|
| 1 | +""" |
| 2 | +Usage examples: |
| 3 | + # Basic usage - show aggregated cache percentages for all vector index files |
| 4 | + python smaps-view.py $(pidof qdrant) '.*/vector_index/.*' |
| 5 | +
|
| 6 | + # Verbose mode - show individual files larger than 1MB |
| 7 | + python smaps-view.py $(pidof qdrant) '.*/vector_storage/.*' -v |
| 8 | +
|
| 9 | +Example output: |
| 10 | + Cache percentages for pattern '.vector_storage.*': |
| 11 | + r--s: 0.00% |
| 12 | + rw-s: 0.06% |
| 13 | +""" |
| 14 | + |
| 15 | +import re |
| 16 | +import sys |
| 17 | +import argparse |
| 18 | +from typing import Dict, Tuple, List |
| 19 | +from dataclasses import dataclass |
| 20 | + |
| 21 | + |
| 22 | +# Example of smap format: |
| 23 | +# |
| 24 | +# |
| 25 | +# 7de883800000-7de885800000 r--s 00000000 00:33 26610006 /qdrant/storage/collections/benchmark/0/segments/6250c760-ccaf-468e-a96f-8e2b03f8c524/vector_storage/vectors/chunk_2.mmap |
| 26 | +# Size: 32768 kB |
| 27 | +# KernelPageSize: 4 kB |
| 28 | +# MMUPageSize: 4 kB |
| 29 | +# Rss: 0 kB |
| 30 | +# Pss: 0 kB |
| 31 | +# Pss_Dirty: 0 kB |
| 32 | +# Shared_Clean: 0 kB |
| 33 | +# Shared_Dirty: 0 kB |
| 34 | +# Private_Clean: 0 kB |
| 35 | +# Private_Dirty: 0 kB |
| 36 | +# Referenced: 0 kB |
| 37 | +# Anonymous: 0 kB |
| 38 | +# KSM: 0 kB |
| 39 | +# LazyFree: 0 kB |
| 40 | +# AnonHugePages: 0 kB |
| 41 | +# ShmemPmdMapped: 0 kB |
| 42 | +# FilePmdMapped: 0 kB |
| 43 | +# Shared_Hugetlb: 0 kB |
| 44 | +# Private_Hugetlb: 0 kB |
| 45 | +# Swap: 0 kB |
| 46 | +# SwapPss: 0 kB |
| 47 | +# Locked: 0 kB |
| 48 | +# THPeligible: 0 |
| 49 | +# ProtectionKey: 0 |
| 50 | +# VmFlags: rd sh mr mw me ms sr sd |
| 51 | +# |
| 52 | + |
| 53 | + |
| 54 | + |
| 55 | +@dataclass |
| 56 | +class MapStats: |
| 57 | + size: int = 0 |
| 58 | + rss: int = 0 |
| 59 | + |
| 60 | +def parse_smaps(smaps_content: str, pattern: str, verbose: bool = False) -> Dict[Tuple[str, str], MapStats]: |
| 61 | + """ |
| 62 | + Parse smaps content and calculate size and RSS for files matching the pattern. |
| 63 | + Optionally logs individual file percentages for files larger than 1MB. |
| 64 | + |
| 65 | + Args: |
| 66 | + smaps_content: Content of /proc/{pid}/smaps file |
| 67 | + pattern: Regular expression pattern to match against file paths |
| 68 | + verbose: Whether to print per-file information |
| 69 | + |
| 70 | + Returns: |
| 71 | + Dictionary mapping (file_path, permissions) tuples to their stats |
| 72 | + """ |
| 73 | + file_stats: Dict[Tuple[str, str], MapStats] = {} |
| 74 | + current_file = None |
| 75 | + current_size = 0 |
| 76 | + current_rss = 0 |
| 77 | + current_permissions = "" |
| 78 | + |
| 79 | + # Compile the regex pattern |
| 80 | + regex = re.compile(pattern) |
| 81 | + |
| 82 | + for line in smaps_content.splitlines(): |
| 83 | + # Check if line starts with an address range (new memory map entry) |
| 84 | + if re.match(r'^[0-9a-f]+-[0-9a-f]+', line): |
| 85 | + # If we have a previous file that matched, add its stats |
| 86 | + if current_file and regex.search(current_file): |
| 87 | + key = (current_file, current_permissions) |
| 88 | + if key not in file_stats: |
| 89 | + file_stats[key] = MapStats() |
| 90 | + |
| 91 | + stats = file_stats[key] |
| 92 | + stats.size += current_size |
| 93 | + stats.rss += current_rss |
| 94 | + |
| 95 | + # Log individual file percentage if verbose mode is enabled and file is larger than 1MB |
| 96 | + if verbose and current_size > 1024: |
| 97 | + percentage = (current_rss / current_size) * 100 |
| 98 | + print(f"File: {current_file} ({current_permissions})") |
| 99 | + print(f" Size: {current_size} kB") |
| 100 | + print(f" RSS: {current_rss} kB") |
| 101 | + print(f" Cache percentage: {percentage:.2f}%") |
| 102 | + print() |
| 103 | + |
| 104 | + # Extract file path and permissions from the line |
| 105 | + parts = line.split() |
| 106 | + current_file = parts[-1] if len(parts) > 5 else None |
| 107 | + current_size = 0 |
| 108 | + current_rss = 0 |
| 109 | + # Store full permissions string |
| 110 | + current_permissions = parts[1] if len(parts) > 1 else "" |
| 111 | + elif line.startswith('Size:'): |
| 112 | + current_size = int(line.split()[1]) |
| 113 | + elif line.startswith('Rss:'): |
| 114 | + current_rss = int(line.split()[1]) |
| 115 | + |
| 116 | + # Don't forget to add the last entry if it matches |
| 117 | + if current_file and regex.search(current_file): |
| 118 | + key = (current_file, current_permissions) |
| 119 | + if key not in file_stats: |
| 120 | + file_stats[key] = MapStats() |
| 121 | + |
| 122 | + stats = file_stats[key] |
| 123 | + stats.size += current_size |
| 124 | + stats.rss += current_rss |
| 125 | + |
| 126 | + # Log individual file percentage for the last entry if verbose mode is enabled and file is larger than 1MB |
| 127 | + if verbose and current_size > 1024: |
| 128 | + percentage = (current_rss / current_size) * 100 if current_size > 0 else 0 |
| 129 | + print(f"File: {current_file} ({current_permissions})") |
| 130 | + print(f" Size: {current_size} kB") |
| 131 | + print(f" RSS: {current_rss} kB") |
| 132 | + print(f" Cache percentage: {percentage:.2f}%") |
| 133 | + print() |
| 134 | + |
| 135 | + return file_stats |
| 136 | + |
| 137 | +def calculate_cache_percentage(pid: int, pattern: str, verbose: bool = False) -> Dict[str, float]: |
| 138 | + """ |
| 139 | + Calculate the percentage of memory that is cached for files matching the pattern. |
| 140 | + Separates by permission type. |
| 141 | + |
| 142 | + Args: |
| 143 | + pid: Process ID to analyze |
| 144 | + pattern: Regular expression pattern to match against file paths |
| 145 | + verbose: Whether to print per-file information |
| 146 | + |
| 147 | + Returns: |
| 148 | + Dictionary mapping permission strings to their cache percentages |
| 149 | + """ |
| 150 | + try: |
| 151 | + with open(f'/proc/{pid}/smaps', 'r') as f: |
| 152 | + content = f.read() |
| 153 | + |
| 154 | + file_stats = parse_smaps(content, pattern, verbose) |
| 155 | + |
| 156 | + if not file_stats: |
| 157 | + return {} |
| 158 | + |
| 159 | + # Calculate percentages for each permission type |
| 160 | + permission_stats: Dict[str, Tuple[int, int]] = {} |
| 161 | + for (_, perms), stats in file_stats.items(): |
| 162 | + if perms not in permission_stats: |
| 163 | + permission_stats[perms] = (0, 0) |
| 164 | + size, rss = permission_stats[perms] |
| 165 | + permission_stats[perms] = (size + stats.size, rss + stats.rss) |
| 166 | + |
| 167 | + # Calculate percentages |
| 168 | + percentages = {} |
| 169 | + for perms, (size, rss) in permission_stats.items(): |
| 170 | + if size > 0: |
| 171 | + percentages[perms] = (rss / size) * 100 |
| 172 | + else: |
| 173 | + percentages[perms] = 0.0 |
| 174 | + |
| 175 | + return percentages |
| 176 | + except FileNotFoundError: |
| 177 | + print(f"Error: Could not find /proc/{pid}/smaps", file=sys.stderr) |
| 178 | + return {} |
| 179 | + except Exception as e: |
| 180 | + print(f"Error processing smaps: {e}", file=sys.stderr) |
| 181 | + return {} |
| 182 | + |
| 183 | +if __name__ == '__main__': |
| 184 | + parser = argparse.ArgumentParser(description='Analyze memory maps and calculate cache percentages') |
| 185 | + parser.add_argument('pid', type=int, help='Process ID to analyze') |
| 186 | + parser.add_argument('pattern', type=str, help='Regular expression pattern to match against file paths') |
| 187 | + parser.add_argument('-v', '--verbose', action='store_true', help='Print per-file information') |
| 188 | + |
| 189 | + args = parser.parse_args() |
| 190 | + |
| 191 | + percentages = calculate_cache_percentage(args.pid, args.pattern, args.verbose) |
| 192 | + print(f"Cache percentages for pattern '{args.pattern}':") |
| 193 | + for perms, percentage in percentages.items(): |
| 194 | + print(f" {perms}: {percentage:.2f}%") |
0 commit comments