Workers and pipelines

Introduction

• Pipelines are run on workers, i.e. nodes with installed Buddy On-Promises.
• The server where Buddy On-Premises is first installed is called the primary worker.
• The primary worker is treated as any other worker, i.e. can be tagged and locked.
• The pipeline will run on a particular worker if its tags match the tag on the worker.

Tag assignment mechanism

A pipeline can have no tags or any number of tags. A worker can have no tags, or one tag maximum.

A. If the pipeline doesn't have any tags > Buddy looks up all unlocked workers with no tags
B. If the pipeline has tags > Buddy looks up all unlocked workers with at least one matching tag

Pipeline assignment rules

A pipeline is assigned to a new worker if one of the following conditions occurs:

1. The pipeline is run for the first time.
2. The average load on the current node exceeds 100% per CPU core over the last 5 minutes.
3. The tag of the worker doesn't match the tags of the pipeline.
4. The current worker is not responding (is offline, is in different version, etc.).
5. The pipeline's cache is cleared.

Example use case

Let's assume you added a worker for time-consuming tasks like integration tests:

Workers tab in On-Premises Admin

To match a pipeline with that worker, you need to select its tag in the pipeline's settings – or immediately when adding a new pipeline.

Below we have a pipeline with several tags. One of them matches the worker, and so it appears on the list in the pipeline settings:

Worker tags in pipeline settings

Filesystem, static files and cache

1. A pipeline is fixed to a specific worker together with its cache and filesystem, as long as no reassignment rule is fulfilled.

2. Whenever a pipeline is switched to another worker, the contents of its cache are cleared. As a result, the first execution on the new worker can take a bit longer, as it needs to redownload the dependencies and regenerate the artifacts.

3. The exception are static files manually uploaded to the filesystem. Static files are stored in the main instance of Buddy On-Premises and are automatically downloaded on the new worker at the time of the first pipeline's run.

Draining

Draining means that the worker is finishing the current pipeline run and does not begin any new, including enqueued. Use this feature if you want to gracefully shut down your worker before update or maintenance.

You can mark the worker to drain by checking the flag in the worker settings:

Draining option in worker settings

Alternatively, you do it by running a command on the machine where the worker is installed:

buddy update-worker --drain
$

To restore the worker to the previous state, run:

buddy update-worker --flood
$

Switching the worker to the draining mode will produce a corresponding label next to it:

Worker in draining mode

Troubleshooting

In case of errors with assigning the pipeline to a worker, you can check the Prepare Environment step of the pipeline's execution logs for details.