coder / coder-v1-cli

Command line for Coder v1. For Coder v2, go to https://github.com/coder/coder

Coder v1 CLI

This is the command line utility for Coder v1. If you are using Coder v2 / Coder OSS, use these instructions to install the CLI.

The Coder v1 CLI is now closed-source. You may download binary releases from this repo.

Coder v2 is open-source and the recommended version for new Coder users.

coder-v1-cli's Issues

Clean up secrets create UX

$ coder secrets create new2 --from-prompt
value: ********

Should just be $ coder secrets create new2 no flag needed. Also it's OK to not star out the value as long it's not shown in terminal history. It would be nice to let the user know again what key they are assigning to walk them through it and confirm that it was successful or not.

e.g.

Enter your secret value for KEY:
...
Key KEY created successfully, access this key in any environment using $ coder get new2

Coder sh freezes

I just had this happen for the first time ever, ~20 minutes after updating to the wsep version. After returning to my terminal after a bit of idling, it's completely unresponsive. No inputs are registered, ctrl+c / ctrl+d won't escape, I had to close the terminal and reconnect.

The old wush implementation would just kick me back out to my local session periodically, so I assume the disconnect that is happening isn't being gracefully treated with the wsep implementation.

Only suggest latest release in README

master may have changes that only work with the unreleased version

Set coder_enterprise permissions to 0600

0620 are too open for OpenSSH....

Mouse inputs treated improperly in coder sh

Happens if I click on my terminal. Doesn't seem to happen in tmux, and mouse mode works fine with it, so not sure if there's anything too problematic here. Just feels a little jank.

Give better error when rsync doesn't exist on remote

right now it looks like

coder sync ~/Code/go/src/go.coder.com/enterprise/. fossel-2:~/enterprise
2020-04-20 14:17:02 INFO	doing initial sync (/Users/garrett/Code/go/src/go.coder.com/enterprise/. -> ~/enterprise)
protocol version mismatch -- is your shell clean?
(see the rsync man page for an explanation)
rsync error: protocol incompatibility (code 2) at /BuildRoot/Library/Caches/com.apple.xbs/Sources/rsync/rsync-54/rsync/compat.c(61) [sender=2.6.9]
2020-04-20 14:17:05 FATAL	sync: rsync: exit status 2

Add create and del subcommands to manage devurls

Proposal to add new subcommands,

• create <env_name> <port> <access>
• del <env_name> <port>

... allowing management of devurls and a full CRUD capability via coder-cli.

• Requires prior login as a user with access to admin <env_name>
• Subject to same site-install/org devurl access policies as doing this through the web frontend
• The create subcommand should be an idempotent operation. Calling create on an existing <env_name> <port> will update the devurl's <access> if a new value is specified, otherwise the command will do nothing and return success.

These commands would be useful both for end users and for integration testing with cemanager, enabling scripted devurl testing to ensure the API endpoints haven't broken by changes either in coder-cli or enterprise.

The create and del subcommands may require work on cdr/enterprise as well but the existing endpoints used by the web FE can/should be usable. should be able to use existing endpoints available for the web frontend.

TBD: subcommand names. create and del may be too generic if coder-cli eventually manages other resources. Will implement as these for now, but suggest addurl and delurl instead?

Add environment status in list

It would be nice to know which envs are running or not in the env list. There is a column for "updating" but not for "running".

Use mutagen over SSH

... Now that we have this new shiny SSH server

cc @cmoog

Link out to enterprise.coder.com docs

Right now we have duplicate documentation.

stdin / stdout behaves really weirdly at random

Came back to a terminal session after a while and got this behavior:

It wasn't showing my input, and line breaks didn't seem to be working. My input was roughly

• echo wtf
• ls -la
• enter a bunch of times
• ls -la

Not sure what caused this, but it happened in two separate instances.

login: Malformed login URL arg breaks login, sh and sync

As a naive user I tried

coder login localhost:8080
coder login master.cdr.dev

The browser auto-launch of the login URL is malformed in the above cases (and the text on terminal if autolaunch fails)

If I specify the full URL eg.
coder login http://localhost:8080

the URL generated is correct, and subsequent coder sh and coder sync ops work.

Recommend adding logic to prepend http:// if not specified by the user.

Running 'coder urls' without any arguments errors

> coder urls
2020-07-30 10:30:44 INFO	found ["work" "nathan" "hello" "commit-judge" "mydev" "nix"]
2020-07-30 10:30:44 FATAL	environment "" not found

We should instead check that the environment name isn't empty. If it is, we should print usage.

Clear screen after disconnect

If you're in a "full screen" command like vim, nano, tmux etc. and you get disconnected, the whole screen keeps the characters from the full screen command. This can be really confusing. I'm pretty sure that ssh's default behavior on a disconnect is to clear the screen, but this might require a little testing about what most people would expect.

Build fails on Darwin

coder login message is busted

[coder@coder ~/enterprise/customer/site (devurl-ui *)]$ coder login https://master.cdr.dev
2020-05-22 17:34:13 INFO        visit {https   master.cdr.dev /internal-auth/  %!s(bool=false) local_service=http://127.0.0.1:45701 } to login

Create foundation for performing bulk actions

Allow admins to give a --user flag that specifies which users's resources should be the target of crud operations. Example:

$ coder secrets --user [email protected] create mysql-password --from-literal 123password

Related: https://github.com/cdr/enterprise/pull/4497

tunnel command

I'm thinking a coder tunnel <env> <env_port>:<local_port> command could be a nice UX even though it would basically just call ssh -L <port>:localhost:<port> coder:<env>.

Does anyone have a strong opinion?

Add input while waiting for oauth callback to allow ssh

Right now, the cli login expects a HTTP callback, so when working over ssh, need to manually copy/paste the link into a different terminal and curl it. Having a text input to do the copy/paste directly on the cli would help a lot. (That's how concourse handles it for instance).
To be clear, that would be in addition of what we have, not instead.

Default to only environment & configurable default environment

If you only have one environment, and you run a command that has an environment target (e.g. coder sh or coder sync) and you omit the target, we should just run the command against your only environment. Likewise, when you have multiple, you should be able to configure a default environment.

gcloud and kubectl both have default namespace / projects, so it would be similar to that.

Suggestion originally comes from @creack.

Make one line install script

Something like https://classic.yarnpkg.com/en/docs/install/#alternatives-stable that makes as few assumptions about the user's system, and works on as many systems as possible.

Support TTY messaging in coder sh

So we can use a native terminal instead of the browser based one

Integrate generate to work with new licensor

Sales should still be able to manually generate licenses

coder url command fails

$ coder login
...
$ coder url russ-env1 3000
2020-06-23 14:49:34 FATAL	json: cannot unmarshal number into Go value of type main.DevURL
$ git log --oneline -1
c69c791 (HEAD -> master, tag: v1.8.0, origin/master, origin/HEAD) Merge pull request #54 from cdr/53-detect-missing-protocol

Add less thorough integration tests that don't require containerization so we can test Windows and MacOS

macOS / Darwin release not signed

The binary release for the Darwin CLI isn't signed, and causes Gatekeeper to throw warnings when run. This is an issue in our environment as we require all installed binaries to be signed for security.

Sign macOS release

This usually means we're not doing any kind of signing, we should sign the release even if we're not certified by Apple since it makes it more clear to the user that they can ignore the warning. This one requires you to know that you can go to security settings to override it.

Host the wush protocol

... And depend on it from Coder Enterprise

Integration tests for coder-cli

As coder-cli runs against the backend (cdr/enterprise) and depends on its APIs, it would make sense to do some tests of the coder-cli tool itself as part of the main integration checks.

Major features could be tested to ensure they still function, eg:

• sync
• sh
• devurls
  • add
  • update
  • list
  • delete

Bad error message for coder sh offline environment

This is what you get if you coder sh into an environment that was turned off by auto off.

Attempt to reconnect when coder sh disconnects randomly

This is by far my biggest annoyance with coder sh, I get disconnected quite frequently. Could we attempt to reconnect for a second or two before dumping the user back to their terminal?

The issue is made even more annoying by #51 since it still looks like you're in your terminal if you're running tmux or some such.

coder sh prints usage when environment is offline

[willo@mbp ~]$ coder sh dev
2020-08-03 08:30:58 FATAL	run command: read pid message: failed to get reader: received close frame: status = StatusInternalError and reason = ""

We should have a better error message for common cases such as an off environment, a rebuilding environment etc.

Use polling based login flow

The existing localhost redirect solutions breaks when the browser is running on a different network than the command line (like in Coder Enterprise).

Instead, the CLI should publish a token to CE and link the browser to authenticate that token, then store that token locally. This is more secure too since the local token is different than the browser token.

A minimal RFC in the enterprise product should lay out the API in more detail first.

'coder sh' is laggy

The feedback is noticeable enough to be frustrating. It's useable but I wouldn't call it a good UX. It's possible this is due to the implementation of wush lite

Add ability to specify custom devurl prefixes

Expose ability to give custom labels to devurls once https://github.com/cdr/enterprise/issues/4122 is implemented.

Re-run project sync if anything changes during

Right now a bug exists where

1. Project sync starts
2. File system changes are made
3. Project sync ends
4. FS Watch starts

And the filesystem changes are never considered. Instead, we should keep an FS watch all the time, and run the project sync concurrently to it.

Add --version flag

To dump version and last built date

Migrate coder sh from wush to wsep

Host documentation here instead of the README

Support and create a version compatible with Windows

Windows release should be a .exe file (and zipped, not tarred)

Currently the file is just called coder, which must be renamed manually to coder.exe to run on windows. Also, since there are no tar/gz utilities on windows by default the release should be distributed as a .zip.

Windows distro uses HOME env var which is commonly not available

The HOME env var is not always set in a windows environments, causing config-ssh to fail not knowing where to put the private key.

Add connection pooling to OpenSSH entries

See how I did it in cmd

Migrate to urfave/cli

Although https://github.com/cdr/cli offers a minimal wrapper, it's becoming clear that it lacks sufficient features for the growing scope of coder-cli. I propose we migrate to https://github.com/spf13/cobra https://github.com/urfave/cli. Consider a few advantages that we'll gain "out of the box".

• Suggestions when "unknown command" happens
• bash, zsh, fish, and powershell completion generation
• well designed and standard command validation abstractions
• better persistent flag abstractions
• no more interface upgrading nonsense
• the model of a separate type per command is flawed and, given the lack of generics, is very difficult to abstract

Any thoughts / objections? cc @scsmithr @nhooyr @ammario

sync -i loops forever

Reconnect automatically on environment rebuild

Right now we just spin on wush errors

config-ssh command

Use github actions to build mac release binaries

Check that rsync protocol versions match before initiating transfer

So we can give the user a better error message than Exit Status 2

coder sync fails if all dirs leading to sync remoteDir don't already exist

With a new environment, I tried

$ coder sync ~/dev/x11wasm russ-dev1:/home/coder/dev/x11wasm
 2020-06-16 11:01:54 INFO doing initial sync (/home/russtopia/dev/x11wasm -> /home/coder/dev/x11wasm)
 2020-06-16 11:01:58 FATAL sync: rsync: exit status 11

For coder sync to succeed, I had to first do

coder sh russ-dev1
$ mkdir dev
$ exit

Might be nice if the coder sync command was able to essentially do a mkdir -p to ensure the rsync can operate on a brand new tree.