Allow the coder agent to tell modules where they can place data, binaries, etc #11131

mafredri · 2023-12-11T15:12:12Z

This is an alternate path to achieving PATH manipulations as discussed in: #10166

The premise is, let the agent decide where a module can store its files (binaries, data, settings, etc).

Let's say the agent discerns that $HOME is a good place to put files, it can create the appropriate folders/permissions and run module script(s) with the following env variables:

CODER_MODULE_DATA_PATH=~/.local/share/coder_module/${MODULE_ID}
CODER_MODULE_BIN_PATH=~/.local/share/coder_module/${MODULE_ID}/bin

Or let's say $HOME isn't writable, we can usually default to the agent directory often located in /tmp (granted, this is ephemeral storage, so a warning may be prudent).

CODER_MODULE_DATA_PATH=/tmp/coder.xxcxcv/module/${MODULE_ID}
CODER_MODULE_BIN_PATH=/tmp/coder.xxcxcv/module/${MODULE_ID}/bin

As long as we can create an association between coder module (name/id) and whatever coder_script it has introduced, we should be able to do this fairly easily in the agent.

We can either use a per-module /bin or a combine bin dir, e.g.

CODER_MODULE_BIN_PATH=~/.local/share/coder_module/bin

Lastly, the agent would make sure to prepend (or append) the module bin dir(s) to PATH for use in scripts, SSH sessions, etc.

The text was updated successfully, but these errors were encountered:

bpmct · 2023-12-12T18:47:26Z

We should avoid using $HOME for the bin but @kylecarbs, @stirby, and I are onboard!

matifali · 2023-12-13T11:54:14Z

@bpmct why not using $HOME? We assume that most of the workspaces will have a persistent $HOME, and I think it is a good idea to store all binaries installed by modules in CODER_MODULE_BIN_PATH=~/.local/share/coder_module/bin. This will allow us to make modules intelligent by checking if the binary is already available.

mafredri · 2023-12-13T12:06:50Z

@bpmct why not using $HOME? We assume that most of the workspaces will have a persistent $HOME, and I think it is a good idea to store all binaries installed by modules in CODER_MODULE_BIN_PATH=~/.local/share/coder_module/bin. This will allow us to make modules intelligent by checking if the binary is already available.

Agreed, storing data in persistent storage can help speed up workspace startup times and reduce flakes, e.g. waiting on curl https://some-awesome-tool, or network hiccups, etc.

I'm not a fan of too many options, but we could have both and a module can make its decision. Like CODER_MODULE_EPHEMERAL_* and CODER_MODULE_PERSISTENT_*.

bpmct · 2024-02-05T15:25:44Z

My concern about $HOME is that a new version of a module may have another version of the script/binary and then it becomes up to the module to look at the checksum, compare, and copy the new one. Or a symlink, but then the binary/script is being re-downloaded

Plus, modules already have to download stuff every time into a non-home dir (e.g. code-server into bin). I agree I'd be good to have some form of caching for all of these utils but I worry about the trade-off "configuration/script drift"

Don't consider this blocking, just one thing to consider. I would be very happy to have a module bin and would happily write a few extra lines to compare a script in $HOME over a janky way to override PATH

The agent is extended with a `--script-data-dir` flag, defaulting to the OS temp dir. This dir is used for storing `coder-script/bin` and `coder-script/[script uuid]`. The former is a place for all scripts to place executable binaries that will be available by other scripts, SSH sessions, etc. The latter is a place for the script to store files. Since we default to OS temp dir, files are ephemeral by default. In the future, we may consider adding new env vars or changing the default storage location. Workspace startup speed could potentially benefit from scripts being able to skip steps that require downloading software. We may also extend this with more env variables (e.g. persistent storage in HOME). Fixes #11131

The agent is extended with a `--script-data-dir` flag, defaulting to the OS temp dir. This dir is used for storing `coder-script-data/bin` and `coder-script/[script uuid]`. The former is a place for all scripts to place executable binaries that will be available by other scripts, SSH sessions, etc. The latter is a place for the script to store files. Since we default to OS temp dir, files are ephemeral by default. In the future, we may consider adding new env vars or changing the default storage location. Workspace startup speed could potentially benefit from scripts being able to skip steps that require downloading software. We may also extend this with more env variables (e.g. persistent storage in HOME). Fixes #11131

mafredri · 2024-02-20T11:33:33Z

@bpmct @matifali

For v1, I only implemented two env variables, $CODER_SCRIPT_DATA_DIR (/tmp/coder-script-data/8d5e8afb-86ec-4a01-ad07-8f53613782ac) and $CODER_SCRIPT_BIN_DIR (/tmp/coder-script-data/bin).

For now, these are ephemeral via /tmp, but it would be safest to consider that they may not be empty when writing modules.

I also decided that all modules share one bin path because otherwise we have 1-2 problems:

Module priorities, one /bin dir will always have higher priority than the other, and if two modules install a command with the same name, that's a problem
The $PATH variable grows unwieldy with, say, 10 modules or more

(The /tmp directory can be changed by setting coder agent --script-data-dir /some/other/path.)

Here's an example of a compatibility layer we can use in our scripts:

#!/bin/sh

BIN_DIR=/usr/local/bin
if [ -n "${CODER_SCRIPT_BIN_DIR}" ]; then
    BIN_DIR="${CODER_SCRIPT_BIN_DIR}"
fi

install_bin() {
	name=$(basename "${1}")
    echo "Installing ${name} to ${BIN_DIR}..."
    cp -f "${1}" "${BIN_DIR}/" || sudo cp -f "${1}" "${BIN_DIR}/"
    chmod +x "${BIN_DIR}/${name}"
}

install_bin vault-v1.0.0/bin/vault

matifali · 2024-02-20T17:38:45Z

The next step would be refactor modules to make use of this new capabilities.
It's certainly a step for a better UX.

bpmct · 2024-04-11T17:45:40Z

@mafredri Is there a way to programatically get the bin dir? For example, on Windows it may be different.

bpmct · 2024-04-11T17:45:53Z

I guess all scripts would break on windows unless they are windows-specific

cdr-bot bot added the feature label Dec 11, 2023

mafredri mentioned this issue Jan 2, 2024

feat(jfrog): add JFrog vscode extension, CLI completion and docker support coder/modules#115

Merged

1 task

mafredri mentioned this issue Feb 5, 2024

allow adding env variables via CLI #11338

Closed

mafredri self-assigned this Feb 5, 2024

mafredri mentioned this issue Feb 19, 2024

refactor(agent/agentssh): move envs to agent and add agentssh config struct #12204

Merged

mafredri mentioned this issue Feb 19, 2024

feat(agent): add script data dir for binaries and files #12205

Merged

mafredri closed this as completed in #12205 Feb 20, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Allow the coder agent to tell modules where they can place data, binaries, etc #11131

Allow the coder agent to tell modules where they can place data, binaries, etc #11131

mafredri commented Dec 11, 2023 •

edited

Loading

bpmct commented Dec 12, 2023 •

edited

Loading

matifali commented Dec 13, 2023

mafredri commented Dec 13, 2023

bpmct commented Feb 5, 2024 •

edited

Loading

mafredri commented Feb 20, 2024 •

edited

Loading

matifali commented Feb 20, 2024

bpmct commented Apr 11, 2024

bpmct commented Apr 11, 2024

Allow the coder agent to tell modules where they can place data, binaries, etc #11131

Allow the coder agent to tell modules where they can place data, binaries, etc #11131

Comments

mafredri commented Dec 11, 2023 • edited Loading

bpmct commented Dec 12, 2023 • edited Loading

matifali commented Dec 13, 2023

mafredri commented Dec 13, 2023

bpmct commented Feb 5, 2024 • edited Loading

mafredri commented Feb 20, 2024 • edited Loading

matifali commented Feb 20, 2024

bpmct commented Apr 11, 2024

bpmct commented Apr 11, 2024

mafredri commented Dec 11, 2023 •

edited

Loading

bpmct commented Dec 12, 2023 •

edited

Loading

bpmct commented Feb 5, 2024 •

edited

Loading

mafredri commented Feb 20, 2024 •

edited

Loading