Please start any new threads on our new
site at https://forums.sqlteam.com. We've got lots of great SQL Server
experts to answer whatever question you can come up with.
| Author |
Topic |
|
robvolk
Most Valuable Yak
15732 Posts |
 Posted - 2010-10-29 : 18:04:40
|
| We experienced a strange mirror failover this afternoon, wondering if anyone can shed some light.Servers: Windows 2003 R2, SP2, SQL Server 2005 SP3, CU10, Version 9.0.4305, 16 GB RAMNY Server: Clustered, 2 nodes, NY02 active, NY01 passive/standby, DB Mirror PrincipalSLV1 Server: standalone, DB Mirror partnerMAS Server: standalone, DB Mirror witnessAccording to our event logs, we had two time related events:2:35:49 PM - The time delta between node NY02 and node NY01 is 54680484(in 100 nanosecs). 2:36:20 PM - The time delta between node NY02 and node NY01 is 1242984(in 100 nanosecs). At 2:36:12 PM, we experienced a mirror failover of 9 databases, that completed at 2:36:59. We found an event where the witness server (MAS) and mirror (SLV1) could not contact the cluster (NY) around this time, initiating mirror failover. The cluster did NOT fail over, and we don't have any specific alerts about network interruptions.We had another mirror failover around 3:34 pm, but no other associated events/warnings/alerts. The only thing we find questionable are the time service discrepancies listed above. Currently the SLV1 machine is primary and the NY machine is mirror, all DBs are synchronized and the witness is connected.Has anyone seen anything like this before? Or can confirm that the time service was not a factor?Thanks. |
|
|
robvolk
Most Valuable Yak
15732 Posts |
Posted - 2010-10-29 : 18:47:14
|
| Update: I need to correct something, we did NOT have a mirror failover at 3:34, we merely lost sync at that time, and re-established mirroring automatically about 1 minute later. Sorry for the confusion.I had perfmon running counters for DB mirroring and network activity for the relevant period. Are there any particular counters I should look at to reveal a problem? |
 |
|
|
robvolk
Most Valuable Yak
15732 Posts |
Posted - 2010-10-29 : 19:08:13
|
| Another update: mirroring timeout was set to the standard 10 seconds. I've gotten advice from a few people to increase it to 30 seconds to avoid problems during cluster failover. Since the cluster is not the principal server, we are leaving the setting at 10 seconds until we fail back to the cluster. |
|
|
|
|
|
|
|
|
DB Mirror & Cluster Question