on 11-03-2009 8:37 AM
Hi all,
SAP Note 405827 "Journaled file system and raw devices on Linux" gives approval for the installation of SAP Systems based on Linux (in our case RedHat 5) and Oracle with ext3 filesystems.
When creating the journal the values for the check of the Filesystem against the Journal are set automatically:
+vmrhel5n01:~ # tune2fs -j /dev/sdb+
+tune2fs 1.38 (30-Jun-2005)+
+Creating journal inode: done+
+This filesystem will be automatically checked every 36 mounts or+
+180 days, whichever comes first. Use tune2fs -c or -i to override.+
Unfortunately I did not find any recommendation how to set this parameter (count of mounts or time value) - neither in the SAP Note mentioned above nor in the SDN.
In an high available environment it would be worse if the check was performed after a reboot of the machine for all filsystems and so delaying the start of the sap instance.
In my opinion there are several possibilities for the configuration in SAP enviroments:
1) Don't change the default setting: This should only be the right choise, if the time for the check is negligible even for large databases / filesystems and the feared start-delay won't occur.
2) Deacativate the regular checks (eg. by setting very large values).
3) Choosing diffrent values for each sapdata filesystem. Implementing that would result that after a reboot/remount only one Filesystem is checked, whereas the check-count for all other sapdata-filesystems is not reached yet.
Which of the three possibilities would you prefer? Is there another one? To which value would you set the parameters?
Thanks in advance
Christoph
We've handled situations like this by shutting the systems down on a scheduled maintenance weekend. That allowed us to reboot these boxes and let any scheduled fsck checks run as needed. It also kept us from running into any sudden surprises if we had a system failure during business hours (we never did encountered one).
That may not be possible if you are running a 24x7 shop but it this is how we handled the issue.
Hope that is helpful.
J. Haynes
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
In an high available environment it would be worse if the check was performed after a reboot of the machine for all filsystems and so delaying the start of the sap instance.
In a high availability environment you shouldn't have to reboot a machine as you will have a fail over. Besides, would you rather risk corrupt data over a file system check that takes place once in a blue moon ? ...doesn't make sense to me. I would rather have peace of mind and let it do it's standard/default checks.
--
Nelis
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
I assume that the counters for the checks are stored in the file system header on the disk-system and not on the operating system. So the count of mounts will not only increase after a reboot (which of course should not happen in an HA szenario) but also wenn performing a cluster switch. In the case the volumes are dismounted on the first node and mounted on the second note ... Here you'll probalby have the same problem when the count is the same for all filesystems. All of them wil lbe checked at once ..
Hi Christoph,
one thing to consider is that databases usually have their own data integrity checks and replay mechanisms, so a journal file system might only create another "integrity layer" below. What journal file systems are definitely useful are all parts that do not hold the database volumes (e.g. standard binary folders, log folders, ...).
Regarding performance, the standard settings for ext3 have been proven as the fastest one during several SAP I/O throughput measurements.
Thanks,
Hannes
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
Hello Hannes
thank you much for your fast reply. Yes it's true that the database has it's own integrity check.
So would you suggest to disable the Filesystem check for all sapdata filesystems or would you keep the values as set with tune2fs by default (36 mounts or 180 days). Don't you expect a start delay due to fileystem checks in this case?
Rgds
Christoph
Hi Christoph,
So would you suggest to disable the Filesystem check for all sapdata filesystems or would you keep the values as set with tune2fs by default (36 mounts or 180 days). Don't you expect a start delay due to fileystem checks in this case?
I'd like to pass over this question to the readers of this forum who use the ext3 file system and who have very long living SAP systems. What are your impressions and thoughts?
Thanks,
Hannes
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.