Cannot (sometimes) find runner by label when multiple self-hosted runners are available #13641

New Issue

GiteaMirror · 2025-11-02T10:49:09-06:00

GiteaMirror commented

2025-11-02 10:49:09 -06:00

Originally created by @tomasmusil on GitHub (Oct 26, 2024).

Description

I am running a self-hosted instance of Gitea with two runners both setup as host (no Docker), one is on the same server as my Gitea instance, one is on another server. Both have ":host" at the end of their labels in .runner file.

I have a workflow that deploys to the local machine.

-- ci.yml --

...
jobs:
  build:
    uses: ./.gitea/workflows/reusable/build.yml
    secrets: inherit
...

-- reusable/build.yml --

...
jobs:
  build:
    runs-on: noxlabs_runner
    steps:
      - name: Git clone to work directory
        uses: actions/checkout@v4
...

With this setup, most of the time, not always the job fails with these logs:

On line 1 it seems to incorrectly find the runner by label (action above should have been executed on noxlabs_runner) and then proceeds to run docker images even though it should be run on the host.
If I try to manually rerun the job, after 10-15 attempts it finds the correct runner and completes the workflow.
It persists whether there is only one running job or multiple concurrent ones.
This issue was not happening when I ran multiple concurrent jobs with only one runner configured.
It seems to be connected simply to the fact that I have two active runners, not to the fact that I am running concurrent jobs.

Gitea Version

1.22.3

Can you reproduce the bug on the Gitea demo site?

No

Log Gist

No response

Screenshots

No response

Git Version

2.34.1

Operating System

Both servers are Ubuntu 22.04 x86_64

How are you running Gitea?

Gitea is deployed from oficial release binary, same as both runners.
Gitea and runners are being run with systemd.

Database

PostgreSQL

Originally created by @tomasmusil on GitHub (Oct 26, 2024). ### Description I am running a self-hosted instance of Gitea with two runners both setup as host (no Docker), one is on the same server as my Gitea instance, one is on another server. Both have ":host" at the end of their labels in .runner file. ![image](https://github.com/user-attachments/assets/f6dd23e7-3a10-4b7e-96a0-afcf97958721) I have a workflow that deploys to the local machine. -- ci.yml -- ``` ... jobs: build: uses: ./.gitea/workflows/reusable/build.yml secrets: inherit ... ``` -- reusable/build.yml -- ``` ... jobs: build: runs-on: noxlabs_runner steps: - name: Git clone to work directory uses: actions/checkout@v4 ... ``` With this setup, **most of the time, not always** the job fails with these logs: ![image](https://github.com/user-attachments/assets/a25fedb9-92ce-47dc-8523-222c55e34dbc) On line 1 it seems to incorrectly find the runner by label (action above should have been executed on noxlabs_runner) and then proceeds to run docker images even though it should be run on the host. If I try to manually rerun the job, after 10-15 attempts it finds the correct runner and completes the workflow. It persists whether there is only one running job or multiple concurrent ones. This issue was not happening when I ran multiple concurrent jobs with only one runner configured. It seems to be connected simply to the fact that I have two active runners, not to the fact that I am running concurrent jobs. ### Gitea Version 1.22.3 ### Can you reproduce the bug on the Gitea demo site? No ### Log Gist _No response_ ### Screenshots _No response_ ### Git Version 2.34.1 ### Operating System Both servers are Ubuntu 22.04 x86_64 ### How are you running Gitea? Gitea is deployed from oficial release binary, same as both runners. Gitea and runners are being run with systemd. ### Database PostgreSQL

GiteaMirror added the topic/gitea-actions type/bug labels 2025-11-02 10:49:09 -06:00

GiteaMirror commented

2025-11-02 10:49:10 -06:00

@marza-sergey commented on GitHub (Nov 28, 2024):

I'm having the same issue with host runner running on windows.
Seems like gitea ignores runs-on: statement in reusable workflow.
If I add runs-on: setting in the original workflow that calls reusable workflow, gitea has no problem finding the right runner.

It can be easily replicated by disabling all the runners except the host one.
This is the workflow I that's not working:
pull-checks.yaml

name: Check PR
run-name: ${{ gitea.actor }} checking PR
on: [pull_request]
jobs:
  test-version:
    uses: http://path/to/my/reusable/workflow/test_version.yaml@main
    with:
      package-name: pyf
      ssh-key: ${{ secrets.DEPLOY_SSH_KEY }}
      my-pypi-token: ${{ secrets.PYPI_TOKEN }}

And corrected:

name: Check PR
run-name: ${{ gitea.actor }} checking PR
on: [pull_request]
jobs:
  test-version:
    uses: http://path/to/my/reusable/workflow/test_version.yaml@main
    runs-on: rocky-8
    with:
      package-name: pyf
      ssh-key: ${{ secrets.DEPLOY_SSH_KEY }}
      marza-pypi-token: ${{ secrets.PYPI_TOKEN }}

test_version.yaml

name: Test version
run-name: ${{ gitea.actor }} testing if version is correctly specified
on:
  workflow_call:
    inputs:
      package-name:
        required: true
        type: string
      package-directory:
        required: true
        type: string
        default: .
      my-repository:
        required: true
        type: string
        default: http://path/to/repo
     my-pypi-token:
        required: true
        type: string
      ssh-key:
        type: string
        required: false
        description: Required if has github submodule

jobs:
  check-version:
    runs-on: rocky-8
    steps:
    ...

And configurations for act runner:
config.yaml

log:
  level: debug

runner:
  file: .runner
  capacity: 1
  env_file: .env
  timeout: 3h
  shutdown_timeout: 0s
  insecure: false
  fetch_timeout: 5s
  fetch_interval: 2s
  labels:
    - "windows:host"

cache:
  enabled: true
  dir: ""
  host: ""
  port: 0
  external_server: ""

host:
  workdir_parent:

.runner

{
  "WARNING": "This file is automatically generated by act-runner. Do not edit it manually unless you know what you are doing. Removing this file will cause act runner to re-register as a new runner.",
  "id": 15,
  "uuid": "<redacted>",
  "name": "<redacted>",
  "token": "<redacted>",
  "address": "<redacted>",
  "labels": [
    "windows:host"
  ]
}

@marza-sergey commented on GitHub (Nov 28, 2024): I'm having the same issue with host runner running on windows. Seems like gitea ignores `runs-on:` statement in reusable workflow. If I add `runs-on:` setting in the original workflow that calls reusable workflow, gitea has no problem finding the right runner. It can be easily replicated by disabling all the runners except the host one. This is the workflow I that's not working: `pull-checks.yaml` ```yaml name: Check PR run-name: ${{ gitea.actor }} checking PR on: [pull_request] jobs: test-version: uses: http://path/to/my/reusable/workflow/test_version.yaml@main with: package-name: pyf ssh-key: ${{ secrets.DEPLOY_SSH_KEY }} my-pypi-token: ${{ secrets.PYPI_TOKEN }} ``` And corrected: ```yaml name: Check PR run-name: ${{ gitea.actor }} checking PR on: [pull_request] jobs: test-version: uses: http://path/to/my/reusable/workflow/test_version.yaml@main runs-on: rocky-8 with: package-name: pyf ssh-key: ${{ secrets.DEPLOY_SSH_KEY }} marza-pypi-token: ${{ secrets.PYPI_TOKEN }} ``` `test_version.yaml` ```yaml name: Test version run-name: ${{ gitea.actor }} testing if version is correctly specified on: workflow_call: inputs: package-name: required: true type: string package-directory: required: true type: string default: . my-repository: required: true type: string default: http://path/to/repo my-pypi-token: required: true type: string ssh-key: type: string required: false description: Required if has github submodule jobs: check-version: runs-on: rocky-8 steps: ... ``` And configurations for act runner: `config.yaml` ```yaml log: level: debug runner: file: .runner capacity: 1 env_file: .env timeout: 3h shutdown_timeout: 0s insecure: false fetch_timeout: 5s fetch_interval: 2s labels: - "windows:host" cache: enabled: true dir: "" host: "" port: 0 external_server: "" host: workdir_parent: ``` `.runner` ```json { "WARNING": "This file is automatically generated by act-runner. Do not edit it manually unless you know what you are doing. Removing this file will cause act runner to re-register as a new runner.", "id": 15, "uuid": "<redacted>", "name": "<redacted>", "token": "<redacted>", "address": "<redacted>", "labels": [ "windows:host" ] } ```

GiteaMirror commented

2025-11-02 10:49:10 -06:00

@tomasmusil commented on GitHub (Nov 29, 2024):

If I add runs-on: setting in the original workflow that calls reusable workflow, gitea has no problem finding the right runner.

Thank you for the tip, in the mean time I just moved all code from reusable workflows to the main workflows as it is not that big of a hassle and I get the added benefit of Gitea actually showing the individual steps in UI and not just combining them into "Set up job". With that I verified that it is an issue with reusable workflows specifically, as without them there seems to be no problem selecting the correct runner.

@tomasmusil commented on GitHub (Nov 29, 2024): > If I add `runs-on:` setting in the original workflow that calls reusable workflow, gitea has no problem finding the right runner. Thank you for the tip, in the mean time I just moved all code from reusable workflows to the main workflows as it is not that big of a hassle and I get the added benefit of Gitea actually showing the individual steps in UI and not just combining them into "Set up job". With that I verified that it is an issue with reusable workflows specifically, as without them there seems to be no problem selecting the correct runner.

GiteaMirror commented

2025-11-02 10:49:10 -06:00

@carobme commented on GitHub (Jan 20, 2025):

I can confirm this issue with 1.23.1. According to actionlint, runs-on isn't allowed when reusable workflows are called:

<stdin>:15:5: when a reusable workflow is called with "uses", "runs-on" is not available. only following keys are allowed: "name", "uses", "with", "secrets", "needs", "if", and "permissions" in job "prepare" [syntax-check]
   |
15 |     runs-on: ubuntu-latest
   |     ^~~~~~~~

But Gitea accepts the runs-on, and without it, our labels seem to be ignored. Our runners map ubuntu-latest to a customized image, but when using reusable workflows, the runner uses docker.io/gitea/runner-images:ubuntu-latest instead of our own image.

@carobme commented on GitHub (Jan 20, 2025): I can confirm this issue with 1.23.1. According to actionlint, `runs-on` isn't allowed when reusable workflows are called: ``` <stdin>:15:5: when a reusable workflow is called with "uses", "runs-on" is not available. only following keys are allowed: "name", "uses", "with", "secrets", "needs", "if", and "permissions" in job "prepare" [syntax-check] | 15 | runs-on: ubuntu-latest | ^~~~~~~~ ``` But Gitea accepts the `runs-on`, and without it, our labels seem to be ignored. Our runners map `ubuntu-latest` to a customized image, but when using reusable workflows, the runner uses `docker.io/gitea/runner-images:ubuntu-latest` instead of our own image.

Sign in to join this conversation.

Branches Tags

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: github-starred/gitea#13641