size of saved docker images for base versus derived

user_2918239 0
Viewed 40

In an empty directory, I have run a sequence of commands shown in the following terminal capture:

$ echo -e "from alpine" > Dockerfile
$ docker build --tag alpine .
Sending build context to Docker daemon  2.048kB
Step 1/1 : from alpine
 ---> c059bfaa849c
Successfully built c059bfaa849c
Successfully tagged alpine:latest
$ docker image save -o alpine.tar alpine
$ echo -e "from alpine \nrun apk add --no-cache tini" > Dockerfile
$ docker build --tag tini .
Sending build context to Docker daemon  11.76MB
Step 1/2 : from alpine
 ---> c059bfaa849c
Step 2/2 : run apk add --no-cache tini
 ---> Using cache
 ---> e50ce83132ec
Successfully built e50ce83132ec
Successfully tagged tini:latest
$ docker image save -o tini.tar tini
$ ls -sh *.tar
 12M alpine.tar  5.7M tini.tar

As expected, the image labeled tini builds a further layer above the one shared with alpine. However, the resulting saved file for the former image is smaller than that for the latter.

Expected is that the image saved in the former file is a superset of the latter, and therefore larger.

Why is this expectation not confirmed by observation?

dockerfile docker containers linux
1 Answers

Best guess is something else was tagged alpine on your host when you did the build, and between the two commands, something pulled the correct alpine image down. But rather than guessing, you have the file to tell you exactly what was in that first alpine image. Extract the tar, review the manifest, and review the config the manifest points to. Here's an example with whatever old version of alpine:latest I happened to have on my machine:

$ cat manifest.json | jq .
[
  {
    "Config": "45683da4f97c23d92b878d2ad15eb66e94c9cab368e54797bcd4a4b20c915815.json",
    "RepoTags": [
      "test-alpine:latest"
    ],
    "Layers": [
      "4833c0e9afedb6022373ef7da2cee392c97a5f4438e7f52a1f940903db618437/layer.tar"
    ]
  }
]

$ cat 45683da4f97c23d92b878d2ad15eb66e94c9cab368e54797bcd4a4b20c915815.json | jq .
{
  "architecture": "amd64",
  "config": {
    "Env": [
      "PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin"
    ],
    "Cmd": [
      "/bin/sh"
    ],
    "OnBuild": null
  },
  "created": "2021-08-27T17:19:45.758611523Z",
  "history": [
    {
      "created": "2021-08-27T17:19:45.553092363Z",
      "created_by": "/bin/sh -c #(nop) ADD file:aad4290d27580cc1a094ffaf98c3ca2fc5d699fe695dfb8e6e9fac20f1129450 in / "
    },
    {
      "created": "2021-08-27T17:19:45.758611523Z",
      "created_by": "/bin/sh -c #(nop)  CMD [\"/bin/sh\"]",
      "empty_layer": true
    }
  ],
  "os": "linux",
  "rootfs": {
    "type": "layers",
    "diff_ids": [
      "sha256:e2eb06d8af8218cfec8210147357a68b7e13f7c485b991c288c2d01dc228bb68"
    ]
  }
}

When you compare the two images, if they were built from the same alpine base, they will have the same initial set of layers. In your case, they almost certainly won't.

user_5962850
Related