update

Signed-off-by: Chenfei Zhang <[email protected]>

Author: chenfeiz0326

jenkins/runPerfSanityTriage.groovy

@@ -22,7 +22,6 @@ def createKubernetesPodConfig(image, arch = "amd64")
             apiVersion: v1
             kind: Pod
             spec:
-                qosClass: Guaranteed
                 nodeSelector:
                   nvidia.com/node_type: builder
                   kubernetes.io/os: linux
@@ -83,9 +82,9 @@ pipeline {
         string(name: "BRANCH", defaultValue: "main", description: "Branch to checkout.")
         string(name: "OPEN_SEARCH_PROJECT_NAME", defaultValue: "swdl-trtllm-infra-ci-prod-perf_sanity_info", description: "OpenSearch project name.")
         string(name: "OPERATION", defaultValue: "", description: "Operation to perform.")
+        string(name: "QUERY_JOB_NUMBER", defaultValue: "1", description: "Number of latest jobs to query.")
         string(name: "SLACK_CHANNEL_ID", defaultValue: "", description: "Slack channel IDs to send messages to.")
-        string(name: "SLACK_BOT_TOKEN", defaultValue: "", description: "Slack bot token for authentication.")
-        string(name: "QUERY_JOB_NUMBER", defaultValue: "", description: "Number of latest jobs to query.")
+        credentials(name: "SLACK_BOT_TOKEN", defaultValue: "", description: "Slack bot token credential ID for authentication.", credentialType: 'org.jenkinsci.plugins.plaincredentials.impl.StringCredentialsImpl')
     }
     stages {
         stage("Run Perf Sanity Script") {
@@ -96,14 +95,16 @@ pipeline {
                         sh "env | sort"
                         trtllm_utils.checkoutSource(LLM_REPO, params.BRANCH, LLM_ROOT, false, false)
                         sh "pip install slack_sdk"
-                        sh """
-                            cd ${LLM_ROOT}/jenkins/scripts/perf && ls -alh && python3 perf_sanity_triage.py \
-                            --project_name "${params.OPEN_SEARCH_PROJECT_NAME}" \
-                            --operation "${params.OPERATION}" \
-                            --channel_id "${params.SLACK_CHANNEL_ID}" \
-                            --bot_token "${params.SLACK_BOT_TOKEN}" \
-                            --query_job_number "${params.QUERY_JOB_NUMBER}"
-                        """
+                        withCredentials([string(credentialsId: params.SLACK_BOT_TOKEN, variable: 'SLACK_TOKEN')]) {
+                            sh """
+                                cd ${LLM_ROOT}/jenkins/scripts/perf && ls -alh && python3 perf_sanity_triage.py \
+                                --project_name "${params.OPEN_SEARCH_PROJECT_NAME}" \
+                                --operation "${params.OPERATION}" \
+                                --channel_id "${params.SLACK_CHANNEL_ID}" \
+                                --bot_token "\${SLACK_TOKEN}" \
+                                --query_job_number "${params.QUERY_JOB_NUMBER}"
+                            """
+                        }
                     }
                 }
             }

jenkins/scripts/open_search_db.py

@@ -251,15 +251,11 @@ def queryFromOpenSearchDB(json_data, project) -> dict:
         use_poc_db = "sandbox" in project
         if not OPEN_SEARCH_DB_BASE_URL:
             OpenSearchDB.logger.info("OPEN_SEARCH_DB_BASE_URL is not set")
-            print(f"OPEN_SEARCH_DB_BASE_URL is not set")
             return None
         if not use_poc_db and project not in READ_ACCESS_PROJECT_NAME:
             OpenSearchDB.logger.info(
                 f"project {project} is not in read access project list: {json.dumps(READ_ACCESS_PROJECT_NAME)}"
             )
-            print(
-                f"project {project} is not in read access project list: {json.dumps(READ_ACCESS_PROJECT_NAME)}"
-            )
             return None
         if not isinstance(json_data, str):
             json_data_dump = json.dumps(json_data)
@@ -281,16 +277,10 @@ def queryFromOpenSearchDB(json_data, project) -> dict:
             OpenSearchDB.logger.info(
                 f"OpenSearchDB query failed, will retry, error:{res.status_code} {res.text}"
             )
-            print(
-                f"OpenSearchDB query failed, will retry, error:{res.status_code} {res.text}"
-            )
             retry_time -= 1
         OpenSearchDB.logger.info(
             f"Fail to queryFromOpenSearchDB after {retry_time} retry: {url}, json: {json_data_dump}, error: {res.text}"
         )
-        print(
-            f"Fail to queryFromOpenSearchDB after {retry_time} retry: {url}, json: {json_data_dump}, error: {res.text}"
-        )
         return None
 
     @staticmethod