Starwind Service Shutdown with Large Cache

[comdot]

Latest released Starwind version running on Server 2012

I have about 65gb of cache in all which takes about 4/5 mins to write out on stopping the starwind service.

Problem is I cannot persuade windows that this length of time is expected - it keeps thinking the service has locked on shutdowns and force closing it, so I'm left having to manually stop the starwind service before restarting a node or face a full sync when it comes up.

Anyone got any ideas - I've tried the normal and extended the timer on the service in the reg but on shutdown there appears to be another service that monitors service shutdown and it kills services that take a long time to stop. I can't seem to find anyway to make that less aggressive!

Anyone any ideas?

[anton (staff)]

Indeed that sounds like an issue if Windows forces service to stop before flush is done. Not critical, just more work for re-sync when the node we'll be up again but we'll check what we could do to improve the things.

[Alex (staff)]

comdot,
Please check, do you have PreshutdownTimeout value in registry key of StarWind service?
It is in HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\services\StarWindService.

This value is set up during StarWind installation and defines wait interval for the service stopping on system shutdown.

[comdot]

Yeah its set to 1800000..

[Alex (staff)]

comdot,
Please, export Windows Event logs for the time when system has been shutdown, and send to support@starwindsoftware.com with StarWind service logs.
We'll try to find out why system is not waiting for the service.

[comdot]

Production system so I can't reboot it and then do a full sync - I've got like 5 TB of data the sync takes an age and I can't risk being on one node for that time unless I have too.

But I have explained - there is another process that runs at shutdown and monitors services it must have its own timer and if a process takes longer to shut down than it expects it kills it think it's part of MS's drive for quicker reboots - I was just wondering if anyone else had experienced it and knew the answer - I can keep digging.

I have live support contract - why don't you replicate it I imagine it would be easy to replicate, create some large devices run yourself up 60-70gb of cache on a server 2012 install and reboot it, the service will get stopped before shutting down gracefully.

Do you know what the preshutdowntimeout value is measured in? is it milliseconds?

[Alex (staff)]

Sure! I've thought, may be logs left from one of previous reboots.

Windows kills service on shutdown after three minutes by default. But StarWind has separate setting and is using special Windows API calls to prevent this.

PreshutdownTimeout is measured in milliseconds, so it's 30 minutes.

Of course, we'll try to reproduce it in our lab. Just trying to collect some more information about environment