ebuse

ebuse

ebuse is an NBD server which uses the new EBS direct APIs to expose an EBS snapshot as a network-accessible block device. You can use this to access data in an EBS snapshot without needing to first restore it to an EBS volume.

This implementation is definitely a bit of a toy, so I probably wouldn't use it for production workloads. There aren't any tests, it's not particularly hardened against error conditions, and it doesn't try to be particularly efficient in data that it fetches. I also can't promise that it will be particularly well maintained.

Also, keep in mind that the EBS direct APIs aren't entirely free, although they are quite cheap. Assume that you'll spend about 0.6c for every gigabyte you read.

Why "ebuse"? Originally I planned to build this using BUSE (well, a Go version of BUSE). "ebuse" is what you get when you smash "EBS" and "BUSE" together. It also seemed appropriate that it sounded abusive - this is probably not exactly what these APIs are intended for.

Usage

You can install ebuse by running

$ go get -u github.com/ebroder/ebuse

ebuse only has one required argument - the snapshot ID you want it to serve. It optionally accepts as flags the AWS region of that snapshot (although that can also be passed in via the AWS_REGION environment variable) and the UNIX socket path to bind the NBD server to (which defaults to somewhere reasonable in XDG_RUNTIME_DIR):

usage: ebuse [flags] snap-12345678
  -region string
    	AWS region of snapshot (default "us-east-1")
  -socket string
    	path to listen on (default "/run/user/1000/nbd.sock")

ebuse depends on having access to AWS credentials from the environment. It supports everything that the AWS SDK supports out of the box.

Once you've started the ebuse NBD server, you can connect to it using something like nbd-client:

$ sudo nbd-client -unix $XDG_RUNTIME_DIR/nbd.sock
Warning: the oldstyle protocol is no longer supported.
This method now uses the newstyle protocol with a default export
Negotiation: ..size = 8192MB
Connected /dev/nbd1

...and voila! You now have a block device (/dev/nbd1 in this case) backed by an EBS snapshot. Note that if you want to mount the block device, you'll need to pass -o ro to mount; otherwise it will try and fail to update the superblock. For journaled filesystems, you may also want -o noload. And if your EBS volume has a partition table, you should automatically get partition-specific block devices (e.g. /dev/nbd1p1) - you'll need to mount those instead.

To disconnect, run

sudo nbd-client -d /dev/nbd1