Chains under the load memory consumption raises

[jhutar]

Expected Behavior

I would expect after some time Chains under the load memory consumption would become constant - it will start freeing memory.

Is this expected, or is this some sort of memory leak?

Actual Behavior

This is a memory graph of a Chains signing 10k very simple TaskRuns with that are just printing "hello world" (Pipeline, PipelineRun)

CHains was started around 15:30 and started signing PRs and TRs:

Chains was configured with this:

kubectl patch TektonConfig/config \
            --type merge \
            -p '{"spec":{"chain":{"artifacts.pipelinerun.format": "slsa/v1"}}}'
kubectl patch TektonConfig/config \
            --type merge \
            -p '{"spec":{"chain":{"artifacts.pipelinerun.storage": "tekton"}}}'
kubectl patch TektonConfig/config \
            --type='merge' \
            -p='{"spec":{"chain":{"artifacts.taskrun.format": "slsa/v1"}}}'
kubectl patch TektonConfig/config \
            --type='merge' \
            -p='{"spec":{"chain":{"artifacts.taskrun.storage": "tekton"}}}'

Steps to Reproduce the Problem

1. Run 10k PipelineRuns and wait for all of them to finish
2. Then start the Chains and let it sign PRs and TRs
3. This was automated in this repo with signing-tr-tekton-bigbang scenario

Additional Info

• Kubernetes version:

Cluster is gone already, but it was ROSA OpenShift 4.14.11 with 5 compute nodes AWS EC2 m6a.2xlarge

• Tekton Pipeline version:

• chains-info: v0.19.0
• pipelines-as-code-info: v0.22.4
• pipelines-info: v0.53.2
• triggers-info: v0.25.3
• openshift-pipelines-operator-cluster-operations: v0.69.0
• openshift-pipelines-operator-lifecycle: 1.13.0

Reported this together with https://github.com/tektoncd/pipeline/issues/7691

[concaf]

@jhutar what's the baseline memory usage of chains in this case when there are no workloads on the cluster?

[wlynch]

https://github.com/tektoncd/plumbing/pull/1840 to enable profiling on our dogfooding instance to help debug this, but if you want to enable this on your cluster and share the pprof output, that might be faster 🙏

Steps:

1. Add profiling.enable: "true" totekton-chains/tekton-chains-config-observability
2. kubectl port-forward -n tekton-chains tekton-chains-controller-794dcd9b65-k9f8d 8008 (replace pod name with your own)
3. wget localhost:8008/debug/pprof/heap

[jhutar]

Oh, thank you Billy! Will try to get a pprof output!

Maybe this is similar to https://github.com/tektoncd/pipeline/issues/7691 : memory used by informer to cache everything.