suryagaddipati/docker-cache-volume-plugin

Docker volume driver to cache container artifacts

Docker volume driver to give your container exclusive view of underlying filesystem cache using overlayfs.Think of it as Read committed Isolation level for your filesystem cache between concurrently running containers. Changes made by container are synced back to cache on container exit.

Note: There is no conflict resolution between changes made by concurrent containers, latest container to exit overwrites all previous changes.

Installation

docker plugin install rapt/cachedriver

Usage

Volume name is expected to be in <cache-name>-<unique-id>

eg:

docker run --rm -it  --name one --volume-driver rapt/cachedriver  -v foo-one:/data busybox sh
/ # ls /data/
/ #  echo one >> /data/foo-one.txt

An underlying cache foo is created at this point but contents written to this cache by the running container is not visible to concurrently running containers. eg:

docker run --rm -it --name two --volume-driver rapt/cachedriver   -v foo-two:/data busybox sh
/ #  ls /data/        #foo-one.txt is not visible There

concurrently running containers one and two get an isolated view of the cache.

Once container one exits changes made by it are written back to the cache.

docker run --rm -it --name three --volume-driver  rapt/cachedriver  -v foo-three:/data busybox sh
/ #  ls /data/
/ #  foo-one.txt

changes made by container one are visible to container three after container one exits.

Use cases

• Build caches in a continuous integration system.

  Build Systems like maven, sbt cache their artifacts but multiple processes cannot use a sigle underlying maven cache due to concurrency issues and exclusive locks. Each maven process requires an isolated view of underlying cache.

  docker run --rm -it --name one --volume-driver rapt/cachedriver  -v foo-one:/~/.m2 busybox sh
  / #  mvn install  
  

  docker run --rm -it --name two --volume-driver  rapt/cachedriver  -v foo-two:/~/.m2 busybox sh
  / #  mvn install   # .m2 cache is primed here by one
  

How does it work?

Driver uses overlayfs to provide exclusive view of lower directory.

 docker run --rm -it --name three --volume-driver  rapt/cachedriver  -v foo-three:/data busybox sh

 / # echo foo-one > /data/foo-one.txt
 / # exit

Inside volume driver container

 / #  docker-runc exec -t 58211af9d85c0e6e095a822f93b6522f0a3fed776c8d9bd72f62a41ebfa7e5c2 sh

 / #  mount | grep foo
overlay on /mnt/cache/merged/foo/three type overlay (rw,relatime,lowerdir=/mnt/cache/lower/foo/0,upperdir=/mnt/cache/upper/foo/three,workdir=/mnt/cache/work/foo/three)

 / #  ls /mnt/cache/lower/foo/
  0    cache-state.json

 / #  cat /mnt/cache/lower/foo/cache-state.json
  {"state":{"latest":"0"}}

After container one exits

/ # ls /mnt/cache/lower/foo/
  0     cache-state.json  three

/ # cat /mnt/cache/lower/foo/cache-state.json
{"state":{"latest":"three"}}

Start another container for foo cache

docker run --rm -it --name four --volume-driver  rapt/cachedriver  -v foo-four:/data busybox sh
/ # ls /data
foo-one.txt

Inside volume container, container four uses three as lower

/ # mount | grep foo
overlay on /mnt/cache/merged/foo/four type overlay (rw,relatime,lowerdir=/mnt/cache/lower/foo/three,upperdir=/mnt/cache/upper/foo/four,workdir=/mnt/cache/work/foo/four)