Plugin auto-update broke

[nora-heydecker-axa]

Description

The auto-update infrastructure broke, preventing us from working.

error: unable to find dependency: "error: HTTP status server error (500 Internal Server Error) for url (https://rover.apollo.dev/tar/router/aarch64-apple-darwin/latest-2)

As a workaround I tried --skip-update

error: unable to find dependency: "error: You do not have any 'router' plugins installed in '/node_modules/.pnpm/@apollo+rover@0.35.0/node_modules/@apollo/rover/binary'.
Re-run this command without the --skip-update flag to install the proper plugin.

This does not work either. I tried manually pnpm rover install --plugin router@v1.61.9 and the binary is in that path but it still breaks.

Steps to reproduce

Run rover.

Expected result

Rover works.

Actual result

The 500 and broken --skip-update.

Environment

Run rover info and paste the results here

Rover Info:
Version: 0.35.0
Install Location: <redacted-repo-path>/node_modules/.pnpm/@apollo+rover@0.35.0/node_modules/@apollo/rover/binary/rover-0.35.0
OS: Mac OS 15.5.0 [64-bit]
Shell: /opt/homebrew/bin/fish

[luflow]

We have a similar issue

unable to find dependency: "error: HTTP status client error (400 Bad Request) for url (https://rover.apollo.dev/tar/supergraph/aarch64-apple-darwin/latest-2)

[abernix]

Hey folks — GitHub has been having some sort of problem. It does appear it's on the road to recovery though.

We have a status page up for this right now: https://manage.statuspage.io/pages/r56zbm15x15h/incidents/5h2z0khms34h

And you can see aquaproj/aqua#4068 (comment) has the same problem and support correspondence.

[nora-heydecker-axa]

I see, thank you for the reply, let's hope it recovers soon.
I still think this is a rover bug though, it shouldn't error out in such cases. Auto-update functionality is not critical to the functionality and failing it should warn, not error and make rover unusable. Or at least --skip-update should work.

[hwillson]

We've received confirmation from GitHub that this issue is now resolved (status page).

@nora-heydecker-axa regarding:

I still think this is a rover bug though, it shouldn't error out in such cases. Auto-update functionality is not critical to the functionality and failing it should warn, not error and make rover unusable. Or at least --skip-update should work.

This is great feedback - we'll discuss this with the team. Thanks!

[andrewmclagan]

We are also experiencing this issue. We have had this issue come and go over the past year. We have attempted to mitigate this (as others have here as well) by "warming up" the rover CLI:

1. Preinstalling federation plugin
2. Adding --skip-update-check flag to commands we run

This flag seems to be ignored:

supergraph config loaded successfully
error: Error when updating Federation Version:
unable to find dependency: "error: error sending request for url (https://rover.apollo.dev/tar/supergraph/x86_64-unknown-linux-gnu/latest-2)

Caused by:
    0: client error (Connect)
    1: Connection reset by peer (os error 104)
    2: Connection reset by peer (os error 104)
"

Caused by:
    unable to find dependency: "error: error sending request for url (https://rover.apollo.dev/tar/supergraph/x86_64-unknown-linux-gnu/latest-2)
    
    Caused by:
        0: client error (Connect)
        1: Connection reset by peer (os error 104)
        2: Connection reset by peer (os error 104)
    "

The above is triggered within the command:

pnpm rover graph check --skip-update-check [...redacted...]

where the following is run before everything:

- name: 'Rover CLI plugin warmup'
  uses: nick-invision/retry@v3
  with:
    timeout_minutes: 3
    max_attempts: 3
    command: pnpm -w exec rover install --elv2-license=accept --plugin=supergraph@=2.11.3

historically we have seen connection errors in the above so we added retries.