Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

I have multipath IO configured server 2012 blade that shows warnings like the following during MPIO path failure:

The IO operation at logical block address 0 for Disk 7 was retried.

I know what is causing the warning to happen so I am not looking for the cause but what does this message actually mean?

Does it mean that if this IO was a write operation then server actually lost data that it was trying to write?

Thank you for any light you can shed on the meaning of this warning message.

share|improve this question
add comment

3 Answers 3

up vote 10 down vote accepted

No it does not mean that the data was lost. It simply means that the IRP (IO Request Packet) timed out while the IO System waited for it to complete, and so it was tried again. When a thread begins any IO operation, the IO manager creates an IRP to represent the operation as it passes through the system.

The IRP gets stored in its initial state in a buffer/look-aside list, so that it can be retried if it fails the first time. That provides the atomicity that one would expect from any transactional system so that we can be more confident that you're not going to get a bunch of corrupted or incomplete data written to your disk.

This event makes perfect sense in the event of an MPIO failure. Say Windows goes to read or write something from SAN storage. The request is dispatched, and at the same instant, I cut one of the cables to the SAN. That request is never going to complete, and so Windows will try the request again, only this time the request will follow the other path.

These events also occur when the disks are overburdened or just really slow. You might notice these messages coincide with scheduled backups, etc. The disk might just be slow and busy, and some random IRP timed out and had to try again. The IRP could be getting stuck in an interrupt service routine, or a deferred procedure call, or whatever.

I could see having a lot of IO filter drivers in your stack exacerbating this issue as well.

It's not that this behavior did not occur just like this in previous versions of Windows, it's just that Microsoft apparently decided to surface these events in Win8/Server 2012.

Edit: You can find the outstanding IRPs of a thread with a kernel debugger: kd> !irp 1a2b3c4d, where you previously found that address by issuing the command kd> !process 8f7d6c4a which will list all the IRPs associated to the threads associated with that process. kd> !process 0 0 to list all the processes running.

Once you list the information about an IRP using the !irp command, you can easily spot which driver last handled the IRP because it will have a > pointing to it in the list. Then to get more information about what that driver was doing with that IRP, do a kd> !devobj 1a2b3c4d5e6f where that is the actual address of the device object.

Then do a kd> dt 0x1a2b3c``3c2b1a _CLASS_PRIVATE_FDO_DATA using the address of the PrivateFdoData structure you got. (Just one backtick; I couldn't get the parser to do it.)

Now you're ready to dump the AllTransferPacketsList data structure you got from PrivateFdoData.

The idea is, you're tracking down what driver was doing what with the IRP the last time it was seen. If the IRP is AWOL for too long, it's timed out and retried from the beginning. This can be caused by so many things... even a stray cosmic ray. But the important thing is that the transaction will be retried from the beginning, and it will not be considered complete until the IO manager says it is.

Oh, and there's also thread-agnostic IO which is a completely different can of worms. :)

For further reading on this topic, I highly recommend chapter 8, I/O System, of Windows Internals 6th edition, from Mark Russinovich, Margosis, et al.

*Edit: * I did finally find the official KB for this error: http://support.microsoft.com/kb/2819485/EN-US

The IO operation should be retried 8 times, once per minute, until Windows gives up.

Edit: As promised: http://blogs.msdn.com/b/ntdebugging/archive/2013/04/30/interpreting-event-153-errors.aspx

share|improve this answer
    
Thank you Ryan, I was hoping that it meant that the request was retired but the data was not lost and another request would created to try writing the data again. Can you reference any of the sources for your answer (books, articles, a note indicating that you have access to the windows source code because your a huge EA customer and did a debug trace to find this information, etc.)? I would love to understand this further. –  Chris Magnuson Apr 14 '13 at 19:36
1  
Edited my post to address your follow up questions. Chances are I'll have more info to add later. –  Ryan Ries Apr 14 '13 at 21:45
1  
Any one who can drop to the Windows Debugger to support their point earns some serious kudos in my book. Couldn't up vote the answer again so upvoting the comment will have to do. I have the Windows Internals 6th edition part 1 and I am off to purchase part 2 with chapter 8 now. Thanks –  Chris Magnuson Apr 17 '13 at 17:01
    
As promised: blogs.msdn.com/b/ntdebugging/archive/2013/04/30/… –  Ryan Ries Apr 30 '13 at 23:38
add comment

No, there would be a different message, and (hopefully) one of the application layers would throw an exception if it failed to successfully save data.

Prior to Windows Server 2012 (or hotfix 2819485 if on Windows Server 2008 R2), the system would silently retry when these timeouts occurred. The purpose of the message is to increase visibility about these occurrences. They may indicate a capacity issue or driver defect, and in the case of iSCSI, other operating system defects may attribute to the delay.

In the case of external (not direct-attached) storage, some vendors in the past have increased the timeout value, for example to 60 seconds. However, given the default number of retries by higher layer components such as the iSCSI initiator, this could mean that several minutes may elapse before the system initiated a failover. That would obviously be suboptimal behavior.

More information:

Registry Entries for SCSI Miniport Drivers
http://msdn.microsoft.com/en-us/library/windows/hardware/ff563970%28v=vs.85%29.aspx

https://blogs.msdn.com/b/san/archive/2011/09/01/the-windows-disk-timeout-value-understanding-why-this-should-be-set-to-a-small-value.aspx


Microsoft has released an update that provides the capability to specify the threshold for storport.sys operations.

After you install this update, you can log an event when the latency time for I/O to storage is equal to, or greater than, a threshold. The threshold value can be set by the user. This operation is performed at the Adapter Driver level so that you can see if there is a performance issue on the SAN. Then, you can contact a storage vendor to address the issue.

Note: This update restores the functionality that was provided in Windows 7 and Windows Server 2008 R2. When the functionality is enabled, the threshold value is measured in 100 nanoseconds (0.0001 milliseconds). Additionally, the following values are logged in the event:

BuildIoDuration: Length of time that the MINIPORT has spent in the build I/O function for this request StartIoDuration: Length of time that the MINIPORT has spent in the start I/O function for this request DataTransferLength: Size of the transfer in bytes

Update that improves the logging capabilities of the Storport.sys driver in Windows Server 2012
http://support.microsoft.com/kb/2819476

Windows 8 and Windows Server 2012 cumulative update: April 2013
http://support.microsoft.com/kb/2822241

share|improve this answer
add comment

Might be a late post, but I have found that it can be caused with VSS. We had a client who was running veeam but had forgot to turn off windows server back up (the disk was removed) It caused a shed load of problems and this error was the main one.

Stopped the back up and wham, no errors.

share|improve this answer
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.