Every developer and every team faces confusion about COPY and ADD in the Dockerfile at some point. When I get this question, first I usually give the technical background, which is this:

Both ADD and COPY copy files and directories from the host machine into a Docker image, the difference is that ADD can also extract and copy local tar archives and it can also download files from URLs (a.k.a. the internet), and copy them into the Docker image. The best practice is to use COPY.

So COPY equals ADD minus the unpacking and URL fetching features. COPY is the preferred way, except if you unpack a local tar archive into a Docker image and you are certain that the local archive has the right format.

You can understand why this is the case looking at some background info. Read on...

Why COPY is preferred

The core purpose of ADD and COPY is to let Dockerfile developers copy files and directories from the host machine into the Docker image during image build.

Extracting archives and downloading files from the internet are common use-cases, these features are built into ADD.

The uncompression feature is described in the official documentation as follows:

If <src> is a local tar archive in a recognized compression format (identity, gzip, bzip2 or xz) then it is unpacked as a directory.

The following note on the same page further explains the behavior:

Note: Whether a file is identified as a recognized compression format is done solely based on the contents of the file, not the name of the file. For example, if an empty file happens to end with .tar.gz this will not be recognized as a compressed file, and will not generate any kind of decompression error message, rather the file will simply be copied to the destination.

This means that your final outcome depends on the contents of the file you intend to copy, and you don't get warnings if something goes wrong. This may make your build pipeline unpredictable.

To make life more reliable, we have the COPY instruction, which is "the same as ADD, but without the tar and URL handling". COPY does one thing and it does it well.

The best practice

Docker best practices suggest to always use COPY when you don't need extraction functionality, because COPY is more transparent.

In real-life projects COPY is sufficient in most scenarios, mainly because we rarely add tarballs to our applications' source code. The main use-case for tarballs, thus ADD, is when we create a base image from a tar archive. This doesn't happen very often. In this case ADD is preferred.

For all other use-cases we use COPY;

• We prefer COPY for copying files from the host machine into a Docker image.
• We use RUN with curl or wget to fetch files from URLs. ADD does not unpack files from the web anyway, so we are better off avoiding it entirely.

Let's see how you can accomplish unpacking and URL fetching.

Unpacking local archives

ADD unpacks archives from the host machine, it does not unpack files from URLs. To unpack an archive you just use it in its default form; ADD <src>... <dest>. Check out this sample Dockerfile:

FROM alpine:3.10

ADD bigfile.tar.xz /tmp/

When you build the image Docker will unpack the archive.

docker build -t yourname/alpine-bigfile .
Sending build context to Docker daemon  4.096kB
Step 1/2 : FROM alpine:3.10
 ---> 4d90542f0623
Step 2/2 : ADD bigfile.tar.xz /tmp/
 ---> 32cfa3eb41f7
Successfully built 32cfa3eb41f7
Successfully tagged yourname/alpine-bigfile:latest

Since the format of ADD is the exact same when you just copy a file or you unpack an archive, this might get tricky. As we mentioned earlier, if Docker does not recognize the archive format during the build, it will copy the archive as it is into the Docker image without warning. You can mitigate the risks by adding a check into your build pipeline.

Our archive in the example was recognized by Docker, so the file is uncompressed in our image:

docker run --rm -ti yourname/alpine-bigfile /bin/ash
/ # ls -al /tmp
total 12
drwxrwxrwt    1 root     root          4096 Jan 31 09:49 .
drwxr-xr-x    1 root     root          4096 Jan 31 09:50 ..
-rw-r--r--    1 501      dialout         29 Jan 31 09:46 bigfile

If you need a solution to share your image as an archive, check out our article How to Transfer/Move a Docker Image to Another System?.

Downloading and unpacking archives from a URL

For downloading and unpacking archives from the internet curl or wget are the better options, because it takes only one image layer to get the results you want. With ADD you'd grab the archive first in one layer, then uncompress it with RUN in another. This is not so efficient.

You can build a Dockerfile to curl an archive and uncompress it like shown below.

FROM alpine:3.10

RUN apk add --no-cache curl && \
  curl -SL https://github.com/yikaus/docker-alpine-base/raw/master/rootfs.tar.xz | tar -xJC /tmp

This takes one image layer and you have full control over the process.

One more thing

One more noteworthy difference between ADD and COPY is that COPY has the --from=<name|index> flag that lets you copy files from a previous build stage in a multi-stage build. ADD does not have this option.

This is another reason to use COPY as your preferred option.