On 10/31/2004 06:44 PM, Benito A. Venegas wrote:
Hi people,
Thanks to everybody for your great effort day by day to maintain this
great project...
Now..my problem.. :(
I'm experiencing some high fragmentations in one of my boxes, running
postgresql database.
Last Friday/Saturday I ran some extra maintenance to my postgres DB
(reindexing, analyze, vacuum full, etc.) due suddenly we started to have
very slow performance in our queries. This maintenance helped, but later
I ran xfs_db -r <partition> -c "frag -f" and the
fragmentation was still high (close to 90%) and xfs_fsr reduced it to 7%
after to complete 10 loops for all my file systems in /etc/mtab.
Today in the evening I took some statistics and fragmentations level was
64% !! :(
#xfs_db -r /dev/sdc1 -c "frag -f"
actual 2531, ideal 906, fragmentation factor 64.20%
Basic info:
-kernel 2.4.19 + xfs 1.2 + some extra patches
-postgres 7.2.3-1PGDG
Wow, those are some truly ancient versions of both the kernel and
postgresql.
#xfs_info /proj
meta-data=/proj isize=256 agcount=34, agsize=262144
blks
data = bsize=4096 blocks=8885945, imaxpct=25
= sunit=0 swidth=0 blks, unwritten=0
naming =version 2 bsize=4096
log =internal bsize=4096 blocks=1200 version=1
= sunit=0 blks
realtime =none extsz=65536 blocks=0, rtextents=0
-HW level
PE2650 PIII 1400 Mhz Dual CPU, 1Gb ram,
/proj is in RAID1 using Perc3/Di (Yes, I know this is a bad raid card,
but IMHO this is not causing the fragmentations)
Only postgres and a small mysql DB are using this partition.
Qs:
-Can I ran xfs_fsr daily instead weekly?
-DO you think if I update my kernel and recreate my partition, using
version 2 (or 3) I can mitigate my fragmentation level? (Risky task)
-Is really postgres the frag killer here (if any one of you is running
postgres and only the data is one paritition..)
I'm running postgresql 7.3.x and 7.4.x on several boxes with XFS using
2.4.x and 2.6.x kernels and i'm not seeing fragmentation at anything
even remotely close to what you're reporting. The worst of my boxes is
under 10%, and most are under 5%.
--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
L. Friedman netllama@xxxxxxxxxxxxx
Linux Step-by-step & TyGeMo: http://netllama.ipfox.com
19:05:00 up 63 days, 9:43, 1 user, load average: 1.36, 1.28, 1.16
|