use runs-on for replay-verify workflows

aptos-labs · Oct 4, 2024 · 6684481 · 6684481
1 parent 41e82f7
commit 6684481
Show file tree

Hide file tree

Showing 4 changed files with 63 additions and 43 deletions.
diff --git a/.github/workflows/module-verify.yaml b/.github/workflows/module-verify.yaml
@@ -38,8 +38,6 @@ jobs:
       BUCKET: aptos-testnet-backup-2223d95b
       SUB_DIR: e1
       BACKUP_CONFIG_TEMPLATE_PATH: terraform/helm/fullnode/files/backup/s3-public.yaml
-      # workflow config
-      RUNS_ON: high-perf-docker-with-local-ssd
       TIMEOUT_MINUTES: 20
 
   verify-modules-mainnet:
@@ -51,8 +49,6 @@ jobs:
       BUCKET: aptos-mainnet-backup-backup-831a69a8
       SUB_DIR: e1
       BACKUP_CONFIG_TEMPLATE_PATH: terraform/helm/fullnode/files/backup/s3-public.yaml
-      # workflow config
-      RUNS_ON: high-perf-docker-with-local-ssd
       TIMEOUT_MINUTES: 20
 
   test-verify-modules:
@@ -64,6 +60,4 @@ jobs:
       BUCKET: aptos-testnet-backup-2223d95b
       SUB_DIR: e1
       BACKUP_CONFIG_TEMPLATE_PATH: terraform/helm/fullnode/files/backup/s3-public.yaml
-      # workflow config
-      RUNS_ON: "high-perf-docker-with-local-ssd"
       TIMEOUT_MINUTES: 20
diff --git a/.github/workflows/replay-verify.yaml b/.github/workflows/replay-verify.yaml
@@ -35,6 +35,9 @@ on:
   schedule:
     - cron: "0 22 * * 0,2,4" # The main branch cadence. This runs every Sun,Tues,Thurs
 
+permissions:
+  id-token: write
+
 # cancel redundant builds
 concurrency:
   # cancel redundant builds on PRs (only on PR, not on branches)
@@ -70,8 +73,6 @@ jobs:
       # 1195000000-122000000: https://github.com/aptos-labs/aptos-core/pull/13832
       RANGES_TO_SKIP: "1195000000-1220000000"
       BACKUP_CONFIG_TEMPLATE_PATH: terraform/helm/fullnode/files/backup/gcs.yaml
-      # workflow config
-      RUNS_ON: "high-perf-docker-with-local-ssd"
       TIMEOUT_MINUTES: 180
       MAX_VERSIONS_PER_RANGE: 2000000
 
@@ -94,8 +95,6 @@ jobs:
       # 1197378568-1198492648: https://github.com/aptos-labs/aptos-core/pull/13832
       RANGES_TO_SKIP: "1197378568-1198492648"
       BACKUP_CONFIG_TEMPLATE_PATH: terraform/helm/fullnode/files/backup/gcs.yaml
-      # workflow config
-      RUNS_ON: "high-perf-docker-with-local-ssd"
       TIMEOUT_MINUTES: 180
       MAX_VERSIONS_PER_RANGE: 800000
 
@@ -115,7 +114,5 @@ jobs:
       # 1195000000-1220000000: https://github.com/aptos-labs/aptos-core/pull/13832
       RANGES_TO_SKIP: "1195000000-1220000000"
       BACKUP_CONFIG_TEMPLATE_PATH: terraform/helm/fullnode/files/backup/gcs.yaml
-      # workflow config
-      RUNS_ON: "high-perf-docker-with-local-ssd"
       TIMEOUT_MINUTES: 120 # increase test replay timeout to capture more flaky errors
       MAX_VERSIONS_PER_RANGE: 2000000
diff --git a/.github/workflows/workflow-run-module-verify.yaml b/.github/workflows/workflow-run-module-verify.yaml
@@ -20,12 +20,57 @@ on:
         description: "The path to the backup config template to use."
         type: string
         required: true
-      # GHA job config
-      RUNS_ON:
-        description: "The runner to use for the job."
+
+on:
+  # This allows the workflow to be triggered from another workflow
+  workflow_call:
+    inputs:
+      GIT_SHA:
+        required: true
+        type: string
+        description: The git SHA1 to test.
+      # replay-verify config
+      BUCKET:
+        required: true
+        type: string
+        description: The bucket to use for the backup. If not specified, it will use the default bucket.
+      SUB_DIR:
+        required: true
+        type: string
+        description: The subdirectory to use for the backup. If not specified, it will use the default subdirectory.
+      HISTORY_START:
+        required: true
+        type: string
+        description: The history start to use for the backup. If not specified, it will use the default history start.
+      TXNS_TO_SKIP:
+        required: false
+        type: string
+        description: The list of transaction versions to skip. If not specified, it will use the default list.
+      RANGES_TO_SKIP:
+        required: false
         type: string
+        description: The optional list of transaction ranges to skip..
+      BACKUP_CONFIG_TEMPLATE_PATH:
+        description: "The path to the backup config template to use."
+        type: string
+        required: true
+      TIMEOUT_MINUTES:
+        description: "Github job timeout in minutes"
+        type: number
+        required: true
+        default: 180
+      MAX_VERSIONS_PER_RANGE:
+        description: "The maximum number of versions to process in a single job."
+        type: number
         required: true
-        default: "high-perf-docker-with-local-ssd"
+  # This allows the workflow to be triggered manually from the Github UI or CLI
+  # NOTE: because the "number" type is not supported, we default to 720 minute timeout
+  workflow_dispatch:
+    inputs:
+      GIT_SHA:
+        required: true
+        type: string
+        description: The git SHA1 to test.
       TIMEOUT_MINUTES:
         description: "Github job timeout in minutes"
         type: number
@@ -36,16 +81,12 @@ jobs:
   module-verify:
     # if we're running on a PR, it's only for testing purposes, so we can set a shorter timeout
     timeout-minutes: ${{ inputs.TIMEOUT_MINUTES }}
-    runs-on: ${{ inputs.RUNS_ON }}
+    runs-on: runs-on,cpu=96,family=c5d+c5ad,image=aptos-ubuntu-x64,run-id=${{ github.run_id }},spot=false
     steps:
       - uses: actions/checkout@v4
         with:
           ref: ${{ inputs.GIT_SHA }}
 
-      - uses: aptos-labs/aptos-core/.github/actions/rust-setup@main
-        with:
-          GIT_CREDENTIALS: ${{ secrets.GIT_CREDENTIALS }}
-
       - name: Install AWS CLI
         shell: bash
         run: |

diff --git a/.github/workflows/workflow-run-replay-verify.yaml b/.github/workflows/workflow-run-replay-verify.yaml
@@ -33,12 +33,6 @@ on:
         description: "The path to the backup config template to use."
         type: string
         required: true
-      # GHA job config
-      RUNS_ON:
-        description: "The runner to use for the job."
-        type: string
-        required: true
-        default: "high-perf-docker-with-local-ssd"
       TIMEOUT_MINUTES:
         description: "Github job timeout in minutes"
         type: number
@@ -81,19 +75,13 @@ on:
         description: "The path to the backup config template to use."
         type: string
         required: true
-      # GHA job config
-      RUNS_ON:
-        description: "The runner to use for the job."
-        type: string
-        required: true
-        default: "high-perf-docker-with-local-ssd"
       MAX_VERSIONS_PER_RANGE:
         description: "The maximum number of versions to process in a single job."
         type: number
         required: true
 jobs:
   prepare:
-    runs-on: ${{ inputs.RUNS_ON }}
+    runs-on: runs-on,cpu=96,family=c5d+c5ad,image=aptos-ubuntu-x64,run-id=${{ github.run_id }},spot=false
     outputs:
       job_ids: ${{ steps.gen-jobs.outputs.job_ids }}
     steps:
@@ -126,11 +114,11 @@ jobs:
           strip -s target/release/aptos-debugger
           cp target/release/aptos-debugger .
 
-      - name: Install GCloud SDK
-        uses: "google-github-actions/setup-gcloud@v2"
+      - name: Authenticate with Google Cloud
+        uses: "google-github-actions/auth@v2"
         with:
-          version: ">= 418.0.0"
-          install_components: "kubectl,gke-gcloud-auth-plugin"
+          workload_identity_provider: ${{ secrets.GCP_WORKLOAD_IDENTITY_PROVIDER }}
+          service_account: ${{ secrets.GCP_SERVICE_ACCOUNT_EMAIL }}
 
       - name: get timestamp to use in cache key
         id: get-timestamp
@@ -178,7 +166,7 @@ jobs:
   replay-verify:
     needs: prepare
     timeout-minutes: ${{ inputs.TIMEOUT_MINUTES || 180 }}
-    runs-on: ${{ inputs.RUNS_ON }}
+    runs-on: runs-on,cpu=96,family=c5d+c5ad,image=aptos-ubuntu-x64,run-id=${{ github.run_id }},spot=false
     strategy:
       fail-fast: false
       matrix:
@@ -208,11 +196,11 @@ jobs:
           key: backup-config-${{ inputs.BUCKET }}/${{ inputs.SUB_DIR }}-${{ github.run_id }}
           fail-on-cache-miss: true
 
-      - name: Install GCloud SDK
-        uses: "google-github-actions/setup-gcloud@v2"
+      - name: Authenticate with Google Cloud
+        uses: "google-github-actions/auth@v2"
         with:
-          version: ">= 418.0.0"
-          install_components: "kubectl,gke-gcloud-auth-plugin"
+          workload_identity_provider: ${{ secrets.GCP_WORKLOAD_IDENTITY_PROVIDER }}
+          service_account: ${{ secrets.GCP_SERVICE_ACCOUNT_EMAIL }}
 
       - name: Run replay-verify in parallel
         env: