March 11, 2013 at 1:39 pm
Hi,
We recently lost a drive the contained the transaction logs for a mirrored server. The principal databases responded to this event in various ways.
Some databases showed as disconnected but where still accessible.
Some databases said they where still principal and synchronizing (although they could not have been)
Some databases said suspect and stayed this way after the reboot. We saw DTC related errors in the windows logs for these db's.
Some said suspended
We are not sure why we had so many variations if they all suffered from the same cause, the loss of a drive with tlogs. Which of those are in theory the result one should expect?
When the drive failed on the mirrored machine, the principal server became very slow. SSMS would time out when we tried to log in from another machine. Expanding the databases folder would freeze up SSMS and applications that used these servers where failing. We had to reboot the entire box twice and manually rebuild mirrors. The drive failure brought the whole infrastructure down.
The drives are SSD so I am nervous this will happen again once they hit there maximum writes.
They are both SQL 2008 servers running with automatic fail-over (synchronous) with a witness. Unfortunately our windows logs start to over write themselves once they got to a certain size so all error logs of the events are gone, but we do have SQL logs. Has anyone had a drive fail in a mirrored environment and remember how SQL Server handled it?
March 14, 2013 at 8:24 am
Is it possible you had a drive failure in an array? I ask because some of the databases may have had no transaction log information on the failed drive, some may have had some data and others mayhave had the entire log on the failed drive.
To be honest, the databases that aren't synchronised/restoring should have the mirror broken and reestablished once the drive has been replaced. The problem is that your transaction logs have no redundancy at the mirror side and even though a mirror is good when the principal falls over, you have just been asked the question, "what happens if something happens to the mirror"?
At a guess, I think your principals were affected adversely by having logs to write and having nowhere to write them to!
We got around this problem by having a storage device divided into 2 LUNs, each an internal mirror of the other. In this way we were able to protect against disk failure in one LUN by failing over to the second. Not exactly cheap but it will help to mitigate against the problem you seem to have.
March 19, 2013 at 6:52 am
It turns out there was a mix of logs and database files on the drive which is why some databases responded differently then the others. Two days ago we had another drive fail but this time on the principal server and it required a reboot. Once the reboot happend some databases came back, but over all we had to take the most recent backups and restore them on the principal and then reconfigure mirroring again. We have had some very bad luck, this happened on St Pattys day so it was very very bad timing.
So, does anyone else have a mirror with witness setup? I am asking because we receive disconnection messages fairly often and we are wondering how normal this is. It does not fail over because it will connect again in time. Last night we received 4 emails about disconnects, we receive at least one a day so we turned the timeout up to 120 seconds.
March 19, 2013 at 7:25 am
The disconnection problems could be a network issue that doesn't cause a failover but is enough to disrupt the mirroring process. Once the connection has been reestablished does mirroring continue normally?
March 19, 2013 at 7:29 am
Yes they do resume. The network guys say they don't see any bad packets coming through the switch interfaces.
March 19, 2013 at 7:49 am
The problem doesn't necessarily have to be anything to do with bad packets but rather with available bandwidth. Are there any bottlenecks occuring at the time of the disconnects?
Viewing 6 posts - 1 through 5 (of 5 total)
You must be logged in to reply to this topic. Login to reply