From 81a2f4268b9e0f25cb6e1ce5c28bba0a34c27604 Mon Sep 17 00:00:00 2001
From: Bo Wu <bo@aptoslabs.com>
Date: Sun, 15 Dec 2024 21:07:17 -0800
Subject: [PATCH] add image-tag to cancel group

---
 .github/workflows/replay-verify.yaml          |  2 +-
 testsuite/replay-verify/README.md             | 10 ++++++++++
 testsuite/replay-verify/archive_disk_utils.py | 10 +++++++++-
 testsuite/replay-verify/main.py               | 10 +++++++++-
 4 files changed, 29 insertions(+), 3 deletions(-)

diff --git a/.github/workflows/replay-verify.yaml b/.github/workflows/replay-verify.yaml
index 8de27ba546a80..6110eede6036b 100644
--- a/.github/workflows/replay-verify.yaml
+++ b/.github/workflows/replay-verify.yaml
@@ -42,7 +42,7 @@ permissions:
 # cancel redundant builds
 concurrency:
   # cancel redundant builds on PRs (only on PR, not on branches)
-  group: ${{ github.workflow }}-${{ (github.event_name == 'pull_request' && github.ref) || github.sha }}
+  group: ${{ github.workflow }}-${{ (github.event_name == 'pull_request' && github.ref) || github.sha }}-${{ inputs.IMAGE_TAG || 'latest' }}
   cancel-in-progress: true
 
 jobs:
diff --git a/testsuite/replay-verify/README.md b/testsuite/replay-verify/README.md
index 1c26e26d39288..fc829bbe77b87 100644
--- a/testsuite/replay-verify/README.md
+++ b/testsuite/replay-verify/README.md
@@ -11,6 +11,11 @@ The main script for executing replay verify tests. This script is responsible fo
 - Running replay verification tests against specified networks (testnet/mainnet)
 - Verifying transaction execution matches expected results
 - Handling test orchestration and reporting
+``` test with cli
+cd testsuite/replay-verify
+poetry shell
+python main.py  --image_tag YOUR_IMAGE_TAG --network testnet 
+```
 
 ### archive_disk_utils.py
 
@@ -19,5 +24,10 @@ A utility script for managing archive storage disks used in replay verification.
 - Provisions Google Cloud Storage disks for storing blockchain archive data
 - Supports both testnet and mainnet networks
 - Is called by GitHub Actions workflows to automatically manage storage resources
+```test with cli
+cd testsuite/replay-verify
+poetry shell
+python archive_disk_utils.py --network mainnet
+```
 
 
diff --git a/testsuite/replay-verify/archive_disk_utils.py b/testsuite/replay-verify/archive_disk_utils.py
index 02eb746899085..0e02c04497f6f 100644
--- a/testsuite/replay-verify/archive_disk_utils.py
+++ b/testsuite/replay-verify/archive_disk_utils.py
@@ -201,7 +201,7 @@ def generate_disk_name(run_id: str, snapshot_name: str, pvc_id: int) -> str:
 @retry(
     stop=stop_after_attempt(3),
     wait=wait_exponential(multiplier=1, min=4, max=10),
-    retry=retry_if_exception_type((ApiException, Exception)),
+    retry=retry_if_exception_type(ApiException),
     before_sleep=lambda retry_state: logger.warning(
         f"Retrying initial disk creation after error: {retry_state.outcome.exception()}"
     ),
@@ -493,6 +493,14 @@ def parse_args() -> argparse.Namespace:
     return args
 
 
+@retry(
+    stop=stop_after_attempt(3),
+    wait=wait_exponential(multiplier=1, min=4, max=10),
+    retry=retry_if_exception_type((ApiException, Exception)),
+    before_sleep=lambda retry_state: logger.warning(
+        f"Retrying creating pvc from snapshot after error: {retry_state.outcome.exception()}"
+    ),
+)
 def create_one_pvc_from_snapshot(
     pvc_name: str, snapshot_name: str, namespace: str, label: str
 ) -> str:
diff --git a/testsuite/replay-verify/main.py b/testsuite/replay-verify/main.py
index 566a850cf1585..dd889745087bd 100644
--- a/testsuite/replay-verify/main.py
+++ b/testsuite/replay-verify/main.py
@@ -237,6 +237,14 @@ def start(self) -> None:
                 )
                 time.sleep(RETRY_DELAY)
 
+    @retry(
+        stop=stop_after_attempt(MAX_RETRIES),
+        wait=wait_fixed(RETRY_DELAY),
+        retry=retry_if_exception_type(ApiException),
+        before_sleep=lambda retry_state: logger.warning(
+            f"Retry {retry_state.attempt_number}/{MAX_RETRIES} failed: {retry_state.outcome.exception()}"
+        ),
+    )
     def delete_pod(self):
         response = self.client.delete_namespaced_pod(
             name=self.name,
@@ -259,7 +267,7 @@ def get_pod_exit_code(self):
         retry=retry_if_exception_type(ApiException),
         before_sleep=lambda retry_state: logger.warning(
             f"Retry {retry_state.attempt_number}/{MAX_RETRIES} failed: {retry_state.outcome.exception()}"
-        )
+        ),
     )
     def get_pod_status(self):
         pod_status = self.client.read_namespaced_pod_status(