Checkout fails on early EOF error. [fetch-pack: invalid index-pack output]

[tamir-kucherov-wdc]

Hello,

During the checkout process, we usually see this error message:
2023-06-09T19:40:18.4811181Z fetch-pack: unexpected disconnect while reading sideband packet
2023-06-09T19:40:18.4849629Z ##[error]fatal: early EOF
2023-06-09T19:40:18.4859845Z ##[error]fatal: fetch-pack: invalid index-pack output
2023-06-09T19:40:18.4902345Z The process 'C:\Program Files\Git\cmd\git.exe' failed with exit code 128
2023-06-09T19:40:18.4903851Z Waiting 15 seconds before trying again

Can you tell what is the reason? How to resolve this?
The connection is stable, and we also tried to run the flow with parameter fetch-depth: 5, but still get the same error.

We tried to run it on our local runner, and it also fails with same error. When we run it manually using git clone on the same machine, all works fine.

Thanks! Tamir

[tamir-kucherov-wdc]

@ericsciple @thboop @TingluoHuang can you take a look? Thanks!

[heathhenley]

I'm seeing this on a windows running with a somewhat large repo with some submodules and lfs files to pull - still actively working on the action but the file looks something like this:

name: Run-Python-Tests
run-name: Running-Python-Tests
on:
  pull_request:
    branches:
      - 'master'
      - '[1-9]+.[0-9]+.[0-9]+'
jobs:
  Run-Python-Tests:
    runs-on: windows-latest-l
    steps:
      - uses: actions/checkout@v4
        with:
          submodules: 'recursive'
          ssh-key: ${{ secrets.SSH_KEY }}
          lfs: true

I also noticed that it seems to not cause a failure all the time, sometimes after the first or second retry it's working - so I assume we're hitting the edge of some threshold to timing out or something? Anything I could adjust?

UPDATE
Adding http.postbuffer update seems like it might have made it more consistent:
#748 (comment)

Another UPDATE:
I tried any git config option suggestions I could find in addition to that buffer update, nothing helped consistently. In the end, I switched to use a PAT and https instead of ssh, and 🤞- it's now consistently working.