How to take a snapshot of a directory on Linux as it was at a particular point in time?

Question

I have a directory containing a lot of result log files (directory size ~ 4GB), and a set of processes running that keep on writing to these files.

To be able to correctly analyze the results at a later time, I want to copy the whole directory to an archive destination, and I cannot stop the processes.

I want a copy of the directory as it was at a particular point of time. As the size directory is huge (which means it takes about 40 seconds to copy) and some of the files are being written to, a normal cp -r does NOT give me a snapshot at a particular point in time, rather a snapshot of files spread over some 40 seconds. This is not good enough.

Is there a way to get an exclusive lock on the directory and all its components while copying?

Answer 1

LVM (Logical Volume Manager) supports snapshots. Since all Linux distribution has LVM builtin, it is the easiest for you. The only downside is you need to recreate your file system.

Here is a good tutorial http://rhcelinuxguide.wordpress.com/2006/06/03/what-is-a-logical-volume-manager-lvm-snapshot-and-how-do-i-use-it/

Answer 2

If cp is taking too long to provide the results you need, you're most probably not going to be satisfied with other options. Unix copy is about as basic as it gets for I/O overhead.

However, if you retarded the I/O processes buy leveraging another management process between the disk and the files as they were changing, the overhead may not matter. What I mean is, you could run the process in a VM (if this is possible), try snapshots with rsync, or even subversion.

The only other thing I can think of is maybe piping the log output to a symbolic link and updating the links to a secondary location with a batch file.

For instance:

file.log -> /mnt/disk1/logarea/file.log

file.log -> /mnt/disk/logarea2/file.log

So that you can update the symbolic link instantly, and the output pipes to the new location.

You might even be able to write a batch file to update the link, count to 45 seconds, and then switch it back - thus isolating the data you need.

Just thoughts, good luck.

Answer 3

A different way of doing it would be to use Logrotate to archive the logs every week/day/hour, then you essentially have snapshots at that interval.

Answer 4

Some filesystems provide snapshot facilities. Creating a filesystem (on an unused area of drive, or in a dedicated data file [loopback]) using one of these filesystems and then mounting that where your logs are written would allow you to snapshot the logs filesystem.

Filesystems known to support snapshots:

• ZFS (fuse based)
• FreeBSD's FFS
• LFS (Log Structured Filesystem)
• XFS

There are probably more, but those are the ones I can think of off the top of my head.

Snapshots don't copy the data, rather the freeze the data at a particular moment in time and create a virtual filesystem containing that data that can be mounted and used like any other filesystem (albeit read-only). The original filesystem continues to function as if nothing had happened.

Answer 5

Frankly, for what you're asking for the only thing that has worked for me is ZFS-FUSE. There is also a native ZFS for Linux but it's not really usable yet.