prombench: Added support for --bench.version flag.

See prometheus/proposals#41 for rationale. Signed-off-by: bwplotka <bwplotka@gmail.com>
prometheus · Dec 17, 2024 · f979141 · f979141
1 parent c037ee0
commit f979141
Show file tree

Hide file tree

Showing 8 changed files with 217 additions and 120 deletions.
diff --git a/prombench/Makefile b/prombench/Makefile
@@ -1,12 +1,5 @@
-INFRA_CMD        ?= ../infra/infra
-
-PROVIDER 		 ?= gke
-
-.PHONY: deploy clean
-deploy: node_create resource_apply
-# GCP sometimes takes longer than 30 tries when trying to delete nodes
-# if k8s resources are not already cleared
-clean: resource_delete node_delete
+INFRA_CMD  ?= ../infra/infra
+PROVIDER   ?= gke
 
 cluster_create:
 	${INFRA_CMD} ${PROVIDER} cluster create -a ${AUTH_FILE} \
@@ -37,50 +30,96 @@ cluster_delete:
 		-v CLUSTER_NAME:${CLUSTER_NAME} -v PR_NUMBER:${PR_NUMBER} \
 		-f manifests/cluster_${PROVIDER}.yaml
 
+# /prombench <...> --bench.directory
+BENCHMARK_DIRECTORY := $(if $(BENCHMARK_DIRECTORY),$(BENCHMARK_DIRECTORY),manifests/prombench)
+# /prombench <...> --bench.version
+BENCHMARK_VERSION := $(if $(BENCHMARK_VERSION),$(BENCHMARK_VERSION),master)
+PROMBENCH_GIT_REPOSITORY ?= git@github.com:prometheus/test-infra.git
+PROMBENCH_DIR ?= .
+
+# maybe_pull_custom_version allows custom benchmarking as designed in
+# https://github.com/prometheus/proposals/pull/41. It allows calling
+# /prombench <release> --bench.version=<@commit or branch> which will cause
+# prombench GH job on Prometheus repo to call infra CLI with the non-master BENCHMARK_VERSION.
+# In such a case we pull a prombench repository for the given branch or commit version
+# and adjust PROMBENCH_DIR. As a result `make deploy` and `make clean` jobs
+# will apply /manifests/ apply custom manifests or even node pools.
+.PHONE: maybe_pull_custom_version
+maybe_pull_custom_version:
+ifeq (${BENCHMARK_VERSION},master)
+	@echo ">> Using standard benchmark configuration, from the docker image"
+else
+	@echo ">> Git pulling custom benchmark configuration from the ${BENCHMARK_VERSION}"
+	@$(eval $@_TMP_DIR=$(shell mktemp -d -t "prombench"))
+	cd ${$@_TMP_DIR} && git clone ${PROMBENCH_GIT_REPOSITORY}
+ifeq ($(subst @,,${BENCHMARK_VERSION}),${BENCHMARK_VERSION})
+	@echo ">> --bench.version is a branch, reseting to origin/${BENCHMARK_VERSION}"
+	cd ${$@_TMP_DIR}/test-infra && git reset --hard origin/${BENCHMARK_VERSION}
+else
+	@echo ">> --bench.version is a commit SHA, reseting to $(subst @,,${BENCHMARK_VERSION})"
+	cd ${$@_TMP_DIR}/test-infra && git reset --hard $(subst @,,${BENCHMARK_VERSION})
+endif
+	$(eval PROMBENCH_DIR=${$@_TMP_DIR}/test-infra/prombench)
+endif
+	@echo ">> Using following files in ${PROMBENCH_DIR}/${BENCHMARK_DIRECTORY}"
+	@ls -lR ${PROMBENCH_DIR}/${BENCHMARK_DIRECTORY}
+
+.PHONE: clean_tmp_dir
+clean_tmp_dir: # Clean after maybe_pull_custom_version
+	[ -z ${maybe_pull_custom_version_TMP_DIR} ] || rm -rf ${maybe_pull_custom_version_TMP_DIR}
+
+.PHONY: deploy
+deploy: maybe_pull_custom_version node_create resource_apply clean_tmp_dir
+
+.PHONE: clean
+# GCP sometimes takes longer than 30 tries when trying to delete nodes
+# if k8s resources are not already cleared
+clean: maybe_pull_custom_version resource_delete node_delete clean_tmp_dir
+
 node_create:
 	${INFRA_CMD} ${PROVIDER} nodes create -a ${AUTH_FILE} \
 		-v ZONE:${ZONE} -v GKE_PROJECT_ID:${GKE_PROJECT_ID} \
 		-v EKS_WORKER_ROLE_ARN:${EKS_WORKER_ROLE_ARN} -v EKS_CLUSTER_ROLE_ARN:${EKS_CLUSTER_ROLE_ARN} \
 		-v EKS_SUBNET_IDS:${EKS_SUBNET_IDS} \
 		-v CLUSTER_NAME:${CLUSTER_NAME} -v PR_NUMBER:${PR_NUMBER} \
-		-f manifests/prombench/nodes_${PROVIDER}.yaml
+		-f ${PROMBENCH_DIR}/${BENCHMARK_DIRECTORY}/nodes_${PROVIDER}.yaml
 
 resource_apply:
 	$(INFRA_CMD) ${PROVIDER} resource apply -a ${AUTH_FILE} \
 		-v ZONE:${ZONE} -v GKE_PROJECT_ID:${GKE_PROJECT_ID} \
 		-v CLUSTER_NAME:${CLUSTER_NAME} \
 		-v PR_NUMBER:${PR_NUMBER} -v RELEASE:${RELEASE} -v DOMAIN_NAME:${DOMAIN_NAME} \
 		-v GITHUB_ORG:${GITHUB_ORG} -v GITHUB_REPO:${GITHUB_REPO} \
-		-f manifests/prombench/benchmark
+		-f ${PROMBENCH_DIR}/${BENCHMARK_DIRECTORY}/benchmark
 
 # Required because namespace and cluster-role are not part of the created nodes
 resource_delete:
 	$(INFRA_CMD) ${PROVIDER} resource delete -a ${AUTH_FILE} \
 		-v ZONE:${ZONE} -v GKE_PROJECT_ID:${GKE_PROJECT_ID} \
 		-v CLUSTER_NAME:${CLUSTER_NAME} -v PR_NUMBER:${PR_NUMBER} \
-		-f manifests/prombench/benchmark/1c_cluster-role-binding.yaml \
-		-f manifests/prombench/benchmark/1a_namespace.yaml
+		-f ${PROMBENCH_DIR}/${BENCHMARK_DIRECTORY}/benchmark/1c_cluster-role-binding.yaml \
+		-f ${PROMBENCH_DIR}/${BENCHMARK_DIRECTORY}/benchmark/1a_namespace.yaml
 
 node_delete:
 	$(INFRA_CMD) ${PROVIDER} nodes delete -a ${AUTH_FILE} \
 		-v ZONE:${ZONE} -v GKE_PROJECT_ID:${GKE_PROJECT_ID} \
 		-v EKS_WORKER_ROLE_ARN:${EKS_WORKER_ROLE_ARN} -v EKS_CLUSTER_ROLE_ARN:${EKS_CLUSTER_ROLE_ARN} \
 		-v EKS_SUBNET_IDS:${EKS_SUBNET_IDS} \
 		-v CLUSTER_NAME:${CLUSTER_NAME} -v PR_NUMBER:${PR_NUMBER} \
-		-f manifests/prombench/nodes_${PROVIDER}.yaml
+		-f ${PROMBENCH_DIR}/${BENCHMARK_DIRECTORY}/nodes_${PROVIDER}.yaml
 
 all_nodes_running:
 	$(INFRA_CMD) ${PROVIDER} nodes check-running -a ${AUTH_FILE} \
 		-v ZONE:${ZONE} -v GKE_PROJECT_ID:${GKE_PROJECT_ID} \
 		-v EKS_WORKER_ROLE_ARN:${EKS_WORKER_ROLE_ARN} -v EKS_CLUSTER_ROLE_ARN:${EKS_CLUSTER_ROLE_ARN} \
 		-v EKS_SUBNET_IDS:${EKS_SUBNET_IDS} -v SEPARATOR:${SEPARATOR} \
 		-v CLUSTER_NAME:${CLUSTER_NAME} -v PR_NUMBER:${PR_NUMBER} \
-		-f manifests/prombench/nodes_${PROVIDER}.yaml	
+		-f ${PROMBENCH_DIR}/${BENCHMARK_DIRECTORY}/nodes_${PROVIDER}.yaml
 
 all_nodes_deleted:
 	$(INFRA_CMD) ${PROVIDER} nodes check-deleted -a ${AUTH_FILE} \
 		-v ZONE:${ZONE} -v GKE_PROJECT_ID:${GKE_PROJECT_ID} \
 		-v EKS_WORKER_ROLE_ARN:${EKS_WORKER_ROLE_ARN} -v EKS_CLUSTER_ROLE_ARN:${EKS_CLUSTER_ROLE_ARN} \
 		-v EKS_SUBNET_IDS:${EKS_SUBNET_IDS} -v SEPARATOR:${SEPARATOR} \
 		-v CLUSTER_NAME:${CLUSTER_NAME} -v PR_NUMBER:${PR_NUMBER} \
-		-f manifests/prombench/nodes_${PROVIDER}.yaml
+		-f ${PROMBENCH_DIR}/${BENCHMARK_DIRECTORY}/nodes_${PROVIDER}.yaml
diff --git a/prombench/README.md b/prombench/README.md
@@ -4,24 +4,24 @@
 
 This setup leverages **GitHub Actions** and **Google Kubernetes Engine (GKE)**, but is designed to be extendable to other Kubernetes providers.
 
-## Overview of Manifest Files
+## Configuration Files
 
-The `/manifest` directory contains Kubernetes manifest files:
+The `./manifest` directory contains configuration files. We can outline :
 
-- **`cluster_gke.yaml`**: Creates the Main Node in GKE.
-- **`cluster_eks.yaml`**: Creates the Main Node in EKS.
-- **`cluster-infra/`**: Contains persistent components of the Main Node.
-- **`prombench/`**: Resources created and destroyed for each Prombench test.
+- **`./manifest/cluster_gke.yaml`**: Creates the Main Node in GKE.
+- **`./manifest/cluster_eks.yaml`**: Creates the Main Node in EKS.
+- **`./manifest/cluster-infra/`**: Contains persistent components of the Main Node.
+- **`./manifest/prombench/`**: Resources created and destroyed for each Prombench test. See [`its README.md`](./manifests/prombench/README.md) for details.
 
-## Setup and Running Prombench
+## Prombench Setup
 
 Prombench can be run on different providers. Follow these instructions based on your provider:
 
 - [Google Kubernetes Engine (GKE)](docs/gke.md)
 - [Kubernetes In Docker (KIND)](docs/kind.md)
 - [Elastic Kubernetes Service (EKS)](docs/eks.md)
 
-## Setting Up GitHub Actions
+### Setting Up GitHub Actions
 
 1. Place a workflow file in the `.github` directory of your repository. Refer to the [Prometheus GitHub repository](https://github.com/prometheus/prometheus) for an example.
 
@@ -30,22 +30,34 @@ Prombench can be run on different providers. Follow these instructions based on
    ```bash
    cat $AUTH_FILE | base64 -w 0
    ```
+
+3. Configure webhook to cluster's comment-monitor as described [here](../tools/comment-monitor/README.md#setting-up-the-github-webhook).
+
+## Prombench Usage
 
 ### Triggering Tests via GitHub Comment
 
 **Starting Tests:**
 
 - `/prombench main` or `/prombench master` - Compare PR with the main/master branch.
 - `/prombench v2.4.0` - Compare PR with a specific release version (e.g., from [quay.io/prometheus/prometheus:releaseVersion](https://quay.io/prometheus/prometheus:releaseVersion)).
+- `/prombench v2.4.0 --bench.version=@aca1803ccf5d795eee4b0848707eab26d05965cc` - Compare with 2.4.0 release, but use a specific `aca1803ccf5d795eee4b0848707eab26d05965cc` commit on this repository for `./manifests/prombench` resources.
+- `/prombench v2.4.0 --bench.version=mybranch` - Compare with 2.4.0 release, but use a specific `mybranch` on this repository for `./manifests/prombench` resources.
+- `/prombench v2.4.0 --bench.directory=manifests/prombench-agent-mode` - Compare with 2.4.0 release, but use a specific resource directory on `master` branch for this repository. Currently there is only `./manifests/prombench` available (default), we might add more modes in the future.
 
 **Restarting Tests:**
 
 - `/prombench restart <release_version>`
+- `/prombench restart <release_version> --bench.version=... --bench.directory...` 
 
 **Stopping Tests:**
 
 - `/prombench cancel`
 
+**Printing available commands:**
+
+- `/prombench help`
+
 ### Building the Docker Image
 
 Build the Docker image with:
@@ -54,3 +66,6 @@ Build the Docker image with:
 docker build -t prominfra/prombench:master .
 ```
 
+
+
+
diff --git a/prombench/manifests/prombench/README.md b/prombench/manifests/prombench/README.md
@@ -0,0 +1,45 @@
+## Prombench Benchmark Scenario Configuration
+
+This directory contains resources that are applied (and cleaned) on every benchmark request
+via `infra` CLI using [`make deploy`](../../Makefile) and cleaned using [`make clean`](../../Makefile).
+
+It assumes running cluster was created via `infra` CLI using `make cluster_create` and `make cluster_delete`.
+
+### Variables
+
+It expects the following templated variables:
+
+* `.PR_NUMBER`: The PR number from which `/prombench` was triggered. This PR number also tells what commit to use for the `prometheus-test-pr-{{ .PR_NUMBER }}` Prometheus image building (in the init container).
+* `.RELEASE`: The argument provided by `/prombench` caller representing the Prometheus version (docker image tag for `quay.io/prometheus/prometheus:{{ .RELEASE }}`) to compare with, deployed as the `prometheus-test-{{ .RELEASE }}`.
+* `.DOMAIN_NAME`
+* `.LOADGEN_SCALE_UP_REPLICAS`
+* `.GITHUB_ORG`
+* `.GITHUB_REPO`
+
+### Customizations
+
+> NOTE: See https://github.com/prometheus/proposals/pull/41 for design.
+
+On the `master` branch, in this directory, we maintain the standard, single benchmarking scenario used
+as an acceptance validation for Prometheus. It's important to ensure it represents common Prometheus configuration.
+
+The only user related parameter for the standard scenario is `RELEASE` version.
+
+However, it's possible to create, a fully custom benchmarking scenarios for `/prombench` via `--bench.version=<branch|@commit>` flag.
+
+Here are an example steps:
+
+1. Create a new branch on https://github.com/prometheus/test-infra e.g. `benchmark/scenario1`.
+2. Modify this directory to your liking e.g. changing query load, metric load of advanced Prometheus configuration. It's also possible to make Prometheus deployments and versions exactly the same, but vary in a single configuration flag, for feature benchmarking.
+
+    > WARN: When customizing this directory, don't change `1a_namespace.yaml` or `1c_cluster-role-binding.yaml` filenames as they are used for cleanup routine. Or, if you change it, know what you're doing in relation to [`make clean` job](../../Makefile).
+
+3. Push changes to the new branch.
+4. From the Prometheus PR comment, call prombench as `/prombench <release> --bench.version=benchmark/scenario1` or `/prombench <release> --bench.version=@<relevant commit SHA from the benchmark/scenario1>` to use configuration files from this custom branch.
+
+Other details:
+
+* Other custom branch modifications other than to this directory do not affect prombench (e.g. to infra CLI or makefiles).
+* `--bench.version` is designed for a short-term or even one-off benchmark scenario configurations. It's not designed for long-term, well maintained scenarios. For the latter reason we can later e.g. maintain multiple `manifests/prombench` directories and introduce a new `--bench.directory` flag.
+* Non-maintainers can follow similar process, but they will need to ask maintainer for a new branch and PR review. We can consider extending `--bench.version` to support remote repositories if this becomes a problem.
+* Custom benchmarking logic is implemented in the [`maybe_pull_custom_version` make job](../../Makefile) and invoked by the prombench GH job on Prometheus repo on `deploy` and `clean`.
diff --git a/...chmark/3b_prometheus-test_deployment.yaml → ...ark/3b_prometheus-test-pr_deployment.yaml b/...chmark/3b_prometheus-test_deployment.yaml → ...ark/3b_prometheus-test-pr_deployment.yaml
@@ -89,7 +89,8 @@ spec:
           name: prometheus-test
       - name: instance-ssd
         hostPath:
-          path: /mnt/disks/ssd0 #gke ssds
+          # /mnt is where GKE keeps it's SSD.
+          path: /mnt/disks/ssd0
       - name: prometheus-executable
         emptyDir: {}
       terminationGracePeriodSeconds: 300
@@ -113,91 +114,3 @@ spec:
   selector:
     app: prometheus
     prometheus: test-pr-{{ .PR_NUMBER }}
----
-apiVersion: apps/v1
-kind: Deployment
-metadata:
-  name: prometheus-test-{{ normalise .RELEASE }}
-  namespace: prombench-{{ .PR_NUMBER }}
-  labels:
-    app: prometheus
-    prometheus: test-{{ normalise .RELEASE }}
-spec:
-  replicas: 1
-  selector:
-    matchLabels:
-      app: prometheus
-      prometheus: test-{{ normalise .RELEASE }}
-  template:
-    metadata:
-      namespace: prombench-{{ .PR_NUMBER }}
-      labels:
-        app: prometheus
-        prometheus: test-{{ normalise .RELEASE }}
-    spec:
-      serviceAccountName: prometheus
-      affinity:
-        podAntiAffinity:
-          requiredDuringSchedulingIgnoredDuringExecution:
-          - topologyKey: kubernetes.io/hostname
-            labelSelector:
-              matchExpressions:
-              - key: app
-                operator: In
-                values:
-                - prometheus
-      securityContext:
-        runAsUser: 0
-      containers:
-      - name: prometheus
-        image: quay.io/prometheus/prometheus:{{ .RELEASE }}
-        imagePullPolicy: Always
-        command: [ "/bin/prometheus" ]
-        args: [
-          "--web.external-url=http://{{ .DOMAIN_NAME }}/{{ .PR_NUMBER }}/prometheus-release",
-          "--storage.tsdb.path=/prometheus",
-          "--config.file=/etc/prometheus/prometheus.yml",
-          "--log.level=debug"
-        ]
-        resources:
-          requests:
-            cpu: 2
-            memory: 20Gi
-        volumeMounts:
-        - name: config-volume
-          mountPath: /etc/prometheus
-        - name: instance-ssd
-          mountPath: /prometheus
-        ports:
-        - name: prom-web
-          containerPort: 9090
-      volumes:
-      - name: config-volume
-        configMap:
-          name: prometheus-test
-      - name: instance-ssd
-        hostPath:
-          # /mnt is where GKE keeps it's SSD
-          # don't change this if you want Prometheus to take advantage of these local SSDs
-          path: /mnt/disks/ssd0
-      terminationGracePeriodSeconds: 300
-      nodeSelector:
-        node-name: prometheus-{{ .PR_NUMBER }}
-        isolation: prometheus
----
-apiVersion: v1
-kind: Service
-metadata:
-  name: prometheus-test-{{ normalise .RELEASE }}
-  namespace: prombench-{{ .PR_NUMBER }}
-  labels:
-    app: prometheus
-    prometheus: test-{{ normalise .RELEASE }}
-spec:
-  ports:
-  - name: prom-web
-    port: 80
-    targetPort: prom-web
-  selector:
-    app: prometheus
-    prometheus: test-{{ normalise .RELEASE }}