4. Frequently asked questions (FAQ)

4.1. About the project

4.1.1. Where did the name Charliecloud come from?

Charlie — Charles F. McMillan was director of Los Alamos National Laboratory from June 2011 until December 2017, i.e., at the time Charliecloud was started in early 2014. He is universally referred to as “Charlie” here.

cloud — Charliecloud provides cloud-like flexibility for HPC systems.

4.1.2. How do you spell Charliecloud?

We try to be consistent with Charliecloud — one word, no camel case. That is, Charlie Cloud and CharlieCloud are both incorrect.

4.2. Errors

4.2.1. How do I read the ch-run error messages?

ch-run error messages look like this:

$ ch-run foo -- echo hello
ch-run[25750]: can't find image: foo: No such file or directory (ch-run.c:107 2)

There is a lot of information here, and it comes in this order:

  1. Name of the executable; always ch-run.

  2. Process ID in square brackets; here 25750. This is useful when debugging parallel ch-run invocations.

  3. Colon.

  4. Main error message; here can't find image: foo. This should be informative as to what went wrong, and if it’s not, please file an issue, because you may have found a usability bug. Note that in some cases you may encounter the default message error; if this happens and you’re not doing something very strange, that’s also a usability bug.

  5. Colon (but note that the main error itself can contain colons too), if and only if the next item is present.

  6. Operating system’s description of the the value of errno; here No such file or directory. Omitted if not applicable.

  7. Open parenthesis.

  8. Name of the source file where the error occurred; here ch-run.c. This and the following item tell developers exactly where ch-run became confused, which greatly improves our ability to provide help and/or debug.

  9. Source line where the error occurred.

  10. Value of errno (see C error codes in Linux for the full list of possibilities).

  11. Close parenthesis.

Note: Despite the structured format, the error messages are not guaranteed to be machine-readable.

4.2.2. Tarball build fails with “No command specified”

The full error from ch-builder2tar or ch-build2dir is:

docker: Error response from daemon: No command specified.

You will also see it with various plain Docker commands.

This happens when there is no default command specified in the Dockerfile or any of its ancestors. Some base images specify one (e.g., Debian) and others don’t (e.g., Alpine). Docker requires this even for commands that don’t seem like they should need it, such as docker create (which is what trips up Charliecloud).

The solution is to add a default command to your Dockerfile, such as CMD ["true"].

4.2.3. ch-run fails with “can’t re-mount image read-only”

Normally, ch-run re-mounts the image directory read-only within the container. This fails if the image resides on certain filesystems, such as NFS (see issue #9). There are two solutions:

  1. Unpack the image into a different filesystem, such as tmpfs or local disk. Consult your local admins for a recommendation. Note that Lustre is probably not a good idea because it can give poor performance for you and also everyone else on the system.

  2. Use the -w switch to leave the image mounted read-write. This may have an impact on reproducibility (because the application can change the image between runs) and/or stability (if there are multiple application processes and one writes a file in the image that another is reading or writing).

4.2.4. ch-image fails with “certificate verify failed”

When ch-image interacts with a remote registry (e.g., via push or pull subcommands), it will verify the registry’s HTTPS certificate. If this fails, ch-image will exit with the error “certificate verify failed”.

This situation tends to arise with self-signed or institutionally-signed certificates, even if the OS is configured to trust them. We use the Python HTTP library Requests, which on many platforms includes its own CA certificates bundle, ignoring the bundle installed by the OS.

Requests can be directed to use an alternate bundle of trusted CAs by setting environment variable REQUESTS_CA_BUNDLE to the bundle path. (See the Requests documentation for details.) For example:

$ export REQUESTS_CA_BUNDLE=/usr/local/share/ca-certificates/registry.crt
$ ch-image pull registry.example.com/image:tag

Alternatively, certificate verification can be disabled entirely with the --tls-no-verify flag. However, users should enable this option only if they have other means to be confident in the registry’s identity.

4.3. Unexpected behavior

4.3.1. What do the version numbers mean?

Released versions of Charliecloud have a pretty standard version number, e.g. 0.9.7.

Work leading up to a released version also has version numbers, to satisfy tools that require them and to give the executables something useful to report on --version, but these can be quite messy. We refer to such versions informally as pre-releases, but Charliecloud does not have formal pre-releases such as alpha, beta, or release candidate.

Pre-release version numbers are not in order, because this work is in a DAG rather than linear, except they precede the version we are working towards. If you’re dealing with these versions, use Git.

Pre-release version numbers are the version we are working towards, followed by: ~pre, the branch name if not master with non-alphanumerics removed, the commit hash, and finally dirty if the working directory had uncommitted changes.


  • 0.2.0 : Version 0.2.0. Released versions don’t include Git information, even if built in a Git working directory.

  • 0.2.1~pre : Some snapshot of work leading up to 0.2.1, built from source code where the Git information has been lost, e.g. the tarballs Github provides. This should make you wary because you don’t have any provenance. It might even be uncommitted work or an abandoned branch.

  • 0.2.1~pre+1a99f42 : Master branch commit 1a99f42, built from a clean working directory (i.e., no changes since that commit).

  • 0.2.1~pre+foo1.0729a78 : Commit 0729a78 on branch foo-1, foo_1, etc. built from clean working directory.

  • 0.2.1~pre+foo1.0729a78.dirty : Commit 0729a78 on one of those branches, plus un-committed changes.

4.3.2. --uid 0 lets me read files I can’t otherwise!

Some permission bits can give a surprising result with a container UID of 0. For example:

$ whoami
$ echo surprise > ~/cantreadme
$ chmod 000 ~/cantreadme
$ ls -l ~/cantreadme
---------- 1 reidpr reidpr 9 Oct  3 15:03 /home/reidpr/cantreadme
$ cat ~/cantreadme
cat: /home/reidpr/cantreadme: Permission denied
$ ch-run /var/tmp/hello cat ~/cantreadme
cat: /home/reidpr/cantreadme: Permission denied
$ ch-run --uid 0 /var/tmp/hello cat ~/cantreadme

At first glance, it seems that we’ve found an escalation – we were able to read a file inside a container that we could not read on the host! That seems bad.

However, what is really going on here is more prosaic but complicated:

  1. After unshare(CLONE_NEWUSER), ch-run gains all capabilities inside the namespace. (Outside, capabilities are unchanged.)

  2. This include CAP_DAC_OVERRIDE, which enables a process to read/write/execute a file or directory mostly regardless of its permission bits. (This is why root isn’t limited by permissions.)

  3. Within the container, exec(2) capability rules are followed. Normally, this basically means that all capabilities are dropped when ch-run replaces itself with the user command. However, if EUID is 0, which it is inside the namespace given --uid 0, then the subprocess keeps all its capabilities. (This makes sense: if root creates a new process, it stays root.)

  4. CAP_DAC_OVERRIDE within a user namespace is honored for a file or directory only if its UID and GID are both mapped. In this case, ch-run maps reidpr to container root and group reidpr to itself.

  5. Thus, files and directories owned by the host EUID and EGID (here reidpr:reidpr) are available for all access with ch-run --uid 0.

This is not an escalation. The quirk applies only to files owned by the invoking user, because ch-run is unprivileged outside the namespace, and thus he or she could simply chmod the file to read it. Access inside and outside the container remains equivalent.


4.3.3. Why does ping not work?

ping fails with “permission denied” or similar under Charliecloud, even if you’re UID 0 inside the container:

$ ch-run $IMG -- ping
PING ( 56 data bytes
ping: permission denied (are you root?)
$ ch-run --uid=0 $IMG -- ping
PING ( 56 data bytes
ping: permission denied (are you root?)

This is because ping needs a raw socket to construct the needed ICMP ECHO packets, which requires capability CAP_NET_RAW or root. Unprivileged users can normally use ping because it’s a setuid or setcap binary: it raises privilege using the filesystem bits on the executable to obtain a raw socket.

Under Charliecloud, there are multiple reasons ping can’t get a raw socket. First, images are unpacked without privilege, meaning that setuid and setcap bits are lost. But even if you do get privilege in the container (e.g., with --uid=0), this only applies in the container. Charliecloud uses the host’s network namespace, where your unprivileged host identity applies and ping still can’t get a raw socket.

The recommended alternative is to simply try the thing you want to do, without testing connectivity using ping first.

4.3.4. Why is MATLAB trying and failing to change the group of /dev/pts/0?

MATLAB and some other programs want pseudo-TTY (PTY) files to be group-owned by tty. If it’s not, Matlab will attempt to chown(2) the file, which fails inside a container.

The scenario in more detail is this. Assume you’re user charlie (UID=1000), your primary group is nerds (GID=1001), /dev/pts/0 is the PTY file in question, and its ownership is charlie:tty (1000:5), as it should be. What happens in the container by default is:

  1. MATLAB stat(2)s /dev/pts/0 and checks the GID.

  2. This GID is nogroup (65534) because tty (5) is not mapped on the host side (and cannot be, because only one’s EGID can be mapped in an unprivileged user namespace).

  3. MATLAB concludes this is bad.

  4. MATLAB executes chown("/dev/pts/0", 1000, 5).

  5. This fails because GID 5 is not mapped on the guest side.

  6. MATLAB pukes.

The workaround is to map your EGID of 1001 to 5 inside the container (instead of the default 1001:1001), i.e. --gid=5. Then, step 4 succeeds because the call is mapped to chown("/dev/pts/0", 1000, 1001) and MATLAB is happy.

4.3.5. ch-builder2tar gives incorrect image sizes

ch-builder2tar often finishes before the progress bar is complete. For example:

$ ch-builder2tar mpihello /var/tmp
 373MiB 0:00:21 [============================>                 ] 65%
146M /var/tmp/mpihello.tar.gz

In this case, the .tar.gz contains 392 MB uncompressed:

$ zcat /var/tmp/mpihello.tar.gz | wc
2740966 14631550 392145408

But Docker thinks the image is 597 MB:

$ sudo docker image inspect mpihello | fgrep -i size
        "Size": 596952928,
        "VirtualSize": 596952928,

We’ve also seen cases where the Docker-reported size is an underestimate:

$ ch-builder2tar spack /var/tmp
 423MiB 0:00:22 [============================================>] 102%
162M /var/tmp/spack.tar.gz
$ zcat /var/tmp/spack.tar.gz | wc
4181186 20317858 444212736
$ sudo docker image inspect spack | fgrep -i size
        "Size": 433812403,
        "VirtualSize": 433812403,

We think that this is because Docker is computing size based on the size of the layers rather than the unpacked image. We do not currently have a fix; see issue #165.

4.3.6. My second-level directory dev is empty

Some image tarballs, such as official Ubuntu Docker images, put device files in /dev. These files prevent unpacking the tarball, because unprivileged users cannot create device files. Further, these files are not needed because ch-run overmounts /dev anyway.

We cannot reliably prevent device files from being included in the tar, because often that is outside our control, e.g. docker export produces a tarball. Thus, we must exclude them at unpacking time.

An additional complication is that ch-tar2dir can handle tarballs both with a single top-level directory and without, i.e. “tarbombs”. For example, best practice use of tar on the command line produces the former, while docker export (perhaps via ch-builder2tar) produces a tarbomb.

Thus, ch-tar2dir uses tar --exclude to exclude from unpacking everything under ./dev and */dev, i.e., directory dev appearing at either the first or second level are forced to be empty.

This yields false positives if you have a tarbomb image with a directory dev at the second level containing stuff you care about. Hopefully this is rare, but please let us know if it is your use case.

4.3.7. My password that contains digits doesn’t work in VirtualBox console

VirtualBox has confusing Num Lock behavior. Thus, you may be typing arrows, page up/down, etc. instead of digits, without noticing because console password fields give no feedback, not even whether a character has been typed.

Try using the number row instead, toggling Num Lock key, or SSHing into the virtual machine.

4.4. How do I ...

4.4.1. My app needs to write to /var/log, /run, etc.

Because the image is mounted read-only by default, log files, caches, and other stuff cannot be written anywhere in the image. You have three options:

  1. Configure the application to use a different directory. /tmp is often a good choice, because it’s shared with the host and fast.

  2. Use RUN commands in your Dockerfile to create symlinks that point somewhere writeable, e.g. /tmp, or /mnt/0 with ch-run --bind.

  3. Run the image read-write with ch-run -w. Be careful that multiple containers do not try to write to the same files.

4.4.2. Which specific sudo commands are needed?

For running images, sudo is not needed at all.

For building images, it depends on what you would like to support. For example, do you want to let users build images with Docker? Do you want to let them run the build tests?

We do not maintain specific lists, but you can search the source code and documentation for uses of sudo and $DOCKER and evaluate them on a case-by-case basis. (The latter includes sudo if needed to invoke docker in your environment.) For example:

$ find . \(   -type f -executable \
           -o -name Makefile \
           -o -name '*.bats' \
           -o -name '*.rst' \
           -o -name '*.sh' \) \
         -exec egrep -H '(sudo|\$DOCKER)' {} \;

4.4.3. OpenMPI Charliecloud jobs don’t work

MPI can be finicky. This section documents some of the problems we’ve seen. mpirun can’t launch jobs

For example, you might see:

$ mpirun -np 1 ch-run /var/tmp/mpihello -- /hello/hello
App launch reported: 2 (out of 2) daemons - 0 (out of 1) procs
[cn001:27101] PMIX ERROR: BAD-PARAM in file src/dstore/pmix_esh.c at line 996

We’re not yet sure why this happens — it may be a mismatch between the OpenMPI builds inside and outside the container — but in our experience launching with srun often works when mpirun doesn’t, so try that. Communication between ranks on the same node fails

OpenMPI has many ways to transfer messages between ranks. If the ranks are on the same node, it is faster to do these transfers using shared memory rather than involving the network stack. There are two ways to use shared memory.

The first and older method is to use POSIX or SysV shared memory segments. This approach uses two copies: one from Rank A to shared memory, and a second from shared memory to Rank B. For example, the sm byte transport layer (BTL) does this.

The second and newer method is to use the process_vm_readv(2) and/or process_vm_writev(2)) system calls to transfer messages directly from Rank A’s virtual memory to Rank B’s. This approach is known as cross-memory attach (CMA). It gives significant performance improvements in benchmarks, though of course the real-world impact depends on the application. For example, the vader BTL (enabled by default in OpenMPI 2.0) and psm2 matching transport layer (MTL) do this.

The problem in Charliecloud is that the second approach does not work by default.

We can demonstrate the problem with LAMMPS molecular dynamics application:

$ srun --cpus-per-task 1 ch-run /var/tmp/lammps_mpi -- \
  lmp_mpi -log none -in /lammps/examples/melt/in.melt
[cn002:21512] Read -1, expected 6144, errno = 1
[cn001:23947] Read -1, expected 6144, errno = 1
[cn002:21517] Read -1, expected 9792, errno = 1
[... repeat thousands of times ...]

With strace(1), one can isolate the problem to the system call noted above:

process_vm_readv(...) = -1 EPERM (Operation not permitted)
write(33, "[cn001:27673] Read -1, expected 6"..., 48) = 48

The man page reveals that these system calls require that the process have permission to ptrace(2) one another, but sibling user namespaces do not. (You can ptrace(2) into a child namespace, which is why gdb doesn’t require anything special in Charliecloud.)

This problem is not specific to containers; for example, many settings of kernels with YAMA enabled will similarly disallow this access.

So what can you do? There are a few options:

  • We recommend simply using the --join family of arguments to ch-run. This puts a group of ch-run peers in the same namespaces; then, the system calls work. See the ch-run man page for details.

  • You can also sometimes turn off single-copy. For example, for vader, set the MCA variable btl_vader_single_copy_mechanism to none, e.g. with an environment variable:

    $ export OMPI_MCA_btl_vader_single_copy_mechanism=none

    psm2 does not let you turn off CMA, but it does fall back to two-copy if CMA doesn’t work. However, this fallback crashed when we tried it.

  • The kernel module XPMEM enables a different single-copy approach. We have not yet tried this, and the module needs to be evaluated for user namespace safety, but it’s quite a bit faster than CMA on benchmarks. I get a bunch of independent rank-0 processes when launching with srun

For example, you might be seeing this:

$ srun ch-run /var/tmp/mpihello -- /hello/hello
0: init ok cn036.localdomain, 1 ranks, userns 4026554634
0: send/receive ok
0: finalize ok
0: init ok cn035.localdomain, 1 ranks, userns 4026554634
0: send/receive ok
0: finalize ok

We were expecting a two-rank MPI job, but instead we got two independent one-rank jobs that did not coordinate.

MPI ranks start as normal, independent processes that must find one another somehow in order to sync up and begin the coupled parallel program; this happens in MPI_Init().

There are lots of ways to do this coordination. Because we are launching with the host’s Slurm, we need it to provide something for the containerized processes for such coordination. OpenMPI must be compiled to use what that Slurm has to offer, and Slurm must be told to offer it. What works for us is a something called “PMI2”. You can see if your Slurm supports it with:

$ srun --mpi=list
srun: MPI types are...
srun: mpi/pmi2
srun: mpi/openmpi
srun: mpi/mpich1_shmem
srun: mpi/mpich1_p4
srun: mpi/lam
srun: mpi/none
srun: mpi/mvapich
srun: mpi/mpichmx
srun: mpi/mpichgm

If pmi2 is not in the list, you must ask your admins to enable Slurm’s PMI2 support. If it is in the list, but you’re seeing this problem, that means it is not the default, and you need to tell Slurm you want it. Try:

$ export SLURM_MPI_TYPE=pmi2
$ srun ch-run /var/tmp/mpihello -- /hello/hello
0: init ok wc035.localdomain, 2 ranks, userns 4026554634
1: init ok wc036.localdomain, 2 ranks, userns 4026554634
0: send/receive ok
0: finalize ok

4.4.4. How do I run X11 apps?

X11 applications should “just work”. For example, try this Dockerfile:

FROM debian:stretch
RUN    apt-get update \
    && apt-get install -y xterm

Build it and unpack it to /var/tmp. Then:

$ ch-run /scratch/ch/xterm -- xterm

should pop an xterm.

If your X11 application doesn’t work, please file an issue so we can figure out why.

4.4.5. How do I specify an image reference?

You must specify an image for many use cases, including FROM instructions, the source of an image pull (e.g. ch-image pull or docker pull), the destination of an image push, and adding image tags. Charliecloud calls this an image reference, but there appears to be no established name for this concept.

The syntax of an image reference is not well documented. This FAQ represents our understanding, which is cobbled together from the Dockerfile reference, the docker tag documentation, and various forum posts. It is not a precise match for how Docker implements it, but it should be close enough.

We’ll start with two complete examples with all the bells and whistles:

  1. example.com:8080/foo/bar/hello-world:version1.0

  2. example.com:8080/foo/bar/hello-world@sha256:f6c68e2ad82a

These references parse into the following components, in this order:

  1. A valid hostname; we assume this matches the regular expression [A-Za-z0-9.-]+, which is very approximate. Optional; here example.com.

  2. A colon followed by a decimal port number. If hostname is given, optional; otherwise disallowed; here 8080.

  3. If hostname given, a slash.

  4. A path, with one or more components separated by slash. Components match the regex [a-z0-9_.-]+. Optional; here foo/bar. Pedantic details:

    • Under the hood, the default path is library, but this is generally not exposed to users.

    • Three or more underscores in a row is disallowed by Docker, but we don’t check this.

  5. If path given, a slash.

  6. The image name (tag), which matches [a-z0-9_.-]+. Required; here hello-world.

  7. Zero or one of:

    • A tag matching the regular expression [A-Za-z0-9_.-]+ and preceded by a colon. Here version1.0 (example 1).

    • A hexadecimal hash preceded by the string @sha256:. Here f6c68e2ad82a (example 2).

      • Note: Digest algorithms other than SHA-256 are in principle allowed, but we have not yet seen any.

Detail-oriented readers may have noticed the following gotchas:

  • A hostname without port number is ambiguous with the leading component of a path. For example, in the reference foo/bar/baz, it is ambiguous whether foo is a hostname or the first (and only) component of the path foo/bar. The resolution rule is: if the ambiguous substring contains a dot, assume it’s a hostname; otherwise, assume it’s a path component.

  • The only character than cannot go in a POSIX filename is slash. Thus, Charliecloud uses image references in filenames, replacing slash with percent (%). Because this character cannot appear in image references, the transformation is reversible.

    An alternate approach would be to replicate the reference path in the filesystem, i.e., path components in the reference would correspond directly to a filesystem path. This would yield a clearer filesystem structure. However, we elected not to do it because it complicates the code to save and clean up image reference-related data, and it does not address a few related questions, e.g. should the host and port also be a directory level.

Usually, most of the components are omitted. For example, you’ll more commonly see image references like:

  • debian, which refers to the tag latest of image debian from Docker Hub.

  • debian:stretch, which is the same except for tag stretch.

  • fedora/httpd, which is tag latest of fedora/httpd from Docker Hub.

See charliecloud.py for a specific grammar that implements this.

4.4.6. Can I build or pull images using a tool Charliecloud doesn’t know about?

Yes. Charliecloud deals in well-known UNIX formats like directories, tarballs, and SquashFS images. So, once you get your image into some format Charliecloud likes, you can enter the workflow.

For example, skopeo is a tool to pull images to OCI format, and umoci can flatten an OCI image to a directory. Thus, you can use the following commands to run an Alpine 3.9 image pulled from Docker hub:

$ skopeo copy docker://alpine:3.9 oci:/tmp/oci:img
$ ls /tmp/oci
blobs  index.json  oci-layout
$ umoci unpack --rootless --image /tmp/oci:img /tmp/alpine:3.9
$ ls /tmp/alpine:3.9
$ ls /tmp/alpine:3.9/rootfs
bin  etc   lib    mnt  proc  run   srv  tmp  var
dev  home  media  opt  root  sbin  sys  usr
$ ch-run /tmp/alpine:3.9/rootfs -- cat /etc/alpine-release

4.4.7. How do I authenticate with SSH during ch-image build?

The simplest approach is to run the SSH agent on the host. ch-image then leverages this with two steps:

  1. pass environment variable SSH_AUTH_SOCK into the build, with no need to put ARG in the Dockerfile or specify --build-arg on the command line; and

  2. bind-mount host /tmp to guest /tmp, which is where the SSH agent’s listening socket usually resides.

Thus, SSH within the container will use this existing SSH agent on the host to authenticate without further intervention.

For example, after making ssh-agent available on the host, which is OS and site-specific:

$ ssh-add
Enter passphrase for /home/charlie/.ssh/id_rsa:
Identity added: /home/charlie/.ssh/id_rsa (/home/charlie/.ssh/id_rsa)
$ ssh-add -l
4096 SHA256:aN4n2JeMah2ekwhyHnb0Ug9bYMASmY+5uGg6MrieaQ /home/charlie/.ssh/id_rsa (RSA)
$ cat ./Dockerfile
FROM alpine:latest
RUN apk add openssh
RUN ssh git@github.com
$ ch-image build -t foo -f ./Dockerfile .
  3 RUN ['/bin/sh', '-c', 'echo $SSH_AUTH_SOCK']
  4 RUN ['/bin/sh', '-c', 'ssh git@github.com']
Hi charlie! You've successfully authenticated, but GitHub does not provide shell access.

Note this example is rather contrived — bare SSH sessions in a Dockerfile rarely make sense. In practice, SSH is used as a transport to fetch something, e.g. with scp(1) or git(1). See the next entry for a more realistic example.

4.4.8. SSH stops ch-image build with interactive queries

This often occurs during an SSH-based Git clone. For example:

FROM alpine:latest
RUN apk add git openssh
RUN git clone git@github.com:hpc/charliecloud.git
$ ch-image build -t foo -f ./Dockerfile .
3 RUN ['/bin/sh', '-c', 'git clone git@github.com:hpc/charliecloud.git']
Cloning into 'charliecloud'...
The authenticity of host 'github.com (' can't be established.
RSA key fingerprint is SHA256:nThbg6kXUpJWGl7E1IGOCspRomTxdCARLviKw6E5SY8.
Are you sure you want to continue connecting (yes/no/[fingerprint])?

At this point, the build stops while SSH waits for input.

This happens even if you have github.com in your ~/.ssh/known_hosts. This file is not available to the build because ch-image runs ch-run with --no-home, so RUN instructions can’t see anything in your home directory.

Solutions include:

  1. Change to anonymous HTTPS clone, if available. Most public repositories will support this. For example:

    FROM alpine:latest
    RUN apk add git
    RUN git clone https://github.com/hpc/charliecloud.git
  2. Approve the connection interactively by typing yes. Note this will record details of the connection within the image, including IP address and the fingerprint. The build also remains interactive.

  3. Edit the image’s system SSH config to turn off host key checking. Note this can be rather hairy, because the SSH config language is quite flexible and the first instance of a directive is the one used. However, often the changes can be simply appended:

    FROM alpine:latest
    RUN apk add git openssh
    RUN printf 'StrictHostKeyChecking=no\nUserKnownHostsFile=/dev/null\n' \
        >> /etc/ssh/ssh_config
    RUN git clone git@github.com:hpc/charliecloud.git

    Check your institutional policy on whether this is permissible, though it’s worth noting that users almost never verify the host fingerprints anyway.

    This will not record details of the connection in the image.

  4. Turn off host key checking on the SSH command line. (See caveats in the previous item.) The wrapping tool should provide a way to configure this command line. For example, for Git:

    FROM alpine:latest
    RUN apk add git openssh
    ARG GIT_SSH_COMMAND="ssh -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null"
    RUN git clone git@github.com:hpc/charliecloud.git
  5. Add the remote host to the system known hosts file, e.g.:

    FROM alpine:latest
    RUN apk add git openssh
    RUN echo 'github.com, ssh-rsa AAAAB3NzaC1yc2EAAAABIwAAAQEAq2A7hRGmdnm9tUDbO9IDSwBK6TbQa+PXYPCPy6rbTrTtw7PHkccKrpp0yVhp5HdEIcKr6pLlVDBfOLX9QUsyCOV0wzfjIJNlGEYsdlLJizHhbn2mUjvSAHQqZETYP81eFzLQNnPHt4EVVUh7VfDESU84KezmD5QlWpXLmvU31/yMf+Se8xhHTvKSCZIFImWwoG6mbUoWf9nzpIoaSjB+weqqUUmpaaasXVal72J+UX2B+2RPW3RcT0eOzQgqlJL3RKrTJvdsjE3JEAvGq3lGHSZXy28G3skua2SmVi/w4yCE6gbODqnTWlg7+wC604ydGXA8VJiS5ap43JXiUFFAaQ==' >> /etc/ssh/ssh_known_hosts
    RUN git clone git@github.com:hpc/charliecloud.git

    This records connection details in both the Dockerfile and the image.

Other approaches could be found with web searches such as “automate unattended SSH” or “SSH in cron jobs”.