CML initialization is broken on GH in container mode

[shcheklein]

Basic CML workflow like this is broken.

I'm getting:

{"level":"error","message":"Command failed: git config --get remote.origin.url","output":[null,{"data":[],"type":"Buffer"},{"data":[],"type":"Buffer"}],"pid":124,"signal":null,"stack":"Error: Command failed: git config --get remote.origin.url\n    at checkExecSyncError (node:child_process:828:11)\n    at execSync (node:child_process:902:15)\n    at gitRemoteUrl (/usr/lib/node_modules/@dvcorg/cml/src/cml.js:29:15)\n    at new CML (/usr/lib/node_modules/@dvcorg/cml/src/cml.js:76:44)\n    at Object.exports.handler (/usr/lib/node_modules/@dvcorg/cml/bin/cml/ci.js:10:15)\n    at /usr/lib/node_modules/@dvcorg/cml/node_modules/yargs/build/index.cjs:1:9054\n    at j (/usr/lib/node_modules/@dvcorg/cml/node_modules/yargs/build/index.cjs:1:4931)\n    at M.applyMiddlewareAndGetResult (/usr/lib/node_modules/@dvcorg/cml/node_modules/yargs/build/index.cjs:1:9023)\n    at M.runCommand (/usr/lib/node_modules/@dvcorg/cml/node_modules/yargs/build/index.cjs:1:7206)\n    at Xt.[runYargsParserAndExecuteCommands] (/usr/lib/node_modules/@dvcorg/cml/node_modules/yargs/build/index.cjs:1:57164)\n    at Xt.parse (/usr/lib/node_modules/@dvcorg/cml/node_modules/yargs/build/index.cjs:1:39275)\n    at Object.<anonymous> (/usr/lib/node_modules/@dvcorg/cml/bin/cml.js:90:4)\n    at Module._compile (node:internal/modules/cjs/loader:1103:14)\n    at Object.Module._extensions..js (node:internal/modules/cjs/loader:1157:10)\n    at Module.load (node:internal/modules/cjs/loader:981:32)\n    at Function.Module._load (node:internal/modules/cjs/loader:822:12)","status":1,"stderr":{"data":[],"type":"Buffer"},"stdout":{"data":[],"type":"Buffer"}}

Command git config --get remote.origin.url that is triggered in CML ctr fails.

Most likely this is due to some breaking changes in action checkout (v2 and v3 both affected). They have recently had to upgrade those due to recent update of Git on GH https://github.blog/2022-04-12-git-security-vulnerability-announced/

(more info for the reference is here: https://iterativeai.slack.com/archives/C01900GSB4J/p1650245948446699 (internal link))

[dacbd]

minimal example:

name: test cml ci
on:
  push:
  workflow_dispatch:

jobs:
  test-v2:
    runs-on: ubuntu-latest
    container: ghcr.io/iterative/cml:latest
    steps:
      - uses: actions/checkout@v2
      - run: cml ci
  test-v3:
    runs-on: ubuntu-latest
    container: ghcr.io/iterative/cml:latest
    steps:
      - uses: actions/checkout@v3
      - run: cml ci
  test-setup-v2:
    runs-on: ubuntu-latest
    steps:
      - uses: iterative/setup-cml@v1
      - uses: actions/checkout@v2
      - run: cml ci
  test-setup-v3:
    runs-on: ubuntu-latest
    steps:
      - uses: iterative/setup-cml@v1
      - uses: actions/checkout@v3
      - run: cml ci

https://github.com/dacbd/err/actions/runs/2182120620

test-v2 / test-v3 fail with an error:

{
  "level": "error",
  "message": "Command failed: git config --get remote.origin.url",
  "output": [
    null,
    {
      "data": [],
      "type": "Buffer"
    },
    {
      "data": [],
      "type": "Buffer"
    }
  ],
  "pid": 126,
  "signal": null,
  "stack": "Error: Command failed: git config --get remote.origin.url\n    at checkExecSyncError (node:child_process:828:11)\n    at execSync (node:child_process:902:15)\n    at gitRemoteUrl (/usr/lib/node_modules/@dvcorg/cml/src/cml.js:29:15)\n    at new CML (/usr/lib/node_modules/@dvcorg/cml/src/cml.js:76:44)\n    at Object.exports.handler (/usr/lib/node_modules/@dvcorg/cml/bin/cml/ci.js:10:15)\n    at /usr/lib/node_modules/@dvcorg/cml/node_modules/yargs/build/index.cjs:1:9054\n    at j (/usr/lib/node_modules/@dvcorg/cml/node_modules/yargs/build/index.cjs:1:4931)\n    at M.applyMiddlewareAndGetResult (/usr/lib/node_modules/@dvcorg/cml/node_modules/yargs/build/index.cjs:1:9023)\n    at M.runCommand (/usr/lib/node_modules/@dvcorg/cml/node_modules/yargs/build/index.cjs:1:7206)\n    at Xt.[runYargsParserAndExecuteCommands] (/usr/lib/node_modules/@dvcorg/cml/node_modules/yargs/build/index.cjs:1:57164)\n    at Xt.parse (/usr/lib/node_modules/@dvcorg/cml/node_modules/yargs/build/index.cjs:1:39275)\n    at Object.<anonymous> (/usr/lib/node_modules/@dvcorg/cml/bin/cml.js:90:4)\n    at Module._compile (node:internal/modules/cjs/loader:1103:14)\n    at Object.Module._extensions..js (node:internal/modules/cjs/loader:1157:10)\n    at Module.load (node:internal/modules/cjs/loader:981:32)\n    at Function.Module._load (node:internal/modules/cjs/loader:822:12)",
  "status": 1,
  "stderr": {
    "data": [],
    "type": "Buffer"
  },
  "stdout": {
    "data": [],
    "type": "Buffer"
  }
}

when I expect them to complain with same token not found as the test-setup-v2/v3

[casperdcl]

Also (/CC @iterative/cml) think we recommend actions/checkout before setup-cml, right?

[DavidGOrtega]

Also (/CC @iterative/cml) think we recommend actions/checkout before setup-cml, right?

Yes, actions/checkout should be prior to anything else. It's what setup the repo within the runner

[0x2b3bfa0]

we recommend actions/checkout before setup-cml, right?

Order doesn't matter, unless iterative/setup-cml does something that it's not supposed (?) to do, like running cml ci on its own.

[0x2b3bfa0]

Infimum example

on: workflow_dispatch
jobs:
  succeeds:
    runs-on: ubuntu-latest
    steps:
      - uses: iterative/setup-cml@v1
      - uses: actions/checkout@v3
      - run: cml ci
        env:
          GITHUB_TOKEN: ${{ github.token }}
  fails:
    runs-on: ubuntu-latest
    container: ghcr.io/iterative/cml
    steps:
      - uses: actions/checkout@v3
      - run: cml ci
        env:
          GITHUB_TOKEN: ${{ github.token }}

[0x2b3bfa0]

Fix

Running git config --global --add safe.directory "$PWD" as part of cml ci is enough to fix this issue. See actions/checkout#766 for more information.

on: push
jobs:
  fix:
    runs-on: ubuntu-latest
    container: ghcr.io/iterative/cml
    steps:
      - uses: actions/checkout@v3
      - run: git config --global --add safe.directory "$PWD"
      - run: cml ci
        env:
          GITHUB_TOKEN: ${{ github.token }}

[0x2b3bfa0]

See Also

• Enforce safe directory actions/checkout#762
• fatal: unsafe repository (REPO is owned by someone else) with ubuntu 20.04 container actions/checkout#760

[0x2b3bfa0]

@iterative/cml & company, what option do you prefer?

1. Add CWD to Git's safe.directory with cml ci #974 and recommend running cml ci as part of every workflow
2. Add CWD to Git's safe.directory with cml ci #974 and run that code automatically for all the cml commands (i.e. implicit cml ci)
3. Pin to git<2.35.2 on all our container images 🙃

[shcheklein]

@0x2b3bfa0 are there other way to get the same result as git config --get remote.origin.url in JS code? that's the only thing that is failing, right? and it looks the only thing that we are using that for is to determine GH/GL/BB? maybe there are also other ways to do thi?

[0x2b3bfa0]

@0x2b3bfa0 are there other way to get the same result as git config --get remote.origin.url in JS code?

Yes, see the parse-git-config package for example, but it's not enough to replace all our other interactions with Git.

that's the only thing that is failing, right?

It's rather the first thing that is failing. Any git command we run will produce the same error:

fatal: unsafe repository ('/path/to/repository' is owned by someone else)
To add an exception for this directory, call:

        git config --global --add safe.directory /path/to/repository

it looks the only thing that we are using that for is to determine GH/GL/BB?

I'd wish! 😅 We use many other git commands for various purposes.

[shcheklein]

It's rather the first thing that is failing

Got it. How does CI works in general now if any git is failing? Does it mean that it'll be fixed by GH soon? Or is it specific to the way action checkout does things?

Pin to git<2.35.2 on all our container images

Seems unreliable - we'll have to deal with this in the non-container mode now or soon?

---

#974

One more option - can we make it part of CML init?