docker存储结构解析

由于aufs并未并入内核，故而目前只有Ubuntu系统上能够使用aufs作为docker的存储引擎，而其他系统上使用lvm thin provisioning（overlayfs是一个和aufs类似的union filesystem，未来有可能进入内核，但目前还没有；Lvm snapshot are useful for doing e.g. backup of a snapshot, but regress badly in performance when you start having many snapshots of the same device.）。为了实现lvm thin provisioning，docker启动时会设置一个100G的sparse文件（/var/lib/docker/devicemapper/devicemapper/data，元数据为/var/lib/docker/devicemapper/devicemapper/metadata），并将其作为devicemapper的存储池，而所有容器都从该存储池中分配默认10G的存储空间使用，如下图所示：




比如创建一个apache容器时devicemapper处理流程如下所示：
Create a snapshot of the base device.
Mount it and apply the changes in the fedora image.
Create a snapshot based on the fedora device.
Mount it and apply the changes in the apache image.
Create a snapshot based on the apache device.
Mount it and use as the root in the new container.

thin provisioning管理
使用lvm工具来创建一个thin pool：
dd if=/dev/zero of=lvm.img bs=1M count=100losetup /dev/loop7 lvm.imglosetup -apvcreate /dev/loop7vgcreate lvm_pool /dev/loop7# create thin poollvcreate -L 80M -T lvm_pool/thin_pool# create volume in thin poollvcreate -T lvm_pool/thin_pool -V 500M -n first_lv
docker启动时创建的默认存储池：#dmsetup table docker-253:1-138011042-pool0 209715200 thin-pool 7:2 7:1 128 32768 1 skip_block_zeroing # 209715200*512/1024/1024/1024=100GB
当启动容器后，会从该池中分配10G出来：#dmsetup table docker-253:1-138011042-641cdebd22b55f2656a560cd250e661ab181dcf2f5c5b78dc306df7ce62231f20 20971520 thin 253:2 166 # 20971520*512/1024/1024/1024=10GB该10G存储的分配过程为：dmsetup message /dev/mapper/docker-253:1-138011042-pool 0 "create_thin 166"dmsetup create docker-253:1-138011042-641cdebd22b55f2656a560cd250e661ab181dcf2f5c5b78dc306df7ce62231f3 --table "0 20971520 thin /dev/mapper/docker-253:1-138011042-pool 166"
创建快照:dmsetup suspend /dev/mapper/thindmsetup message /dev/mapper/yy_thin_pool 0 "create_snap 1 0"dmsetup resume /dev/mapper/thindmsetup create snap --table "0 40960 thin /dev/mapper/yy_thin_pool 1"
docker服务在启动的时候可以配置devicemapper的启动参数，docker -d --storage-opt dm.foo=bar，可选参数有以下几个：
dm.basesize 默认为10G，限制容器和镜像的大小
dm.loopdatasize 存储池大小，默认为100G
dm.datadev 存储池设备，默认生成一个/var/lib/docker/devicemapper/devicemapper/data文件
dm.loopmetadatasize 元数据大小，默认为2G
dm.metadatadev 元数据设备，默认生成一个/var/lib/docker/devicemapper/devicemapper/metadata文件
dm.fs 文件系统，默认ext4
dm.blocksize blocksize默认64K
dm.blkdiscard 默认true

最后看看启动一个容器后，该容器的配置是如何组织的。
每个容器创建后都会将其基本配置写入到/var/lib/docker/containers/中：#ls /var/lib/docker/containers/49f19ee979f6bf125c62779dcabf3bdce310b13d22e5c826752db202e509154e -ltotal 20-rw------- 1 root root 0 Nov 18 16:31 49f19ee979f6bf125c62779dcabf3bdce310b13d22e5c826752db202e509154e-json.log-rw-r--r-- 1 root root 1741 Nov 18 16:31 config.json-rw-r--r-- 1 root root 368 Nov 18 16:31 hostconfig.json-rw-r--r-- 1 root root 13 Nov 18 16:31 hostname-rw-r--r-- 1 root root 175 Nov 18 16:31 hosts-rw-r--r-- 1 root root 325 Nov 18 16:31 resolv.conf分配10G空间后会将容器存储配置写入到以下两个文件中：# cd /var/lib/docker#cat ./devicemapper/metadata/49f19ee979f6bf125c62779dcabf3bdce310b13d22e5c826752db202e509154e-init
{"device_id":174,"size":10737418240,"transaction_id":731,"initialized":false}
#cat ./devicemapper/metadata/49f19ee979f6bf125c62779dcabf3bdce310b13d22e5c826752db202e509154e
{"device_id":175,"size":10737418240,"transaction_id":732,"initialized":false}而容器的rootfs会mount到/var/lib/docker/devicemapper/mnt/container_id下：#mount | grep 49f1/dev/mapper/docker-253:1-138011042-49f19ee979f6bf125c62779dcabf3bdce310b13d22e5c826752db202e509154e on /var/lib/docker/devicemapper/mnt/49f19ee979f6bf125c62779dcabf3bdce310b13d22e5c826752db202e509154e type ext4 (rw,relatime,discard,stripe=16,data=ordered)

参考文档
http://blogs.gnome.org/alexl/2013/10/15/adventures-in-docker-land//article/4821616.htmlhttps://hustcat.github.io/docker-devicemapper2/https://github.com/docker/docker/tree/master/daemon/graphdriver/devmapper