This is the mail archive of the cygwin mailing list for the Cygwin project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

1.7 preallocate issue?

On Nov 14 10:38, Corinna Vinschen wrote:
> It shouldn't be hard to replicate if you modify the rsync statement to
> rsync only one such file.  The strace would be interesting in the first
> place.  You should get an error code from a function called
> fhandler_disk_file::ftruncate.

I can't replicate it the error on demand.  I've used the exact servers and
files that caused the issue in my testing and I cannot get it to repeat on
demand.  The issue, however, still occurs daily.  My analysis of the log
files from rsync do not yield any definitive clues on the issue either (e.g.
specific load, affecting multiple rsync proceses at the same time,
etc...none of these things are happening).  Other posts on "resource
temporarily unavailable" indicates a lack of threads or memory and kernel
modifications address it.  Memory is not the issue here per my monitoring.
Threads do not seem to make sense since the only time the error ever occurs
is when the posix_fallocate function is used in the rsync.

> - Are you perhaps really out of space?
Over 5TB's free on the array

> - Are quotas enabled on the target filesystem and you're hitting your
>   quota?
No OS limitations...the problem only occurred when we moved to more recent
builds of 1.7.  I also think it is related to the build we are using, Cygwin
1.7.0-28, since when I compiled the exact same rsync build/patch/code under
a previous version I did not have the problem (e.g. build date 7/19/08).
> - Is the fiber array a local FS or a remote FS?  NTFS/FAT32?  
Direct fiber attached, local FS, NTFS
>   If you don't find any other reason, we would have to check if
>   there are some OS or FS restrictions. 
No OS restrictions...been using the same configuration for two years without
this issue although clearly not on 1.7 with posix_fallocate for that entire


Unsubscribe info:
Problem reports:

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]