Copy Files to a pod

[ltamaster]

Is there a way to copy files to a pod?, I am trying with connect_get_namespaced_pod_exec but I think is very a limited workaround.

Thanks

[aogier]

Hi, unfortunately it doesn't seem there is another way to cope with that as even the kubectl implementation or this ticket seems to suggest this is the only way.
Have you managed to copy files via exec() ?

[ltamaster]

Hi @aogier,
Yes, I used the exec to copy files. My user case is coping scripts to pods (a rundeck plugin), so using exec is enough. However, my workaround won't work with binary files.

[aogier]

Hi @ltamaster, AFAICT current examples/implementation use tar on a standard stream pipe, like:

source -> tar -> stdout -> kubernetes -> stdin -> tar -> target

so I'd expect it should work for any file type, as long as tar can cope with them. If you don't mind share your current implementation I could try to help you make things work and maybe promote it to an example script to give to others too :)

[ltamaster]

Hi @aogier ,

I am doing something like this:

    # Calling exec interactively.
    exec_command = [shell]
    resp = stream(api.connect_get_namespaced_pod_exec, name, namespace,
                  command=exec_command,
                  stderr=True, stdin=True,
                  stdout=True, tty=False,
                  _preload_content=False)

    file = open(source_file, "r")

    commands = []
    commands.append("cat <<'EOF' >" + destination_file + "\n")
    commands.append(file.read())
    commands.append("EOF\n")

    while resp.is_open():
        resp.update(timeout=1)
        if resp.peek_stdout():
            print("STDOUT: %s" % resp.read_stdout())
        if resp.peek_stderr():
            print("STDERR: %s" % resp.read_stderr())

        if commands:
            c = commands.pop(0)
            resp.write_stdin(c)
        else:
            break

    resp.close()

[aogier]

Ah, I've seen that now. There are some problems with strings send over the websocket, this will in turn also become a python 2/3 issue. I think the problem reside in kubernetes.stream.WSClient implementation, I've trying patching his write_channel() method that way:

def write_channel(self, channel, data):
    """Write data to a channel."""
    self.sock.send(bytes(chr(channel), 'utf-8') + data)

now your modified example does not give errors:

# Copying file
exec_command = ['/bin/sh']
resp = stream(api.connect_get_namespaced_pod_exec, name, 'default',
              command=exec_command,
              stderr=True, stdin=True,
              stdout=True, tty=False,
              _preload_content=False)

source_file = '/bin/sh'
destination_file = '/tmp/sh'

file = open(source_file, "rb")

buffer = b''
with open(source_file, "rb") as file:
    buffer += file.read()

commands = []
commands.append(bytes("cat <<'EOF' >" + destination_file + "\n", 'utf-8'))
commands.append(buffer)
commands.append(bytes("EOF\n", 'utf-8'))
commands.append(bytes("date\n", 'utf-8'))

while resp.is_open():
    resp.update(timeout=1)
    if resp.peek_stdout():
        print("STDOUT: %s" % resp.read_stdout())
    if resp.peek_stderr():
        print("STDERR: %s" % resp.read_stderr())
    if commands:
        c = commands.pop(0)
        #print("Running command... %s\n" % c)
        resp.write_stdin(c)
    else:
        break

resp.write_stdin(bytes("date\n", 'utf-8'))
sdate = resp.readline_stdout(timeout=3)
print("Server date command returns: %s" % sdate)
resp.write_stdin(bytes("whoami\n", 'utf-8'))
user = resp.readline_stdout(timeout=3)
print("Server user is: %s" % user)
resp.close()

however, output is unexpected:

Server date command returns: None
Server user is: None

and copied file is truncated. I'd like to use tar rather than shell heredocs just for.

What do you think ? Ciao

[aogier]

Ok this PoC works well:

# Copying file
exec_command = ['tar', 'xvf', '-', '-C', '/']
resp = stream(api.connect_get_namespaced_pod_exec, name, 'default',
              command=exec_command,
              stderr=True, stdin=True,
              stdout=True, tty=False,
              _preload_content=False)

source_file = '/tmp/dash.tar'
destination_file = '/tmp/sh'

file = open(source_file, "rb")

buffer = b''
with open(source_file, "rb") as file:
    buffer += file.read()

commands = []
commands.append(buffer)

while resp.is_open():
    resp.update(timeout=1)
    if resp.peek_stdout():
        print("STDOUT: %s" % resp.read_stdout())
    if resp.peek_stderr():
        print("STDERR: %s" % resp.read_stderr())
    if commands:
        c = commands.pop(0)
        #print("Running command... %s\n" % c)
        resp.write_stdin(c)
    else:
        break
resp.close()

Things to do:

• programmatically produce tar stream
• send a PR against python-base repo

Some preliminary tests will be needed, though, as well as six'ing websocket/example code

HTH, ciao !

[aogier]

real tar file PoC implemented here, let's wait for kubernetes-client/python-base#52 needed for this to properly work.

ATM only client -> pod binary traffic works on py3, either direction seems to work in py2, ymmv

ciao

[ltamaster]

Hi @aogier ,

I tested your code and It worked great. I was trying to figure out how I can set a particular destination path on the tar command, is that possible?

Thanks
Luis

[aogier]

Hi @ltamaster, first of all I'm very glad this is working yeah :)

Defining destination path is indeed possibile and it depends on how you want to do it. I'll suggest you couple of option starting from what i prefer the most:

Using proposed tar -C /

From manual, tar's -C flag:

Change to DIR before performing any operations.
This option is order-sensitive, i.e. it affects all options that follow.

so if you want to put under /usr/local/foo/bar/:

oggei@cane:/tmp$ cd $(mktemp -d)
oggei@cane:/tmp/tmp.u9nN0XILDP$ mkdir -p usr/local/foo/bar
oggei@cane:/tmp/tmp.u9nN0XILDP$ echo hello > usr/local/foo/bar/my_file
oggei@cane:/tmp/tmp.u9nN0XILDP$ tar cv . > ../package.tar
./
./usr/
./usr/local/
./usr/local/foo/
./usr/local/foo/bar/
./usr/local/foo/bar/my_file
oggei@cane:/tmp/tmp.u9nN0XILDP$

That way, you have a /tmp/package.tar that when decompressed place a usr/local/foo/bar/my_file under your root filesystem. Tar being tar, this is not limited to a single file BTW (in fact even the example above ships 4 dir and 1 file).

A similar concept is used eg. in Debian packaging system, where a filesystem root is produced under debian/$pkg_name and then compressed in .deb package.

Runtime -C flag definition

You could write your code so you define operations' root via -C flag, and then you tar a hierarchy relative to this root. It's more or less like the above, except you start from a custom root and not / (so you can tar c file > archive.tar and then place it where you want via -C). Seems less straightforward to me, but it works.

OK that's all, happy hacking on that :)

[ltamaster]

Thanks @aogier

Adding this on your code made the trick:
tar.add(name=source_file,arcname=destination_path+"/"+file)

[aogier]

ah brilliant, this is a lot simpler ... I admit it, didn't studied well the tar module 🐐

thanks for sharing !

[ltamaster]

BTW, I got an error message on the line resp.close():

ERROR: websocket: close status: 891

The file is copied OK; the message is produced when that line is called.

[aogier]

I've pushed a new feature/476-copy-example branch with an updated submodule revision pointer could you please try this one ? No problem happens to me with either python 2 and 3, double check you're using 7b133cf for repo and kubernetes-client/python-base@7d67d12 for base submodule ! :)

[ltamaster]

Ok, I will try that