diff --git a/tools/torchci/test_insights/file_report_generator.py b/tools/torchci/test_insights/file_report_generator.py
new file mode 100644
index 0000000000..13938ab473
--- /dev/null
+++ b/tools/torchci/test_insights/file_report_generator.py
@@ -0,0 +1,671 @@
+#!/usr/bin/env python3
+"""
+File Report Generator
+
+This script generates comprehensive file reports by comparing test data between
+two commit SHAs. It fetches all test data, calculates diffs for all files, and
+groups results by owner labels from test_owner_labels.json.
+
+Usage:
+ python file_report_generator.py --sha1 abc123 --sha2 def456
+"""
+
+import argparse
+from functools import lru_cache
+import gzip
+import json
+import multiprocessing
+import random
+import re
+import urllib.request
+from datetime import datetime, timezone
+from typing import Dict, List, Optional, Any
+from dataclasses import dataclass
+from pathlib import Path
+import os
+from torchci.clickhouse import query_clickhouse
+import concurrent.futures
+import time
+
+def get_temp_dir() -> Path:
+ """Create a temporary directory for processing files"""
+ temp_dir = Path("/tmp/file_report_generator")
+ temp_dir.mkdir(parents=True, exist_ok=True)
+ (temp_dir / "intermediate_ind").mkdir(parents=True, exist_ok=True)
+ (temp_dir / "intermediate_status_changes").mkdir(parents=True, exist_ok=True)
+ return temp_dir
+
+@dataclass
+class FileReportData:
+ file: str
+ owner_labels: List[str]
+ timestamp: int
+ errors: int = 0
+ failures: int = 0
+ skipped: int = 0
+ successes: int = 0
+ tests: int = 0
+ time: float = 0.0
+ cost: float = 0.0
+ workflows: List[Dict[str, Any]] = None
+
+ def __post_init__(self):
+ if self.workflows is None:
+ self.workflows = []
+
+
+class FileReportGenerator:
+ """Generator for file reports based on owner labels"""
+
+ # S3 URL for EC2 pricing data
+ EC2_PRICING_URL = "https://ossci-metrics.s3.us-east-1.amazonaws.com/ec2_pricing.json.gz"
+
+ def __init__(self, reports_dir: str, test_owners_file: str = "test_owner_labels.json"):
+ """Initialize the generator with the test owners file path"""
+ self.test_owners_file = test_owners_file
+ self.base_dir = Path(__file__).parent.parent.parent.parent # Navigate to repo root
+ self.test_owners_path = self.base_dir / test_owners_file
+ self.reports_dir = Path(reports_dir)
+ if not self.reports_dir.exists():
+ self.reports_dir.mkdir(parents=True)
+
+ @lru_cache
+ def load_runner_costs(self) -> Dict[str, float]:
+ """Load runner costs from the S3 endpoint"""
+ print("Fetching EC2 pricing data from S3...")
+ with urllib.request.urlopen(self.EC2_PRICING_URL) as response:
+ compressed_data = response.read()
+
+ decompressed_data = gzip.decompress(compressed_data)
+ pricing_data = {}
+ for line in decompressed_data.decode('utf-8').splitlines():
+ if line.strip():
+ line_json = json.loads(line)
+ pricing_data[line_json[0]] = float(line_json[2])
+ return pricing_data
+
+ @lru_cache
+ def load_test_owners(self) -> List[Dict[str, Any]]:
+ """Load the test owner labels JSON file from S3"""
+ S3_URL = "https://ossci-metrics.s3.us-east-1.amazonaws.com/test_owner_labels/test_owner_labels.json.gz"
+ print(f"Fetching test owner labels from S3: {S3_URL}")
+ with urllib.request.urlopen(S3_URL) as response:
+ compressed_data = response.read()
+ decompressed_data = gzip.decompress(compressed_data)
+ test_owners = []
+ for line in decompressed_data.decode('utf-8').splitlines():
+ if line.strip():
+ test_owners.append(json.loads(line))
+ return test_owners
+
+
+ def get_runner_cost(self, runner_label: str) -> float:
+ """Get the cost per hour for a given runner"""
+ runner_costs = self.load_runner_costs()
+ if runner_label.startswith("lf."):
+ runner_label = runner_label[3:]
+ return runner_costs.get(runner_label, 0.0)
+
+ def _get_first_suitable_sha(self, shas: list[str]) -> Optional[str]:
+ """Get the first suitable SHA from a list of SHAs."""
+ lens = []
+ for sha in shas:
+ head_sha = sha['head_sha']
+ test_data = self._get_all_test_data_for_sha(head_sha)
+
+ has_no_job_name = False
+ for entry in test_data:
+ if not entry.get("job_id"):
+ has_no_job_name = True
+ break
+ if has_no_job_name:
+ print(f"Has entries with no job IDs for {head_sha}")
+ continue
+
+ lens.append((head_sha, len(test_data)))
+ del test_data
+
+ if len(lens) > 1:
+ lens.sort(key=lambda x: x[1], reverse=True)
+ if abs(lens[0][1] - lens[1][1]) * 2/ (lens[0][1] + lens[1][1]) < 0.1:
+ return lens[0][0]
+ return None
+
+ def find_suitable_shas(self, start_date: str, stop_date: str) -> list[str]:
+ """
+ Auto-select suitable SHAs from PyTorch main branch for a given date window.
+ Usage:
+ - Provide a date range (start and end) to select SHAs within that window.
+ - Returns 1 sha per day
+ Criteria:
+ - SHA is from main branch
+ - Workflow jobs are successful (green)
+ - S3 test data is available
+ - All test entries have job names
+ """
+
+ print("Searching for suitable SHAs from PyTorch main branch...")
+
+ params = {
+ "start_date": start_date + ' 00:00:00',
+ "stop_date": stop_date + ' 23:59:59',
+ }
+
+ # Single query with conditional logic using CASE expressions
+ query = """
+ SELECT
+ w.head_sha,
+ toUnixTimestamp(w.head_commit.'timestamp') as push_date
+ FROM default.workflow_run w
+ WHERE w.head_branch = 'main'
+ AND w.repository.full_name = 'pytorch/pytorch'
+ AND w.name in ('pull', 'trunk', 'inductor', 'slow')
+ AND w.conclusion = 'success'
+ AND w.head_commit.'timestamp' >= {start_date: DateTime}
+ AND w.head_commit.'timestamp' <= {stop_date: DateTime}
+ GROUP BY
+ w.head_sha, w.head_commit.'timestamp'
+ HAVING count(*) >= 4
+ ORDER BY
+ min(w.head_commit.'timestamp') DESC
+ """
+ print(f"Querying ClickHouse for successful workflow runs...")
+ candidates = query_clickhouse(query, params)
+
+ print(f"Found {len(candidates)} candidate workflow runs")
+
+
+ # Test each candidate SHA
+ group_by_day = {}
+ for candidate in candidates:
+ day = datetime.fromtimestamp(candidate['push_date'], timezone.utc).strftime('%Y-%m-%d')
+ if day not in group_by_day:
+ group_by_day[day] = []
+ group_by_day[day].append(candidate)
+
+ with multiprocessing.get_context("spawn").Pool(processes=2) as pool:
+ results = pool.map(self._get_first_suitable_sha, group_by_day.values())
+
+ return [sha for sha in results if sha is not None]
+
+ def _get_workflow_jobs_for_sha(self, sha: str) -> List[Dict[str, Any]]:
+ """Get workflow runs for a specific SHA using ClickHouse."""
+ query = """
+ SELECT DISTINCT
+ j.id as job_id,
+ j.name as job_name,
+ j.labels as job_labels,
+ j.run_id as workflow_id,
+ j.run_attempt,
+ j.workflow_name as workflow_name
+ FROM default.workflow_job j
+ WHERE j.head_sha = {sha: String}
+ AND j.workflow_name in ('pull', 'trunk', 'inductor', 'slow')
+ """
+
+ params = {'sha': sha}
+
+ print(f"Querying ClickHouse for workflow runs with SHA: {sha}")
+ result = query_clickhouse(query, params)
+
+ for row in result:
+ row["short_job_name"] = f"{row.get('workflow_name')} / {self._parse_job_name(row.get('job_name', ''))}"
+ row["runner_type"] = self._get_runner_label_from_job_info(row)
+ row["cost"] = self.get_runner_cost(row.get("runner_type", 0)) * self.get_frequency(row.get("workflow_name"))
+
+ return result
+
+ @lru_cache
+ def _get_frequency(self) -> float:
+ query = """
+ select
+ count(*) as count,
+ name
+ from
+ workflow_run j
+ where
+ j.created_at > now() - interval 8 day
+ and j.created_at < now() - interval 1 day
+ group by
+ name
+ """
+ params = {}
+ return query_clickhouse(query, params)
+
+ def get_frequency(self, workflow_name: str) -> float:
+ res = self._get_frequency()
+ for row in res:
+ if row["name"] == workflow_name:
+ return int(row["count"])
+ return 1
+
+ def _parse_job_name(self, job_name: str) -> str:
+ """
+ Parse job name to remove shard information.
+ Example: 'linux-jammy-py3.10-clang18-asan / test (default, 1, 6, linux.4xlarge)'
+ becomes: 'linux-jammy-py3.10-clang18-asan / test (default)'
+ """
+ if not job_name:
+ return "unknown"
+
+ # Replace with just the first part in parentheses
+ # First extract the part before the comma if it exists
+ match = re.search(r'\(([^,]+),.*\)', job_name)
+ if match:
+ base_part = job_name[:job_name.find('(')]
+ first_param = match.group(1)
+ return f"{base_part}({first_param})"
+
+ return job_name
+
+ def _get_runner_label_from_job_info(self, job_info: Dict[str, Any]) -> str:
+ """
+ Extract runner label from job information.
+ Tries multiple sources: runner_name, job_labels with 'self-hosted' patterns.
+ """
+ # Then try to find runner info in job_labels
+ job_labels = job_info.get("job_labels", [])
+ for label in job_labels:
+ if label.startswith("lf."):
+ label = label[3:]
+ if label in self.load_runner_costs():
+ return label
+
+ return "unknown"
+
+ def _fetch_test_data_from_s3(self, workflow_run_id: int, workflow_run_attempt: int) -> Optional[list[dict[str, Any]]]:
+ """
+ Use local cache for a specific workflow run.
+ """
+ bucket = "ossci-raw-job-status"
+ key = f"test_run/{workflow_run_id}/{workflow_run_attempt}"
+ try:
+ start_time = time.time()
+ file_loc = get_temp_dir() / f"cache_{workflow_run_id}_{workflow_run_attempt}"
+ if file_loc.exists():
+ print(f"Using cached download for {file_loc}")
+ compressed_data = file_loc.read_bytes()
+ else:
+ url = f"https://{bucket}.s3.amazonaws.com/{key}"
+ with urllib.request.urlopen(url) as response:
+ compressed_data = response.read()
+
+ with open(file_loc, 'wb') as f:
+ f.write(compressed_data)
+
+ decompressed_data = gzip.decompress(compressed_data)
+ text_data = decompressed_data.decode('utf-8')
+
+ test_data = []
+ for line in text_data.strip().split('\n'):
+ if line.strip():
+ test_data.append(json.loads(line))
+
+ print(f"Fetched {len(test_data)} test entries from {key}, took {time.time() - start_time:.2f} seconds")
+ return test_data
+ except Exception as e:
+ print(f"Failed to fetch from {key}: {e}")
+ return None
+
+ def process_all_test_data(self, all_test_data):
+ files = {}
+
+ for entry in all_test_data:
+ # Get file name (prefer 'file' over 'invoking_file')
+ file_name = entry.get("file_name") or entry.get("invoking_file")
+ if not file_name:
+ continue
+
+ job_id = str(entry.get("job_id", "unknown"))
+ job_info = entry.get("_job_info", {})
+
+ # Use pre-computed short job name if available, otherwise parse
+ job_name = job_info.get("short_job_name")
+ if not job_name:
+ raw_job_name = job_info.get("job_name", f"unknown_job")
+ job_name = self._parse_job_name(raw_job_name)
+
+ test_name = f"{entry.get('classname', '')}::{entry.get('name', '')}"
+ key = (file_name, job_name, test_name)
+
+ if key not in files:
+ files[key] = {
+ "test_name": test_name,
+ "job_name": job_name,
+ "file_name": file_name,
+ "count": 0,
+ "time": 0.0,
+ "cost": 0.0,
+ "errors": 0,
+ "failures": 0,
+ "skipped": 0,
+ "successes": 0,
+ "job_info": job_info,
+ "job_ids": set() # Track which job IDs contribute to this job name
+ }
+
+ # Update test data
+ files[key]["count"] += 1
+ files[key]["job_ids"].add(job_id)
+
+ # Handle time conversion safely
+ time_value = entry.get("time", 0)
+ if isinstance(time_value, (int, float)):
+ files[key]["time"] += time_value
+ elif isinstance(time_value, str):
+ try:
+ files[key]["time"] += float(time_value)
+ except (ValueError, TypeError):
+ # Skip invalid time values
+ pass
+
+ # Determine test outcome
+ if "failure" in entry:
+ files[key]["failures"] += 1
+ elif "error" in entry:
+ files[key]["errors"] += 1
+ elif "skipped" in entry:
+ files[key]["skipped"] += 1
+ else:
+ files[key]["successes"] += 1
+
+ # Calculate cost using pre-computed cost per hour if available
+ cost_per_hour = job_info.get("cost_per_hour")
+ if cost_per_hour is None:
+ # Fallback to manual calculation if not pre-computed
+ runner_label = self._get_runner_label_from_job_info(job_info)
+ cost_per_hour = self.get_runner_cost(runner_label)
+
+ try:
+ time_float = float(time_value)
+ files[key]["cost"] += time_float * cost_per_hour / 3600.0
+ except (ValueError, TypeError):
+ pass
+
+ for val in files.values():
+ val["job_ids"] = list(val["job_ids"])
+ return list(files.values())
+
+ def _get_all_test_data_for_sha(self, sha: str) -> List[Dict[str, Any]]:
+ """
+ Fetch all test data for a given SHA once and cache it.
+ Returns a flat list of test entries with job info embedded.
+ """
+ workflow_jobs = self._get_workflow_jobs_for_sha(sha)
+
+ # Create job_info mapping from the returned job data
+ job_info = {}
+ workflow_runs = set()
+
+ for job_data in workflow_jobs:
+ job_id = str(job_data.get("job_id"))
+ job_info[job_id] = {
+ "job_name": job_data.get("job_name", ""),
+ "short_job_name": job_data.get("short_job_name", ""),
+ "job_labels": job_data.get("job_labels", []),
+ "runner_type": job_data.get("runner_type", ""),
+ "cost_per_hour": job_data.get("cost", 0.0),
+ "workflow_name": job_data.get("workflow_name", "")
+ }
+ workflow_runs.add((job_data.get("workflow_id"), job_data.get("run_attempt", 1)))
+
+ all_test_data = []
+
+ # Use threads instead of processes for IO-bound S3 fetching
+ results = []
+ with concurrent.futures.ThreadPoolExecutor(max_workers=4) as executor:
+ futures = [
+ executor.submit(self._fetch_test_data_from_s3, run_id, run_attempt)
+ for run_id, run_attempt in workflow_runs
+ ]
+ # Maintain order to match workflow_runs
+ results = [f.result() for f in futures]
+ for test_data in results:
+ if test_data:
+ # Enrich each test entry with job info
+ for entry in test_data:
+ job_id = str(entry.get("job_id", "unknown"))
+ entry["_job_info"] = job_info.get(job_id, {})
+ all_test_data.extend(test_data)
+
+ return all_test_data
+
+ def group_by_job_file(self, tests, sha, push_date):
+ job_file_map = {}
+ for test in tests:
+ job_name = test["job_name"]
+ file_name = test["file_name"]
+ key = (job_name, file_name)
+ if key not in job_file_map:
+ job_file_map[key] = {
+ "job_name": job_name,
+ "file_name": file_name,
+ "count": 0,
+ "time": 0.0,
+ "cost": 0.0,
+ # "tests": set(),
+ "errors": 0,
+ "failures": 0,
+ "skipped": 0,
+ "successes": 0,
+ # "job_info": [],
+ # "job_ids": set(),
+ # "duplicate_tests": [], # Track tests with multiple entries
+ "labels": self.get_label_for_file(file_name),
+ "sha": sha,
+ "push_date": push_date
+ }
+ job_file_map[key]["count"] += 1
+ job_file_map[key]["time"] += test["time"]
+ job_file_map[key]["cost"] += test["cost"]
+ job_file_map[key]["errors"] += test["errors"]
+ job_file_map[key]["failures"] += test["failures"]
+ job_file_map[key]["skipped"] += test["skipped"]
+ job_file_map[key]["successes"] += test["successes"]
+
+ # # Track tests with multiple entries
+ # if test["count"] > 1:
+ # job_file_map[key]["duplicate_tests"].append({
+ # "test_name": test["test_name"],
+ # "count": test["count"]
+ # })
+
+ # for val in job_file_map.values():
+ # val["tests"] = list(val["tests"])
+ # val["job_ids"] = list(val["job_ids"])
+
+ return list(job_file_map.values())
+
+ def get_label_for_file(self, file_name: str) -> List[str]:
+ for row in self.load_test_owners():
+ if row["file"] == f"{file_name.replace('.', '/')}.py":
+ return row["owner_labels"]
+ return []
+
+ def get_status_changes(self, sha1: str, sha2: str, sha2_push_date: str) -> Dict[str, Any]:
+ """
+ Compare test data between two pre-fetched datasets.
+ Returns a dictionary with file_name as keys and job diffs as values.
+ """
+
+ def save_to_final(data):
+ if len(data) > 50:
+ # Too much data... sampling
+ shuffled = random.sample(data, len(data))
+ data = shuffled[:50]
+ with open(self.reports_dir / f"status_changes.json", 'a') as f:
+ json.dump(data, f)
+ f.write("\n")
+
+ cache_key = get_temp_dir() / "intermediate_status_changes" / f"{sha1} {sha2}"
+ if cache_key.exists():
+ with open(cache_key) as f:
+ data = json.load(f)
+ save_to_final(data)
+ return data
+
+ tests1 = self.get_data_for_sha(sha1)["process_all_test_data"]
+ tests2 = self.get_data_for_sha(sha2)["process_all_test_data"]
+
+ # Group by key
+ map1 = {(v["job_name"], v["file_name"], v["test_name"]): v for v in tests1}
+ map2 = {(v["job_name"], v["file_name"], v["test_name"]): v for v in tests2}
+
+ status_changes = []
+
+ for key in map1.keys() | map2.keys():
+ status = None
+
+ if key in map1 and key not in map2:
+ status = "removed"
+ elif key not in map1 and key in map2:
+ status = "added"
+ else:
+ skipped1 = map1[key].get("skipped", 0) > 0
+ skipped2 = map2[key].get("skipped", 0) > 0
+ if not skipped1 and skipped2:
+ status = "started_skipping"
+ elif skipped1 and not skipped2:
+ status = "stopped_skipping"
+ if status is not None:
+ status_changes.append({
+ "job_name": key[0],
+ "file_name": key[1],
+ "test_name": key[2],
+ "status": status,
+ "labels": self.get_label_for_file(key[1]),
+ "sha": sha2,
+ "push_date": sha2_push_date
+ })
+
+ with open(get_temp_dir() / "intermediate_status_changes" / f"{sha1} {sha2}", 'w') as f:
+ json.dump(status_changes, f)
+ save_to_final(status_changes)
+
+ return status_changes
+
+ def get_data_for_sha(self, sha: str) -> Dict[str, Any]:
+ def save_to_final(data):
+ with open(self.reports_dir / f"data.json", 'a') as f:
+ json.dump(data["group_by_job_file"], f)
+ f.write("\n")
+
+ json_path = get_temp_dir() / "intermediate_ind" / f"{sha}"
+ if json_path.exists():
+ with open(json_path, 'r') as f:
+ data = json.load(f)
+ save_to_final(data)
+ return data
+
+ push_date = self.get_push_date_for_sha(sha)
+ _get_all_test_data_for_sha = self._get_all_test_data_for_sha(sha)
+ process_all_test_data = self.process_all_test_data(_get_all_test_data_for_sha)
+ group_by_job_file = self.group_by_job_file(process_all_test_data, sha, push_date)
+ data = {
+ "sha": sha,
+ "push_date": push_date,
+ # "_get_all_test_data_for_sha": _get_all_test_data_for_sha,
+ "process_all_test_data": process_all_test_data,
+ "group_by_job_file": group_by_job_file
+ }
+ with open(json_path, 'w') as f:
+ json.dump(data, f, indent=2)
+ save_to_final(data)
+ return data
+
+
+ def get_push_date_for_sha(self, sha: str) -> Optional[str]:
+ """
+ Get the push date for a given SHA from ClickHouse push table.
+ Returns the date as an ISO string, or None if not found.
+ """
+ query = """
+ SELECT toUnixTimestamp(min(p.head_commit.timestamp)) as pushed_at
+ FROM default.push p
+ WHERE p.after = {sha: String}
+ AND p.ref = 'refs/heads/main'
+ AND p.repository.full_name = 'pytorch/pytorch'
+ """
+ params = {"sha": sha}
+ result = query_clickhouse(query, params)
+ if result:
+ return result[0]["pushed_at"]
+ return None
+
+def main():
+ """Main function to run the file report generator"""
+ parser = argparse.ArgumentParser(
+ description="Generate comprehensive file reports grouped by owner labels",
+ formatter_class=argparse.RawDescriptionHelpFormatter,
+ epilog="""
+Examples:
+ # Generate report with two specific dates (auto-select SHAs for each date)
+ python file_report_generator.py --day1 2025-08-15 --day2 2025-08-20
+ """
+ )
+
+ parser.add_argument("--start-date", help="Start date for filtering (YYYY-MM-DD)")
+ parser.add_argument("--stop-date", help="Stop date for filtering (YYYY-MM-DD)")
+ parser.add_argument("--shas", nargs="+", help="List of commit SHAs to compare in sequence")
+ parser.add_argument("--output", help="Output folder to write results to")
+ parser.add_argument("--test-pricing", action="store_true", help="Test fetching pricing data from S3 and exit")
+ parser.add_argument("--list-runners", action="store_true", help="List all available runners and their costs, then exit")
+ parser.add_argument("--test-owners-file", default="test_owner_labels.json", help="Path to the test owners file")
+
+ args = parser.parse_args()
+
+ # Generate default output filename if not provided
+ if not args.output:
+ timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
+ reports_dir = os.path.join(os.getcwd(), f"file_reports_{timestamp}")
+ if not os.path.exists(reports_dir):
+ os.makedirs(reports_dir)
+ args.output = str(reports_dir)
+
+ generator = FileReportGenerator(args.output, args.test_owners_file)
+
+ if args.test_pricing:
+ print("Testing pricing data fetch...")
+ runner_costs = generator.load_runner_costs()
+ print(f"Successfully loaded {len(runner_costs)} runner costs")
+ return
+
+ if args.list_runners:
+ print("Available runners and costs:")
+ runner_costs = generator.load_runner_costs()
+ for runner, cost in sorted(runner_costs.items()):
+ print(f" {runner:<50} ${cost:>8.4f}/hour")
+ return
+
+ if (args.start_date and not args.stop_date) or (not args.start_date and args.stop_date):
+ parser.error("Must provide both --start-date and --stop-date for auto-selection")
+ if args.shas and (args.start_date or args.stop_date):
+ parser.error("Cannot mix --shas with --start-date/--stop-date parameters")
+ if args.shas and len(args.shas) < 2:
+ parser.error("Either provide --shas with at least 2 SHAs, or --start-date and --stop-date for auto-selection")
+
+ # Handle the start-date/stop-date parameters
+ if args.start_date:
+ args.shas = generator.find_suitable_shas(args.start_date, args.stop_date)
+
+ print(f"Using SHAs: {args.shas}")
+
+ # Load data to get dates/ordering
+ shas_with_push_date = []
+ for sha in args.shas:
+ data = generator.get_data_for_sha(sha)
+ shas_with_push_date.append({
+ "sha": sha,
+ "push_date": data["push_date"]
+ })
+ del data
+
+ shas_with_push_date = sorted(shas_with_push_date, key=lambda x: x["push_date"])
+
+ print("Calculating diffs for all files and grouping by labels...")
+ for i in range(1, len(shas_with_push_date)):
+ generator.get_status_changes(shas_with_push_date[i-1]["sha"], shas_with_push_date[i]["sha"], shas_with_push_date[i]["push_date"])
+
+
+if __name__ == "__main__":
+ main()
diff --git a/torchci/components/common/TimeUtils.tsx b/torchci/components/common/TimeUtils.tsx
index 3aeff7126c..672367fc89 100644
--- a/torchci/components/common/TimeUtils.tsx
+++ b/torchci/components/common/TimeUtils.tsx
@@ -1,3 +1,4 @@
+import { Tooltip } from "@mui/material";
import dayjs from "dayjs";
import utc from "dayjs/plugin/utc";
import { useEffect, useState } from "react";
@@ -21,12 +22,11 @@ export function LocalTimeHuman({ timestamp }: { timestamp: string }) {
}
}, [timestamp]);
return (
-
- {time}
-
+ {time}
+
);
}
diff --git a/torchci/pages/tests/fileReport.tsx b/torchci/pages/tests/fileReport.tsx
new file mode 100644
index 0000000000..edb645efaf
--- /dev/null
+++ b/torchci/pages/tests/fileReport.tsx
@@ -0,0 +1,797 @@
+import {
+ Box,
+ Button,
+ ButtonGroup,
+ Stack,
+ TextField,
+ Typography,
+} from "@mui/material";
+import {
+ DataGrid,
+ GridColDef,
+ GridRenderCellParams,
+ GridTreeNodeWithRender,
+} from "@mui/x-data-grid";
+import LoadingPage from "components/common/LoadingPage";
+import { durationDisplay } from "components/common/TimeUtils";
+import ReactECharts from "echarts-for-react";
+import _ from "lodash";
+import { useRouter } from "next/router";
+import { useEffect, useRef, useState } from "react";
+
+function formatTimestamp(ts: number) {
+ return new Date(ts * 1000).toLocaleDateString().slice(0, 10);
+}
+
+function roundedCostCell(params: GridRenderCellParams) {
+ if (params.value != null) {
+ return <>{params.value.toFixed(2)} ;
+ }
+ return <>;
+}
+
+function renderTimeCell(
+ params: GridRenderCellParams
+) {
+ if (isNaN(params.value)) {
+ return "";
+ }
+ const value = parseFloat(params.value);
+
+ return durationDisplay(value);
+}
+
+function Diffs({
+ data,
+ setFileFilter,
+ setJobFilter,
+}: {
+ data: { [key: string]: any }[];
+ setFileFilter: (v: string) => void;
+ setJobFilter: (v: string) => void;
+}) {
+ const groupByOptions = {};
+ // Compute diffs for every row (except the earliest) in each (file_name, job_name) group
+ const groupedDiff = _.groupBy(data, (d) => `${d.file_name}|||${d.job_name}`);
+ // Map from id (row) to diff object for every row (except the first in group)
+ const rowDiffs: Record = {};
+ Object.entries(groupedDiff).forEach(([key, arr]) => {
+ // Sort by push_date ascending (oldest to newest)
+ const sorted = _.sortBy(arr, (d) => d.push_date);
+ for (let i = 1; i < sorted.length; ++i) {
+ const curr = sorted[i];
+ const prev = sorted[i - 1];
+ function diff(field: string) {
+ if (!curr || !prev) return null;
+ return (curr[field] || 0) - (prev[field] || 0);
+ }
+ rowDiffs[curr.id] = {
+ count_diff: diff("count"),
+ cost_diff: diff("cost"),
+ time_diff: diff("time"),
+ skipped_diff: diff("skipped"),
+ errors_diff: diff("errors"),
+ failures_diff: diff("failures"),
+ successes_diff: diff("successes"),
+ };
+ }
+ });
+
+ const columns: GridColDef[] = [
+ { field: "sha", headerName: "SHA", flex: 1 },
+ {
+ field: "push_date",
+ headerName: "Push Date",
+ flex: 1,
+ renderCell: (params: any) => formatTimestamp(params.value),
+ },
+ {
+ field: "file_name",
+ headerName: "File",
+ flex: 4,
+ renderCell: (params: any) => (
+ setFileFilter(params.value)}
+ title="Double-click to filter by this file"
+ >
+ {params.value}
+
+ ),
+ },
+ {
+ field: "job_name",
+ headerName: "Job",
+ flex: 4,
+ renderCell: (params: any) => (
+ setJobFilter(params.value)}
+ title="Double-click to filter by this job"
+ >
+ {params.value}
+
+ ),
+ },
+ { field: "count", headerName: "Count", flex: 1 },
+ {
+ field: "count_diff",
+ headerName: "Δ Count",
+ flex: 1,
+ cellClassName: (params: any) => {
+ const value = parseFloat(params.value);
+ const base = parseFloat(params.row?.count);
+ if (!isNaN(value) && base && Math.abs(value) / base > 0.2) {
+ return "highlight";
+ }
+ if (Math.abs(value) > 20) {
+ return "highlight";
+ }
+ return "change";
+ },
+ },
+ {
+ field: "time",
+ headerName: "Duration",
+ flex: 1,
+ renderCell: renderTimeCell,
+ },
+ {
+ field: "time_diff",
+ headerName: "Δ Duration",
+ flex: 1,
+ cellClassName: (params: any) => {
+ const value = parseFloat(params.value);
+ const base = parseFloat(params.row?.time);
+ if (!isNaN(value) && base && Math.abs(value) / base > 0.2) {
+ return "highlight";
+ }
+ if (Math.abs(value) > 30 * 60) {
+ return "highlight";
+ }
+ return "change";
+ },
+ renderCell: renderTimeCell,
+ },
+ {
+ field: "cost",
+ headerName: "Cost ($)",
+ flex: 1,
+ renderCell: roundedCostCell,
+ },
+ {
+ field: "cost_diff",
+ headerName: "Δ Cost",
+ flex: 1,
+ cellClassName: (params: any) => {
+ const value = parseFloat(params.value);
+ const base = parseFloat(params.row?.cost);
+ if (!isNaN(value) && base && Math.abs(value) / base > 0.2) {
+ return "highlight";
+ }
+ if (Math.abs(value) > 500) {
+ return "highlight";
+ }
+ return "change";
+ },
+ renderCell: roundedCostCell,
+ },
+ // { field: "errors", headerName: "Errors", flex: 1 },
+ // {
+ // field: "errors_diff",
+ // headerName: "Δ Errors",
+ // flex: 1,
+ // getCellClassName: () => "change",
+ // },
+ // { field: "failures", headerName: "Failures", flex: 1 },
+ // {
+ // field: "failures_diff",
+ // headerName: "Δ Failures",
+ // flex: 1,
+ // getCellClassName: () => "change",
+ // },
+ { field: "skipped", headerName: "Skipped", flex: 1 },
+ {
+ field: "skipped_diff",
+ headerName: "Δ Skipped",
+ flex: 1,
+ cellClassName: (params: any) => {
+ const value = parseFloat(params.value);
+ const base = parseFloat(params.row?.skipped);
+ if (!isNaN(value) && base && Math.abs(value) / base > 0.2) {
+ return "highlight";
+ }
+ if (Math.abs(value) > 20) {
+ return "highlight";
+ }
+ return "change";
+ },
+ },
+ // { field: "successes", headerName: "Successes", flex: 1 },
+ // {
+ // field: "successes_diff",
+ // headerName: "Δ Successes",
+ // flex: 1,
+ // getCellClassName: () => "change",
+ // },
+ ];
+
+ const styling = {
+ "& .total-row": {
+ fontWeight: "bold",
+ backgroundColor: "rgba(213, 213, 213, 0.25)",
+ },
+ "& .change": {
+ backgroundColor: "rgba(213, 213, 213, 0.25)",
+ },
+ "& .highlight": {
+ backgroundColor: "#ffe085ff",
+ },
+ };
+ return (
+
+ File Test Counts
+
+ This table displays test run statistics for each test file and job
+ combination. The Δ (delta) columns show the change in each metric
+ compared to the previous commit for the same file and job. Double click
+ on the file or job to filter by that value. Highlighted cells are large
+ changes (either by percent or absolute value) and may indicate
+ regressions or improvements.
+
+
+
+ Pricing is approximate and weighted by workflow frequency to represent
+ the cost over an entire week. Some pricing data may be missing (ex mac,
+ rocm), in those cases the cost will be 0.
+
+
+ ({
+ ...row,
+ ...(rowDiffs[row.id] || {}),
+ }))}
+ sx={styling}
+ columns={columns}
+ initialState={{
+ sorting: {
+ sortModel: [{ field: "cost_diff", sort: "desc" }],
+ },
+ }}
+ />
+
+
+ );
+}
+
+function Overview({
+ data,
+ setFileFilter,
+ setJobFilter,
+ setLabelFilter,
+}: {
+ data: { [key: string]: any }[];
+ setFileFilter: (_: string) => void;
+ setJobFilter: (_: string) => void;
+ setLabelFilter: (_: string) => void;
+}) {
+ const groupByOptions = {
+ file: {
+ headerName: "File",
+ field: "file_name",
+ buttonText: "Group by File",
+ onDoubleClick: (value: any) => setFileFilter(value),
+ onDoubleClickHelpText: "Double-click to filter by this file",
+ groupByKey: (v: any) => [v.file_name],
+ },
+ job: {
+ headerName: "Job",
+ field: "job_name",
+ buttonText: "Group by Job",
+ onDoubleClick: (value: any) => setJobFilter(value),
+ onDoubleClickHelpText: "Double-click to filter by this job",
+ groupByKey: (v: any) => [v.job_name],
+ },
+ label: {
+ headerName: "Label",
+ field: "labels",
+ buttonText: "Group by Label",
+ onDoubleClick: (value: any) => setLabelFilter(value),
+ onDoubleClickHelpText: "Double-click to filter by this label",
+ groupByKey: (v: any) => v.labels,
+ },
+ total: {
+ headerName: "Total",
+ field: "total",
+ buttonText: "Total",
+ onDoubleClick: () => {},
+ onDoubleClickHelpText: "",
+ groupByKey: (_: any) => ["total"],
+ },
+ };
+ const [groupBy, setGroupBy] = useState("file");
+ const columns: any[] = [
+ {
+ field: groupByOptions[groupBy].field,
+ headerName: groupByOptions[groupBy].headerName,
+ flex: 4,
+ renderCell: (params: any) => (
+
+ groupByOptions[groupBy].onDoubleClick(params.value)
+ }
+ title={groupByOptions[groupBy].onDoubleClickHelpText}
+ >
+ {params.value}
+
+ ),
+ },
+ { field: "count", headerName: "Count", flex: 1 },
+ {
+ field: "time",
+ headerName: "Duration",
+ flex: 1,
+ renderCell: renderTimeCell,
+ },
+ {
+ field: "cost",
+ headerName: "Cost ($)",
+ flex: 1,
+ renderCell: roundedCostCell,
+ },
+ { field: "skipped", headerName: "Skipped", flex: 1 },
+ ];
+
+ const groupByTarget = _.reduce(
+ data,
+ (acc, row) => {
+ const keys = groupByOptions[groupBy].groupByKey(row) as string[];
+ keys.forEach((key) => {
+ acc[key] = acc[key] || [];
+ acc[key].push(row);
+ });
+ return acc;
+ },
+ {} as Record
+ );
+
+ const groupedRows = _.map(groupByTarget, (rows, key) => {
+ // Sum within sha
+ const summedBySha = _.map(_.groupBy(rows, "sha"), (shaRows) => {
+ return _.reduce(
+ shaRows,
+ (acc, row) => {
+ acc.count += row.count || 0;
+ acc.time += row.time || 0;
+ acc.cost += row.cost || 0;
+ acc.skipped += row.skipped || 0;
+ return acc;
+ },
+ { count: 0, time: 0, cost: 0, skipped: 0 }
+ );
+ });
+ // the reduce across shas for average
+ return _.reduce(
+ summedBySha,
+ (acc, summed) => {
+ acc.count += summed.count;
+ acc.time += summed.time;
+ acc.cost += summed.cost;
+ acc.skipped += summed.skipped;
+ return acc;
+ },
+ {
+ id: rows[0].id,
+ file_name: rows[0].file_name,
+ job_name: rows[0].job_name,
+ labels: key,
+ count: 0,
+ time: 0,
+ cost: 0,
+ skipped: 0,
+ }
+ );
+ });
+
+ return (
+
+ Overview
+
+ This section provides an overview of the test statistics. Values are
+ summed within a commit, then averaged.
+
+
+ Pricing is approximate and weighted by workflow frequency to represent
+ the cost over an entire week. Some pricing data may be missing (ex mac,
+ rocm), in those cases the cost will be 0.
+
+
+
+ {Object.entries(groupByOptions).map(([key, setting]) => (
+
+ ))}
+
+
+
+
+
+
+ );
+}
+
+function CommitInfo({ data }: { data: any[] }) {
+ const commits = _.reduce(
+ data,
+ (acc, row) => {
+ const key = row.sha;
+ acc[key] = row.push_date;
+ return acc;
+ },
+ {} as Record
+ );
+ return (
+
+ Commits
+
+ These are the commits that are included in the report.
+
+ ({
+ id: sha,
+ push_date: pushDate,
+ }))}
+ columns={[
+ { field: "id", headerName: "SHA", flex: 1 },
+ {
+ field: "push_date",
+ headerName: "Push Date",
+ flex: 1,
+ renderCell: (params: any) => formatTimestamp(params.value),
+ },
+ ]}
+ initialState={{
+ sorting: {
+ sortModel: [{ field: "push_date", sort: "asc" }],
+ },
+ }}
+ />
+
+ );
+}
+
+function Graphs({ data }: { data: any[] }) {
+ // Map selector value to field and label
+ const groupByOptions = {
+ file: {
+ getGroupByField: (d: any) => d.file_name,
+ groupByButtonText: "Group by File",
+ },
+ job: {
+ getGroupByField: (d: any) => d.job_name,
+ groupByButtonText: "Group by Job",
+ },
+ filejob: {
+ getGroupByField: (d: any) => `${d.job_name} | ${d.file_name}`,
+ groupByButtonText: "Group by File + Job",
+ },
+ total: {
+ getGroupByField: (_: any) => `total`,
+ groupByButtonText: "Total",
+ },
+ };
+ const metricOptions = {
+ count: { label: "Count", field: "count" },
+ cost: { label: "Cost", field: "cost" },
+ duration: { label: "Duration", field: "time" },
+ skips: { label: "Skips", field: "skipped" },
+ };
+
+ const [metric, setMetric] = useState("count");
+ const [groupBy, setGroupBy] = useState("file");
+
+ const chartData = _.map(
+ // Group by the sha and the option that is selected
+ _.groupBy(data, (d) => {
+ return [d.sha, groupByOptions[groupBy].getGroupByField(d)];
+ }),
+ // Sum over each group
+ (rows, key) => {
+ return {
+ push_date: rows[0].push_date,
+ key: key.split(",")[1],
+ [metricOptions[metric].field]: _.sumBy(
+ rows,
+ (d) => d[metricOptions[metric].field]
+ ),
+ };
+ }
+ );
+ // Convert to series
+ const echartData = _.map(_.groupBy(chartData, "key"), (rows) => ({
+ name: rows[0].key,
+ type: "line",
+ data: rows.map((r) => [r.push_date, r[metricOptions[metric].field]]),
+ }));
+ const option = {
+ tooltip: { trigger: "axis" },
+ legend: {
+ type: "scroll",
+ orient: "vertical",
+ right: 10,
+ selector: [
+ {
+ type: "all",
+ title: "All",
+ },
+ {
+ type: "inverse",
+ title: "Inv",
+ },
+ ],
+ },
+ xAxis: { type: "time", name: "Push Date" },
+ yAxis: { type: "value", name: metricOptions[metric].label },
+ series: echartData,
+ };
+
+ return (
+
+ Graphs
+
+ The charts below visualize trends in the selected metric (Count, Cost,
+ Duration, or Skips) over time, grouped by file name, job name, or both.
+ Use these charts to spot regressions, improvements, or anomalies in test
+ performance across recent commits.
+
+
+ {Object.entries(metricOptions).map(([key, option]) => (
+
+ ))}
+
+
+ {Object.entries(groupByOptions).map(([key, option]) => (
+
+ ))}
+
+
+
+
+
+ );
+}
+
+function TestStatus({ data }: { data: { [key: string]: any }[] }) {
+ const columns: any[] = [
+ { field: "status", headerName: "Status", flex: 1 },
+ { field: "file_name", headerName: "File", flex: 4 },
+ { field: "test_name", headerName: "Test", flex: 4 },
+ { field: "job_name", headerName: "Job", flex: 4 },
+ { field: "sha", headerName: "SHA", flex: 1 },
+ {
+ field: "push_date",
+ headerName: "Push Date",
+ flex: 1,
+ renderCell: (params: any) => formatTimestamp(params.value),
+ },
+ ];
+
+ return (
+
+
+
+ );
+}
+
+// Custom hook to fetch real data from the local JSON file
+function useData(link: string) {
+ const [data, setData] = useState([]);
+
+ useEffect(() => {
+ fetch(link)
+ .then((response) => response.text())
+ .then((text) => {
+ const final = [];
+ for (const line of text.split("\n")) {
+ if (line.trim()) {
+ final.push(...JSON.parse(line));
+ }
+ }
+ setData(final.map((item, index) => ({ ...item, id: index })));
+ });
+ }, [link]);
+ return data;
+}
+
+export default function Page() {
+ const [fileFilter, setFileFilter] = useState("");
+ const [jobFilter, setJobFilter] = useState("");
+ const [labelFilter, setLabelFilter] = useState("");
+ const fileInputRef = useRef(null);
+ const jobInputRef = useRef(null);
+ const labelInputRef = useRef(null);
+
+ // Keep input fields in sync when filters are set programmatically
+ useEffect(() => {
+ if (fileInputRef.current && fileInputRef.current.value !== fileFilter) {
+ fileInputRef.current.value = fileFilter;
+ }
+ }, [fileFilter]);
+ useEffect(() => {
+ if (jobInputRef.current && jobInputRef.current.value !== jobFilter) {
+ jobInputRef.current.value = jobFilter;
+ }
+ }, [jobFilter]);
+ useEffect(() => {
+ if (labelInputRef.current && labelInputRef.current.value !== labelFilter) {
+ labelInputRef.current.value = labelFilter;
+ }
+ }, [labelFilter]);
+
+ const router = useRouter();
+ let data = useData("/data.json").map((item) => ({
+ // Hopefully get rid of this eventually
+ ...item,
+ file_name: item.file_name.replace(".", "/") + ".py",
+ }));
+ let statusChangeData = useData("/status_changes.json");
+
+ useEffect(() => {
+ if (router.query.label) {
+ setLabelFilter(router.query.label as string);
+ }
+ }, [router.query.label]);
+
+ if (!router.isReady) {
+ return ;
+ }
+ // Filter data by file and job name
+ data = data.filter((row) => {
+ const fileMatch = fileFilter === "" || row.file_name == fileFilter;
+ const jobMatch = jobFilter === "" || row.job_name == jobFilter;
+ const labelMatch =
+ !labelFilter || (row.labels && row.labels.includes(labelFilter));
+ return fileMatch && jobMatch && labelMatch;
+ });
+
+ // Apply the same file/job filter to statusChangeData
+ statusChangeData = statusChangeData?.filter((row) => {
+ const labelMatch =
+ !labelFilter || (row.labels && row.labels.includes(labelFilter));
+ const fileMatch = fileFilter === "" || row.file_name == fileFilter;
+ const jobMatch = jobFilter === "" || row.job_name == jobFilter;
+ return fileMatch && jobMatch && labelMatch;
+ });
+
+ return (
+
+ Test Reports
+ ) => {
+ e.preventDefault();
+ if (fileInputRef.current) setFileFilter(fileInputRef.current.value);
+ if (jobInputRef.current) setJobFilter(jobInputRef.current.value);
+ if (labelInputRef.current)
+ setLabelFilter(labelInputRef.current.value);
+ }}
+ >
+ {[
+ {
+ text: "Filter by Label",
+ inputRef: labelInputRef,
+ defaultValue: labelFilter,
+ },
+ {
+ text: "Filter by File",
+ inputRef: fileInputRef,
+ defaultValue: fileFilter,
+ },
+ {
+ text: "Filter by Job",
+ inputRef: jobInputRef,
+ defaultValue: jobFilter,
+ },
+ ].map(({ text, inputRef, defaultValue }) => (
+
+ ))}
+
+
+
+
+
+
+
+
+
+ Status Changes
+
+ This table lists the tests that were added, removed, started skipping,
+ or stopped skipping. This will only show at most 50 entries per commit
+ pair due to file size.
+
+
+
+