-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1
Thierry Florac wrote:
I'm currently trying to migrate an old Zope-2.6.1 application to
Zope-2.9.9.
The new application platform is a Sun Sparc Solaris 10 server (dual
processors).
Application setup includes a ZEO server managing two databases (a main
one for common data and a catalog to store catalog), and two Zope
clients (one as main frontend and another one for uncommon
administration tasks, both with standard threads count of 4).
Database sizes (when packed) are of 280 and 400 MB ; Zope clients caches
are of 300 and 500 MB, for 4 and 8 objects.
All data are stored on two mirrored (via RAID 1) Hitachi SAN with quite
good writing speed.
Because you need fast *seek* speed, too, I would recommend trying your
app with the ZODB on local disk, first. In fact, I never recommend
putting the main ZODB on a non-local disk, for just this reason. The
usage pattern of FileStorage (one big file, writing only at the end, but
with lots of random-access reads throughout) is nearly worst-case for
such deployments.
I have one client whose setup is similar (Linux, but using a big iron
SAN), where the first phase of incremental repozo backup (the part
checking whether incremental is possible) takes longer by a factor of
two than just doing a full backup. Putting it on local disk both makes
writes go faster and makes incremental backup feasible.
If you must stay with the SAN, I would investigate whether
DirectoryStorage, which uses lots of litte files, is a better fit.
I actually have two problems with this configuration :
- storing an object into ZODB is VERY slow ; it can take more than 30
seconds when the object wasn't saved recently, event under light load. I
actually suspect that it's the catalog full-text search index which is
long to load, update and store. The old application under Zope-2.6.1
didn't have this problem (but also didn't use ZEO !). Is there a
specific way to handle catalogs with ZEO, should I avoid this kind of
configuration ?
You can change the application to use QueueCatalog, which defers writes
to a configurable set of indexes (typically the text index(es)) for
later processing. I often run that processing via a cron job or other
long-running, non-server process (running as a ZEO client).
- sometimes (but most often under quite heavy load) the Zope clients
completely hang and stop responding, without any access or error log,
and without any disk or CPU activity. I suppose it's a kind of deadlock,
how can I detect the source of the problem ?
Many thanks for any help or advise, as the resolution of these problems
is quite urgent...
As Alan said, the DeadlockDebugger product is designed to help diagnose
such issues.
Tres.
- --
===
Tres Seaver +1 540-429-0999 [EMAIL PROTECTED]
Palladion Software Excellence by Designhttp://palladion.com
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.6 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
iD8DBQFIfp7e+gerLs4ltQ4RAgmfAJ9K9TtMfLc3i0hndANvvrY/PaA9qwCgrof8
YAGIUADOgpkIJvb2A7c5JJM=
=76B6
-END PGP SIGNATURE-
___
For more information about ZODB, see the ZODB Wiki:
http://www.zope.org/Wikis/ZODB/
ZODB-Dev mailing list - ZODB-Dev@zope.org
http://mail.zope.org/mailman/listinfo/zodb-dev
