMENTAL delayed logging feature - use at your own
> > risk". Ok, i know it's experimental, but what kind of problem could i
> > have using delaylog?

Basically you could hit a race or lockup in the code under high stress
or unusual workloads.  So far we just had one possible lockup under very
high dbench load.

> ... and what problems in case of system hang or power loss when compared to 
> nodelaylog mode?

The same as with the old log code - if you crash recently written data
might be lost.  Unless a really severe bugs shows up (in either the old
or new code) that only includes data since the last fsync/sync.  The
quantitative difference is that a lot more metadata is now cached in
core, so on a crash you can lose more recently written but not synced
metadata.

<Prev in Thread] Current Thread [Next in Thread>