Update arch_mismatch_detector_agent to fetch job start and end times from GCS, and replace loki_query mcp function call by the grafana_loki_ query function call

Author: sherine-k

sub_agents/arch_mismatch_detector/agent.py

@@ -4,38 +4,94 @@
 from google.adk.models.lite_llm import LiteLlm
 from . import prompt
 from google.adk.tools.mcp_tool.mcp_toolset import MCPToolset, StreamableHTTPConnectionParams
+from dotenv import load_dotenv
 
 
 import asyncio
 import httpx
-import re
 import threading
 import concurrent.futures
-import re
 import os
-from typing import Dict, Any, Optional
+from typing import Dict, Any
+from datetime import datetime
 
 GCS_URL = "https://gcsweb-ci.apps.ci.l2s4.p1.openshiftapps.com/gcs/test-platform-results/logs"
 
 MODEL = os.environ.get("MODEL", "qwen3:4b")
 
+load_dotenv()
+
+def run_async_in_thread(coro):
+    """Run async function in a thread to avoid event loop conflicts."""
+    
+    def run_in_thread():
+        loop = asyncio.new_event_loop()
+        asyncio.set_event_loop(loop)
+        try:
+            return loop.run_until_complete(coro)
+        finally:
+            loop.close()
+    
+    with concurrent.futures.ThreadPoolExecutor() as executor:
+        future = executor.submit(run_in_thread)
+        return future.result()
+
+def get_job_start_and_end_time_tool(job_name: str, build_id: str)-> Dict[str, Any]:
+    """Get the start and end time of a job."""
+    return run_async_in_thread(get_job_start_and_end_time_async(job_name, build_id))
+
+async def get_job_start_and_end_time_async(job_name: str, build_id: str)-> Dict[str, Any]:
+    """Get the start and end time of a job."""
+    start_time = None
+    end_time = None
+    url_started = f"{GCS_URL}/{job_name}/{build_id}/started.json"
+    url_finished = f"{GCS_URL}/{job_name}/{build_id}/finished.json"
+    try:
+        async with httpx.AsyncClient() as client:
+            response_started = await client.get(url_started)
+            response_started.raise_for_status()
+            data_started = response_started.json()
+            if not data_started:
+                return {"error": "No response from Prow API for started.json"}
+            start_time = data_started["timestamp"]
+        async with httpx.AsyncClient() as client:
+            response_finished = await client.get(url_finished)
+            response_finished.raise_for_status()
+            data_finished = response_finished.json()
+            if not data_finished:
+                return {"error": "No response from Prow API for finished.json"}
+            end_time = data_finished["timestamp"]   
+        # Convert epoch timestamps to RFC 3339 format
+        if start_time:
+            start_time = datetime.fromtimestamp(start_time).strftime('%Y-%m-%dT%H:%M:%SZ')
+        if end_time:
+            end_time = datetime.fromtimestamp(end_time).strftime('%Y-%m-%dT%H:%M:%SZ')
+        return {
+            "start_time": start_time,
+            "end_time": end_time
+        }
+    except Exception as e:
+        return {"error": f"Failed to fetch job start and end time: {str(e)}"}
 
 arch_mismatch_detector_agent = Agent(
     model=LiteLlm(model=MODEL),
+    # model="gemini-2.0-flash",
     name="arch_mismatch_detector_agent",
     instruction=prompt.ARCH_MISMATCH_DETECTOR_PROMPT,
     output_key="installation_analysis_output",
     tools=[
+        get_job_start_and_end_time_tool,
         MCPToolset(
             connection_params=StreamableHTTPConnectionParams(
-                url="http://127.0.0.1:9000/mcp",
-                tool_filter=[
-                    "loki_query",
-                    "loki_label_names",
-                    "loki_label_values",
-                ],
+                url="http://127.0.0.1:8888/stream"
             ),
+            tool_filter=[
+                "grafana_loki_query",
+            ],
+            
         ),
 
     ],
-)
+)
+
+root_agent = arch_mismatch_detector_agent

sub_agents/arch_mismatch_detector/prompt.py

@@ -1,40 +1,41 @@
 """Prompts for Installation Analyst Agent."""
 
-ARCH_MISMATCH_DETECTOR_PROMPT = """
-You are the Arch Mismatch Detector agent.
+ARCH_MISMATCH_DETECTOR_PROMPT = f"""
+You are the Arch Mismatch Detector agent. You are a grafana loki expert.
 
 Objective:
 - Retrieve and analyze Grafana Loki logs for a specific CI job invocation to identify architecture mismatch errors, specifically messages matching the case-insensitive pattern "exec format".
 
 Required user inputs in each request:
-- start_time: the absolute start of the time range to search.
-- end_time: the absolute end of the time range to search.
 - job_name: the name of the job to search for.
 - build_id: the id of the build to search for.
 
-Time handling rules:
-- Accept start_time and end_time as either Unix epoch milliseconds or ISO 8601 / RFC3339 timestamps.
-- If the user does not provide both, ask them to provide both before proceeding.
-- Convert any non-epoch-millisecond timestamps to epoch milliseconds before calling tools.
-- If start_time >= end_time, ask the user to correct the range.
-
-Data source and query:
-- Always query Grafana Loki using the loki_query tool.
-- Always set orgId to 1.
-- Build the invoker as: "openshift-internal-ci/" + job_name + "/" + build_id
-- Construct the expression exactly (preserve content; whitespace changes are okay):
-  {invoker="openshift-internal-ci/{job_name}/{build_id}"} |~ "(?i)exec format"
+Workflow:
+- Use the get_job_start_and_end_time_tool to get the start and end time of the job from the prow job_name and build_id.
+- Prepare the inputs for the grafana_loki_query tool 
+  - orgId: 1
+  - datasource uid: PCEB727DF2F34084E (DPCR Loki)
+  - url: https://grafana-loki.ci.openshift.org/api/ds/query
+  - expr:
+    - Replace %job_name%, %build_id% by the values provided in the user inputs in the following expression:
+      {{invoker="openshift-internal-ci/%job_name%/%build_id%"}} |~ "(?i)exec format"
+- Set the start_time and end_time to the values provided by the get_job_start_and_end_time_tool.
+- Use the grafana_loki_query tool to query the logs for the job.
+- Analyze the logs for the job.
+- Return the analysis.
 
-Tool invocation contract (loki_query):
+Tool invocation contract (grafana_loki_query):
 - Parameters you must provide:
+  - datasource uid: PCEB727DF2F34084E (DPCR Loki)
   - orgId: 1
-  - url: https://grafana-loki.ci.openshift.org
-  - expr: the expression above
-  - start: start_time in Unix epoch milliseconds
-  - end: end_time in Unix epoch milliseconds
+  - url: https://grafana-loki.ci.openshift.org/api/ds/query
+  - expr: the expression above prepared by the workflow
+  - start: start_time in ISO 8601 / RFC3339 format
+  - end: end_time in ISO 8601 / RFC3339 format
 
 Response style:
 - Keep outputs concise and focused on the error pattern.
+- When no logs are found, this indicates that there are no arch mismatch errors in the job.
 - Report total matches, and surface 3–5 representative lines with timestamps.
 - Briefly note any repeated message patterns or clusters.
 - Provide a convenience link for further inspection in Grafana Explore with orgId=1 and the requested time range, e.g., https://grafana-loki.ci.openshift.org/explore?orgId=1