Using self-hosted runners in a workflow

For information on creating custom and default labels, see "Using labels with self-hosted runners."

Using self-hosted runners in a workflow

Labels allow you to send workflow jobs to specific types of self-hosted runners, based on their shared characteristics. For example, if your job requires a particular hardware component or software package, you can assign a custom label to a runner and then configure your job to only execute on runners with that label.

Para especificar um executor auto-hospedado para o seu trabalho, configure runs-on no seu arquivo de fluxo de trabalho com etiquetas de executores auto-hospedados.

Todos os executores auto-hospedados têm a etiqueta auto-hospedado. O uso apenas esta etiqueta selecionará qualquer executor auto-hospedado. Para selecionar executores que atendem a certos critérios, como sistema operacional ou arquitetura, recomendamos fornecer uma matriz de etiquetas que começam com auto-hospedado (isso deve ser listado primeiro) e, em seguida, inclui etiquetas adicionais, conforme necessário. Ao especificar uma matriz de etiquetas, os trabalhos serão colocados na fila em executores que tenham todas as etiquetas especificadas.

Embora a etiqueta de auto-hospedado não seja obrigatória, recomendamos especificá-lo quando estiver usando executores auto-hospedados para garantir que seu trabalho não especifique nenhum executor atual ou futuro de GitHub.

For more information, see "Workflow syntax for GitHub Actions."

Using default labels to route jobs

A self-hosted runner automatically receives certain labels when it is added to GitHub Actions. These are used to indicate its operating system and hardware platform:

• self-hosted: Default label applied to all self-hosted runners.
• linux, windows, or macOS: Applied depending on operating system.
• x64, ARM, or ARM64: Applied depending on hardware architecture.

You can use your workflow's YAML to send jobs to a combination of these labels. In this example, a self-hosted runner that matches all three labels will be eligible to run the job:

runs-on: [self-hosted, linux, ARM64]

• self-hosted - Run this job on a self-hosted runner.
• linux - Only use a Linux-based runner.
• ARM64 - Only use a runner based on ARM64 hardware.

The default labels are fixed and cannot be changed or removed. Consider using custom labels if you need more control over job routing.

Using custom labels to route jobs

You can create custom labels and assign them to your self-hosted runners at any time. Custom labels let you send jobs to particular types of self-hosted runners, based on how they're labeled.

For example, if you have a job that requires a specific type of graphics hardware, you can create a custom label called gpu and assign it to the runners that have the hardware installed. A self-hosted runner that matches all the assigned labels will then be eligible to run the job.

This example shows a job that combines default and custom labels:

runs-on: [self-hosted, linux, x64, gpu]

• self-hosted - Run this job on a self-hosted runner.
• linux - Only use a Linux-based runner.
• x64 - Only use a runner based on x64 hardware.
• gpu - This custom label has been manually assigned to self-hosted runners with the GPU hardware installed.

These labels operate cumulatively, so a self-hosted runner’s labels must match all four to be eligible to process the job.

Routing precedence for self-hosted runners

When routing a job to a self-hosted runner, GitHub looks for a runner that matches the job's runs-on labels:

1. GitHub first searches for a runner at the repository level, then at the organization level, then at the enterprise level.
2. The job is then sent to the first matching runner that is online and idle.
   • If all matching online runners are busy, the job will queue at the level with the highest number of matching online runners.
   • If all matching runners are offline, the job will queue at the level with the highest number of matching offline runners.
   • If there are no matching runners at any level, the job will fail.
   • If the job remains queued for more than 24 hours, the job will fail.