Informer Cache Optimization

This page describes the informer cache transform functions that Pipelines-as-Code uses to reduce the memory footprint of its watcher controller. These transforms are applied automatically and require no configuration.

Background

The PAC watcher controller maintains in-memory caches (via Kubernetes informers) for Repository and PipelineRun objects. In clusters with many repositories or long-running PipelineRuns, these caches can consume a significant amount of memory because each cached object carries fields that the watcher never reads — managedFields, last-applied-configuration annotations, embedded specs, and status details.

Pipelines-as-Code registers TransformFunc callbacks on each informer. A TransformFunc is called on every object before it is stored in the cache, allowing unnecessary fields to be stripped while the object is still in use.

This is the same approach used by the Tekton Pipelines controller to reduce its own cache memory usage.

What Gets Stripped

Repository Objects

Benchmark result: ~89% JSON size reduction per Repository object.

PipelineRun Objects

Fields preserved (required for reconciliation)

Fields stripped

When the reconciler needs the full PipelineRun (for example, during postFinalStatus or GetStatusFromTaskStatusOrFromAsking), it fetches the complete object directly from the API server.

Benchmark result: ~94% JSON size reduction per PipelineRun object.

Memory Impact

Benchmarks using realistic object sizes from production clusters show the following per-object savings:

For a cluster with 1000 Repositories and 700 PipelineRuns, the estimated watcher cache reduction is approximately 12 MB.

Graceful Degradation

The transform functions are designed to degrade gracefully:

• If an object is wrapped in a DeletedFinalStateUnknown tombstone (which happens when the watcher misses a delete event), the transform unwraps it, strips the inner object, and re-wraps it.
• If the transform receives an unexpected type, it returns the object unmodified rather than returning an error.
• If any error occurs during transformation, the original object is returned unchanged so that the informer cache continues to function.

Developer Notes

If you add new reconciliation logic that reads a field from cached objects (via listers), you must verify that the field is not stripped by these transforms. Fields stripped from cached objects will be nil or empty even though they exist in etcd.

If you need a stripped field, fetch the full object via the API client instead of the lister:

// Don't do this — spec.params is stripped from the cache:
pr, _ := pipelineRunLister.PipelineRuns(ns).Get(name)
params := pr.Spec.Params // always nil!

// Do this instead — fetch from the API server:
pr, _ := tektonClient.TektonV1().PipelineRuns(ns).Get(ctx, name, metav1.GetOptions{})
params := pr.Spec.Params // full object from etcd

The transform functions and their benchmarks live in pkg/informer/transform/.

To run the benchmarks yourself:

go test -bench=. -benchmem -v ./pkg/informer/transform/

To see the size reduction report:

go test -v -run 'TestMeasure.*TransformSavings' ./pkg/informer/transform/