Es/taskify reexecution

.github/actions/c-chain-reexecution-benchmark/action.yml

@@ -2,30 +2,18 @@ name: 'C-Chain Re-Execution Benchmark'
 description: 'Run C-Chain re-execution benchmark'
 
 inputs:
+  task:
+    description: 'Task name to execute from Taskfile.yml'
+    required: true
   runner_name:
     description: 'The name of the runner to use and include in the Golang Benchmark name.'
     required: true
-  config:
-    description: 'The config to pass to the VM for the benchmark. See BenchmarkReexecuteRange for details.'
-    default: ''
-  start-block:
-    description: 'The start block for the benchmark.'
-    default: '101'
-  end-block:
-    description: 'The end block for the benchmark.'
-    default: '250000'
-  block-dir-src:
-    description: 'The source block directory. Supports S3 directory/zip and local directories.'
-    default: 's3://avalanchego-bootstrap-testing/cchain-mainnet-blocks-1m-ldb/**'
-  current-state-dir-src:
-    description: 'The current state directory. Supports S3 directory/zip and local directories.'
-    default: 's3://avalanchego-bootstrap-testing/cchain-current-state-hashdb-full-100/**'
   aws-role:
     description: 'AWS role to assume for S3 access.'
     required: true
   aws-region:
     description: 'AWS region to use for S3 access.'
-    required: true
+    default: 'us-east-2'
   aws-role-duration-seconds:
     description: 'The duration of the AWS role to assume for S3 access.'
     required: true
@@ -56,64 +44,121 @@ inputs:
   push-github-action-benchmark:
     description: 'Whether to push the benchmark result to GitHub.'
     required: true
-    default: false
   push-post-state:
     description: 'S3 destination to copy the current-state directory after completing re-execution. If empty, this will be skipped.'
     default: ''
+  # The following inputs need never be provided by the caller. They
+  # default to context values that the action's steps are unable to
+  # access directly.
+  repository-owner:
+    default: ${{ github.repository_owner }}
+  repository-name:
+    default: ${{ github.event.repository.name }}
+  workflow:
+    default: ${{ github.workflow }}
+  run-id:
+    default: ${{ github.run_id }}
+  run-number:
+    default: ${{ github.run_number }}
+  run-attempt:
+    default: ${{ github.run_attempt }}
+  job:
+    default: ${{ github.job }}
+  grafana-dashboard-id:
+    default: 'Gl1I20mnk/c-chain'
 
 runs:
   using: composite
   steps:
-    - name: Set task env
+    - uses: cachix/install-nix-action@02a151ada4993995686f9ed4f1be7cfbb229e56f #v31
+      with:
+        github_access_token: ${{ inputs.github-token }}
+    - run: $GITHUB_ACTION_PATH/nix-develop.sh --command echo "dependencies installed"
+      shell: bash
+    # Cache Go modules (architecture-independent)
+    - uses: actions/cache@v4
+      id: go-mod-cache
+      with:
+        path: ~/go/pkg/mod
+        key: ${{ runner.os }}-go-mod-${{ hashFiles('go.sum') }}
+        restore-keys: ${{ runner.os }}-go-mod-
+    # Cache Go build cache (architecture-specific)
+    - uses: actions/cache@v4
+      with:
+        path: ~/.cache/go-build
+        key: ${{ runner.os }}-${{ runner.arch }}-go-build-${{ hashFiles('go.sum') }}
+        restore-keys: ${{ runner.os }}-${{ runner.arch }}-go-build-
+    # Download modules only on cache miss
+    - run: $GITHUB_ACTION_PATH/nix-develop.sh --command go mod download
+      if: steps.go-mod-cache.outputs.cache-hit != 'true'
+      shell: bash
+    - run: |
+        if [[ -f tools/go.mod ]]; then
+          $GITHUB_ACTION_PATH/nix-develop.sh --command go mod download -modfile=tools/go.mod
+        fi
+      if: steps.go-mod-cache.outputs.cache-hit != 'true'
+      shell: bash
+    - name: Notify of metrics availability
+      if: inputs.prometheus-username != ''
       shell: bash
       run: |
-        {
-          echo "EXECUTION_DATA_DIR=${{ inputs.workspace }}/reexecution-data"
-          echo "BENCHMARK_OUTPUT_FILE=output.txt"
-          echo "START_BLOCK=${{ inputs.start-block }}"
-          echo "END_BLOCK=${{ inputs.end-block }}"
-          echo "BLOCK_DIR_SRC=${{ inputs.block-dir-src }}"
-          echo "CURRENT_STATE_DIR_SRC=${{ inputs.current-state-dir-src }}"
-        } >> $GITHUB_ENV
+        metrics_url=$($GITHUB_ACTION_PATH/output-metrics-url.sh)
+        echo "Grafana: ${metrics_url}"
+        echo "🔗 [View Grafana Dashboard](${metrics_url})" >> "$GITHUB_STEP_SUMMARY"
+      env:
+        GRAFANA_URL: https://grafana-poc.avax-dev.network/d/${{ inputs.grafana_dashboard_id }}?orgId=1&refresh=10s&var-filter=is_ephemeral_node%7C%3D%7Cfalse&var-filter=gh_repo%7C%3D%7C${{ inputs.repository_owner }}%2F${{ inputs.repository_name }}&var-filter=gh_run_id%7C%3D%7C${{ inputs.run_id }}&var-filter=gh_run_attempt%7C%3D%7C${{ inputs.run_attempt }}
+        GH_JOB_ID: ${{ inputs.job }}
+    - name: Warn that collection of metrics and logs will not be performed
+      if: inputs.prometheus-username == ''
+      shell: bash
+      run: echo "::warning::Monitoring credentials not found. Skipping collector start. Is the PR from a fork branch?"
     - name: Configure AWS Credentials
       uses: aws-actions/configure-aws-credentials@v4
       with:
         role-to-assume: ${{ inputs.aws-role }}
         aws-region: ${{ inputs.aws-region }}
         role-duration-seconds: ${{ inputs.aws-role-duration-seconds }}
-    - name: Run C-Chain Re-Execution
-      uses: ./.github/actions/run-monitored-tmpnet-cmd
-      with:
-        run: |
-          ./scripts/run_task.sh reexecute-cchain-range-with-copied-data \
-            CONFIG=${{ inputs.config }} \
-            EXECUTION_DATA_DIR=${{ env.EXECUTION_DATA_DIR }} \
-            BLOCK_DIR_SRC=${{ env.BLOCK_DIR_SRC }} \
-            CURRENT_STATE_DIR_SRC=${{ env.CURRENT_STATE_DIR_SRC }} \
-            START_BLOCK=${{ env.START_BLOCK }} \
-            END_BLOCK=${{ env.END_BLOCK }} \
-            LABELS=${{ env.LABELS }} \
-            BENCHMARK_OUTPUT_FILE=${{ env.BENCHMARK_OUTPUT_FILE }} \
-            RUNNER_NAME=${{ inputs.runner_name }} \
-            METRICS_SERVER_ENABLED=true \
-            METRICS_COLLECTOR_ENABLED=true
-        prometheus_url: ${{ inputs.prometheus-url }}
-        prometheus_push_url: ${{ inputs.prometheus-push-url }}
-        prometheus_username: ${{ inputs.prometheus-username }}
-        prometheus_password: ${{ inputs.prometheus-password }}
-        grafana_dashboard_id: 'Gl1I20mnk/c-chain'
-        runtime: "" # Set runtime input to empty string to disable log collection
-
+    - name: Set task env
+      shell: bash
+      run: |
+        TIMESTAMP=$(date '+%Y%m%d-%H%M%S')
+        EXEC_DIR="/tmp/reexecution-data-${TIMESTAMP}"
+        echo "EXECUTION_DATA_DIR=${EXEC_DIR}" >> "$GITHUB_ENV"
+    - name: Run C-Chain Re-execution Benchmark
+      shell: bash
+      run: |
+        $GITHUB_ACTION_PATH/nix-develop.sh --impure --command bash -x ./scripts/run_task.sh ${{ inputs.task }} \
+          BENCHMARK_OUTPUT_FILE="benchmark-output.txt" \
+          EXECUTION_DATA_DIR="${EXEC_DIR}"
+      env:
+        RUNNER_NAME: ${{ inputs.runner_name }}
+        TMPNET_START_METRICS_COLLECTOR: ${{ inputs.prometheus-username != '' }}
+        TMPNET_CHECK_METRICS_COLLECTED: ${{ inputs.prometheus-username != '' }}
+        METRICS_SERVER_ENABLED: ${{ inputs.prometheus-username != '' }}
+        METRICS_COLLECTOR_ENABLED: ${{ inputs.prometheus-username != '' }}
+        PROMETHEUS_URL: ${{ inputs.prometheus-url }}
+        PROMETHEUS_PUSH_URL: ${{ inputs.prometheus-push-url }}
+        PROMETHEUS_USERNAME: ${{ inputs.prometheus-username }}
+        PROMETHEUS_PASSWORD: ${{ inputs.prometheus-password }}
+        GH_REPO: ${{ inputs.repository_owner }}/${{ inputs.repository_name }}
+        GH_WORKFLOW: ${{ inputs.workflow }}
+        GH_RUN_ID: ${{ inputs.run_id }}
+        GH_RUN_NUMBER: ${{ inputs.run_number }}
+        GH_RUN_ATTEMPT: ${{ inputs.run_attempt }}
+        GH_JOB_ID: ${{ inputs.job }}
     - name: Compare Benchmark Results
       uses: benchmark-action/github-action-benchmark@v1
       with:
         tool: 'go'
-        output-file-path: ${{ env.BENCHMARK_OUTPUT_FILE }}
+        output-file-path: benchmark-output.txt
         summary-always: true
         github-token: ${{ inputs.github-token }}
         auto-push: ${{ inputs.push-github-action-benchmark }}
-
-    - name: Push Post-State to S3 (if not exists)
-      if: ${{ inputs.push-post-state != '' }}
-      shell: nix develop --command bash -x {0}
-      run: ./scripts/run_task.sh export-dir-to-s3 SRC=${{ env.EXECUTION_DATA_DIR }}/current-state/ DST=${{ inputs.push-post-state }}
+    - name: Push Post-State to S3
+      if: inputs.push-post-state != ''
+      shell: bash
+      run: |
+        $GITHUB_ACTION_PATH/nix-develop.sh --command bash -x \
+          ./scripts/run_task.sh export-dir-to-s3 \
+          SRC=${{ env.EXECUTION_DATA_DIR }}/current-state/ \
+          DST=${{ inputs.push-post-state }}

.github/actions/c-chain-reexecution-benchmark/nix-develop.sh

@@ -0,0 +1,35 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+
+if [[ -f "flake.nix" ]]; then
+  echo "Starting nix shell for local flake"
+  FLAKE=
+else
+  echo "No local flake found, will attempt to use avalanchego flake"
+
+  # Get module details from go.mod
+  MODULE_DETAILS="$(go list -m "github.com/ava-labs/avalanchego" 2>/dev/null)"
+
+  # Extract the version part
+  AVALANCHE_VERSION="$(echo "${MODULE_DETAILS}" | awk '{print $2}')"
+
+  if [[ -z "${AVALANCHE_VERSION}" ]]; then
+    echo "Failed to get avalanchego version from go.mod"
+    exit 1
+  fi
+
+  # Check if the version matches the pattern where the last part is the module hash
+  # v*YYYYMMDDHHMMSS-abcdef123456
+  #
+  # If not, the value is assumed to represent a tag
+  if [[ "${AVALANCHE_VERSION}" =~ ^v.*[0-9]{14}-[0-9a-f]{12}$ ]]; then
+    # Use the module hash as the version
+    AVALANCHE_VERSION="$(echo "${AVALANCHE_VERSION}" | cut -d'-' -f3)"
+  fi
+
+  FLAKE="github:ava-labs/avalanchego?ref=${AVALANCHE_VERSION}"
+  echo "Starting nix shell for ${FLAKE}"
+fi
+
+nix develop "${FLAKE}" "${@}"

.github/actions/c-chain-reexecution-benchmark/output-metrics-url.sh

@@ -0,0 +1 @@
+../run-monitored-tmpnet-cmd/output-metrics-url.sh

.github/workflows/c-chain-reexecution-benchmark-config.json

@@ -0,0 +1,46 @@
+{
+  "pull_request": {
+    "include": [
+      {
+        "task": "c-chain-reexecution-hashdb-101-250k",
+        "runner": "ubuntu-latest",
+        "self_hosted": false,
+        "timeout-minutes": 30
+      },
+      {
+        "task": "c-chain-reexecution-hashdb-101-250k",
+        "runner": "avalanche-avalanchego-runner-2ti",
+        "self_hosted": true,
+        "timeout-minutes": 30
+      },
+      {
+        "task": "c-chain-reexecution-hashdb-archive-101-250k",
+        "runner": "blacksmith-4vcpu-ubuntu-2404",
+        "self_hosted": false,
+        "timeout-minutes": 30
+      },
+      {
+        "task": "c-chain-reexecution-hashdb-101-250k",
+        "runner": "blacksmith-4vcpu-ubuntu-2404",
+        "self_hosted": false,
+        "timeout-minutes": 30
+      }
+    ]
+  },
+  "schedule": {
+    "include": [
+      {
+        "task": "c-chain-reexecution-hashdb-33m-33m500k",
+        "runner": "avago-runner-m6i-4xlarge-ebs-fast",
+        "self_hosted": true,
+        "timeout-minutes": 1440
+      },
+      {
+        "task": "c-chain-reexecution-hashdb-33m-33m500k",
+        "runner": "avago-runner-i4i-4xlarge-local-ssd",
+        "self_hosted": true,
+        "timeout-minutes": 1440
+      }
+    ]
+  }
+}