Taking Snapshots

The following section is from Sébastien Han's “OpenStack: Perform Consistent Snapshots” blog entry.

A snapshot captures the state of the file system, but not the state of the memory. Therefore, to ensure your snapshot contains the data that you want, before your snapshot you need to ensure that:

To ensure that important services have written their contents to disk (such as databases), we recommend that you read the documentation for those applications to determine what commands to issue to have them sync their contents to disk. If you are unsure how to do this, the safest approach is to simply stop these running services normally.

To deal with the "dirty" buffer issue, we recommend using the sync command before snapshotting:

# sync

Running sync writes dirty buffers (buffered blocks that have been modified but not written yet to the disk block) to disk.

Just running sync is not enough to ensure that the file system is consistent. We recommend that you use the fsfreeze tool, which halts new access to the file system, and create a stable image on disk that is suitable for snapshotting. The fsfreeze tool supports several file systems, including ext3, ext4, and XFS. If your virtual machine instance is running on Ubuntu, install the util-linux package to get fsfreeze:

# apt-get install util-linux

If your operating system doesn't have a version of fsfreeze available, you can use xfs_freeze instead, which is available on Ubuntu in the xfsprogs package. Despite the "xfs" in the name, xfs_freeze also works on ext3 and ext4 if you are using a Linux kernel version 2.6.29 or greater, since it works at the virtual file system (VFS) level starting at 2.6.29. The xfs_freeze version supports the same command-line arguments as fsfreeze.

Consider the example where you want to take a snapshot of a persistent block storage volume, detected by the guest operating system as /dev/vdb and mounted on /mnt. The fsfreeze command accepts two arguments:

To freeze the volume in preparation for snapshotting, you would do the following, as root, inside the instance:

# fsfreeze -f /mnt

You must mount the file system before you run the fsfreeze command.

When the fsfreeze -f command is issued, all ongoing transactions in the file system are allowed to complete, new write system calls are halted, and other calls that modify the file system are halted. Most importantly, all dirty data, metadata, and log information are written to disk.

Once the volume has been frozen, do not attempt to read from or write to the volume, as these operations hang. The operating system stops every I/O operation and any I/O attempts are delayed until the file system has been unfrozen.

Once you have issued the fsfreeze command, it is safe to perform the snapshot. For example, if your instance was named mon-instance and you wanted to snapshot it to an image named mon-snapshot, you could now run the following:

$ nova image-create mon-instance mon-snapshot

When the snapshot is done, you can thaw the file system with the following command, as root, inside of the instance:

# fsfreeze -u /mnt

If you want to back up the root file system, you can't simply run the preceding command because it will freeze the prompt. Instead, run the following one-liner, as root, inside the instance:

# fsfreeze -f / && sleep 30 && fsfreeze -u /