misc/munin/tahoe_stats

   1 #!/usr/bin/env python
   2
   3 import os
   4 import pickle
   5 import re
   6 import sys
   7 import time
   8
   9 STAT_VALIDITY = 300 # 5min limit on reporting stats
  10
  11 PLUGINS = {
  12     # LOAD AVERAGE
  13     'tahoe_runtime_load_avg':
  14         { 'statid': 'load_monitor.avg_load',
  15           'category': 'stats',
  16           'configheader': '\n'.join(['graph_title Tahoe Runtime Load Average',
  17                                      'graph_vlabel load',
  18                                      'graph_category tahoe',
  19                                      'graph_info This graph shows average reactor delay',
  20                                      ]),
  21           'graph_config': '\n'.join(['%(name)s.label %(name)s',
  22                                      '%(name)s.draw LINE1',
  23                                      ]),
  24           'graph_render': '\n'.join(['%(name)s.value %(value)s',
  25                                      ]),
  26         },
  27
  28     'tahoe_runtime_load_peak':
  29         { 'statid': 'load_monitor.max_load',
  30           'category': 'stats',
  31           'configheader': '\n'.join(['graph_title Tahoe Runtime Load Peak',
  32                                      'graph_vlabel load',
  33                                      'graph_category tahoe',
  34                                      'graph_info This graph shows peak reactor delay',
  35                                      ]),
  36           'graph_config': '\n'.join(['%(name)s.label %(name)s',
  37                                      '%(name)s.draw LINE1',
  38                                      ]),
  39           'graph_render': '\n'.join(['%(name)s.value %(value)s',
  40                                      ]),
  41         },
  42
  43     # STORAGE ALLOCATION (BYTES)
  44     'tahoe_storage_consumed':
  45         { 'statid': 'storage_server.consumed',
  46           'category': 'stats',
  47           'configheader': '\n'.join(['graph_title Tahoe Storage Server Space Consumed',
  48                                      'graph_vlabel bytes',
  49                                      'graph_category tahoe_storage_server',
  50                                      'graph_info This graph shows space consumed',
  51                                      'graph_args --base 1024',
  52                                      ]),
  53           'graph_config': '\n'.join(['%(name)s.label %(name)s',
  54                                      '%(name)s.draw LINE1',
  55                                      ]),
  56           'graph_render': '\n'.join(['%(name)s.value %(value)s',
  57                                      ]),
  58         },
  59
  60     'tahoe_storage_allocated':
  61         { 'statid': 'storage_server.allocated',
  62           'category': 'stats',
  63           'configheader': '\n'.join(['graph_title Tahoe Storage Server Space Allocated',
  64                                      'graph_vlabel bytes',
  65                                      'graph_category tahoe_storage_server',
  66                                      'graph_info This graph shows space allocated',
  67                                      'graph_args --base 1024',
  68                                      ]),
  69           'graph_config': '\n'.join(['%(name)s.label %(name)s',
  70                                      '%(name)s.draw LINE1',
  71                                      ]),
  72           'graph_render': '\n'.join(['%(name)s.value %(value)s',
  73                                      ]),
  74         },
  75
  76     'tahoe_storage_bytes_added':
  77         { 'statid': 'storage_server.bytes_added',
  78           'category': 'counters',
  79           'configheader': '\n'.join(['graph_title Tahoe Storage Server Bytes Added',
  80                                      'graph_vlabel bytes',
  81                                      'graph_category tahoe_storage_server',
  82                                      'graph_info This graph shows cummulative bytes added',
  83                                      ]),
  84           'graph_config': '\n'.join(['%(name)s.label %(name)s',
  85                                      '%(name)s.draw LINE1',
  86                                      ]),
  87           'graph_render': '\n'.join(['%(name)s.value %(value)s',
  88                                      ]),
  89         },
  90
  91     'tahoe_storage_bytes_freed':
  92         { 'statid': 'storage_server.bytes_freed',
  93           'category': 'counters',
  94           'configheader': '\n'.join(['graph_title Tahoe Storage Server Bytes Removed',
  95                                      'graph_vlabel bytes',
  96                                      'graph_category tahoe_storage_server',
  97                                      'graph_info This graph shows cummulative bytes removed',
  98                                      ]),
  99           'graph_config': '\n'.join(['%(name)s.label %(name)s',
 100                                      '%(name)s.draw LINE1',
 101                                      ]),
 102           'graph_render': '\n'.join(['%(name)s.value %(value)s',
 103                                      ]),
 104         },
 105
 106     'tahoe_storage_operations_allocate':
 107         { 'statid': 'storage_server.allocate',
 108           'category': 'counters',
 109           'configheader': '\n'.join(['graph_title Tahoe Storage Server Allocate_Bucket Operations',
 110                                      'graph_vlabel operations per second',
 111                                      'graph_category tahoe_storage_server',
 112                                      'graph_info This graph shows how many allocate_buckets operations occured per second. Each immutable file upload causes one such operation per server.',
 113                                      ]),
 114           'graph_config': '\n'.join(['%(name)s.label %(name)s',
 115                                      '%(name)s.type DERIVE',
 116                                      '%(name)s.min 0',
 117                                      '%(name)s.draw LINE1',
 118                                      ]),
 119           'graph_render': '\n'.join(['%(name)s.value %(value)s',
 120                                      ]),
 121         },
 122
 123     'tahoe_storage_operations_get':
 124         { 'statid': 'storage_server.get',
 125           'category': 'counters',
 126           'configheader': '\n'.join(['graph_title Tahoe Storage Server get_bucket Operations',
 127                                      'graph_vlabel operations per second',
 128                                      'graph_category tahoe_storage_server',
 129                                      'graph_info This graph shows how many get_bucket operations occured per second. Each immutable file download/check causes one such operation per server.',
 130                                      ]),
 131           'graph_config': '\n'.join(['%(name)s.label %(name)s',
 132                                      '%(name)s.type DERIVE',
 133                                      '%(name)s.min 0',
 134                                      '%(name)s.draw LINE1',
 135                                      ]),
 136           'graph_render': '\n'.join(['%(name)s.value %(value)s',
 137                                      ]),
 138         },
 139
 140     'tahoe_storage_operations_writev':
 141         { 'statid': 'storage_server.writev',
 142           'category': 'counters',
 143           'configheader': '\n'.join(['graph_title Tahoe Storage Server writev Operations',
 144                                      'graph_vlabel operations per second',
 145                                      'graph_category tahoe_storage_server',
 146                                      'graph_info This graph shows how many writev operations occured per second. Each mutable file / dirnode write causes one such operation per server.',
 147                                      ]),
 148           'graph_config': '\n'.join(['%(name)s.label %(name)s',
 149                                      '%(name)s.type DERIVE',
 150                                      '%(name)s.min 0',
 151                                      '%(name)s.draw LINE1',
 152                                      ]),
 153           'graph_render': '\n'.join(['%(name)s.value %(value)s',
 154                                      ]),
 155         },
 156
 157     'tahoe_storage_operations_readv':
 158         { 'statid': 'storage_server.readv',
 159           'category': 'counters',
 160           'configheader': '\n'.join(['graph_title Tahoe Storage Server readv Operations',
 161                                      'graph_vlabel operations per second',
 162                                      'graph_category tahoe_storage_server',
 163                                      'graph_info This graph shows how many readv operations occured per second. Each dirnode read causes one such operation per server.',
 164                                      ]),
 165           'graph_config': '\n'.join(['%(name)s.label %(name)s',
 166                                      '%(name)s.type DERIVE',
 167                                      '%(name)s.min 0',
 168                                      '%(name)s.draw LINE1',
 169                                      ]),
 170           'graph_render': '\n'.join(['%(name)s.value %(value)s',
 171                                      ]),
 172         },
 173
 174
 175     # HELPER
 176     'tahoe_helper_incoming_files':
 177         { 'statid': 'chk_upload_helper.incoming_count',
 178           'category': 'stats',
 179           'configheader': '\n'.join(['graph_title Tahoe Upload Helper Incoming File Count',
 180                                      'graph_vlabel n files',
 181                                      'graph_category tahoe_helper',
 182                                      'graph_info This graph shows number of incoming files',
 183                                      ]),
 184           'graph_config': '\n'.join(['%(name)s.label %(name)s',
 185                                      '%(name)s.draw LINE1',
 186                                      ]),
 187           'graph_render': '\n'.join(['%(name)s.value %(value)s',
 188                                      ]),
 189         },
 190     'tahoe_helper_incoming_filesize':
 191         { 'statid': 'chk_upload_helper.incoming_size',
 192           'category': 'stats',
 193           'configheader': '\n'.join(['graph_title Tahoe Upload Helper Incoming File Size',
 194                                      'graph_vlabel bytes',
 195                                      'graph_category tahoe_helper',
 196                                      'graph_info This graph shows total size of incoming files',
 197                                      ]),
 198           'graph_config': '\n'.join(['%(name)s.label %(name)s',
 199                                      '%(name)s.draw LINE1',
 200                                      ]),
 201           'graph_render': '\n'.join(['%(name)s.value %(value)s',
 202                                      ]),
 203         },
 204     'tahoe_helper_incoming_files_old':
 205         { 'statid': 'chk_upload_helper.incoming_size_old',
 206           'category': 'stats',
 207           'configheader': '\n'.join(['graph_title Tahoe Upload Helper Incoming Old Files',
 208                                      'graph_vlabel bytes',
 209                                      'graph_category tahoe_helper',
 210                                      'graph_info This graph shows total size of old incoming files',
 211                                      ]),
 212           'graph_config': '\n'.join(['%(name)s.label %(name)s',
 213                                      '%(name)s.draw LINE1',
 214                                      ]),
 215           'graph_render': '\n'.join(['%(name)s.value %(value)s',
 216                                      ]),
 217         },
 218
 219     'tahoe_helper_encoding_files':
 220         { 'statid': 'chk_upload_helper.encoding_count',
 221           'category': 'stats',
 222           'configheader': '\n'.join(['graph_title Tahoe Upload Helper Encoding File Count',
 223                                      'graph_vlabel n files',
 224                                      'graph_category tahoe_helper',
 225                                      'graph_info This graph shows number of encoding files',
 226                                      ]),
 227           'graph_config': '\n'.join(['%(name)s.label %(name)s',
 228                                      '%(name)s.draw LINE1',
 229                                      ]),
 230           'graph_render': '\n'.join(['%(name)s.value %(value)s',
 231                                      ]),
 232         },
 233     'tahoe_helper_encoding_filesize':
 234         { 'statid': 'chk_upload_helper.encoding_size',
 235           'category': 'stats',
 236           'configheader': '\n'.join(['graph_title Tahoe Upload Helper Encoding File Size',
 237                                      'graph_vlabel bytes',
 238                                      'graph_category tahoe_helper',
 239                                      'graph_info This graph shows total size of encoding files',
 240                                      ]),
 241           'graph_config': '\n'.join(['%(name)s.label %(name)s',
 242                                      '%(name)s.draw LINE1',
 243                                      ]),
 244           'graph_render': '\n'.join(['%(name)s.value %(value)s',
 245                                      ]),
 246         },
 247     'tahoe_helper_encoding_files_old':
 248         { 'statid': 'chk_upload_helper.encoding_size_old',
 249           'category': 'stats',
 250           'configheader': '\n'.join(['graph_title Tahoe Upload Helper Encoding Old Files',
 251                                      'graph_vlabel bytes',
 252                                      'graph_category tahoe_helper',
 253                                      'graph_info This graph shows total size of old encoding files',
 254                                      ]),
 255           'graph_config': '\n'.join(['%(name)s.label %(name)s',
 256                                      '%(name)s.draw LINE1',
 257                                      ]),
 258           'graph_render': '\n'.join(['%(name)s.value %(value)s',
 259                                      ]),
 260         },
 261
 262     'tahoe_helper_active_uploads':
 263         { 'statid': 'chk_upload_helper.active_uploads',
 264           'category': 'stats',
 265           'configheader': '\n'.join(['graph_title Tahoe Upload Helper Active Files',
 266                                      'graph_vlabel n files',
 267                                      'graph_category tahoe_helper',
 268                                      'graph_info This graph shows number of files actively being processed by the helper',
 269                                      ]),
 270           'graph_config': '\n'.join(['%(name)s.label %(name)s',
 271                                      '%(name)s.draw LINE1',
 272                                      ]),
 273           'graph_render': '\n'.join(['%(name)s.value %(value)s',
 274                                      ]),
 275         },
 276
 277     'tahoe_helper_upload_requests':
 278         { 'statid': 'chk_upload_helper.upload_requests',
 279           'category': 'counters',
 280           'configheader': '\n'.join(['graph_title Tahoe Upload Helper Upload Requests',
 281                                      'graph_vlabel requests',
 282                                      'graph_category tahoe_helper',
 283                                      'graph_info This graph shows the number of upload requests arriving at the helper',
 284                                      ]),
 285           'graph_config': '\n'.join(['%(name)s.label %(name)s',
 286                                      '%(name)s.type DERIVE',
 287                                      '%(name)s.min 0',
 288                                      '%(name)s.draw LINE1',
 289                                      ]),
 290           'graph_render': '\n'.join(['%(name)s.value %(value)s',
 291                                      ]),
 292         },
 293     'tahoe_helper_upload_already_present':
 294         { 'statid': 'chk_upload_helper.upload_already_present',
 295           'category': 'counters',
 296           'configheader': '\n'.join(['graph_title Tahoe Upload Helper Uploads Already Present',
 297                                      'graph_vlabel requests',
 298                                      'graph_category tahoe_helper',
 299                                      'graph_info This graph shows the number of uploads whose files are already present in the grid',
 300                                      ]),
 301           'graph_config': '\n'.join(['%(name)s.label %(name)s',
 302                                      '%(name)s.type DERIVE',
 303                                      '%(name)s.min 0',
 304                                      '%(name)s.draw LINE1',
 305                                      ]),
 306           'graph_render': '\n'.join(['%(name)s.value %(value)s',
 307                                      ]),
 308         },
 309     'tahoe_helper_upload_need_upload':
 310         { 'statid': 'chk_upload_helper.upload_need_upload',
 311           'category': 'counters',
 312           'configheader': '\n'.join(['graph_title Tahoe Upload Helper Uploads Needing Upload',
 313                                      'graph_vlabel requests',
 314                                      'graph_category tahoe_helper',
 315                                      'graph_info This graph shows the number of uploads whose files are not already present in the grid',
 316                                      ]),
 317           'graph_config': '\n'.join(['%(name)s.label %(name)s',
 318                                      '%(name)s.type DERIVE',
 319                                      '%(name)s.min 0',
 320                                      '%(name)s.draw LINE1',
 321                                      ]),
 322           'graph_render': '\n'.join(['%(name)s.value %(value)s',
 323                                      ]),
 324         },
 325     'tahoe_helper_encoded_bytes':
 326         { 'statid': 'chk_upload_helper.encoded_bytes',
 327           'category': 'counters',
 328           'configheader': '\n'.join(['graph_title Tahoe Upload Helper Encoded Bytes',
 329                                      'graph_vlabel bytes',
 330                                      'graph_category tahoe_helper',
 331                                      'graph_info This graph shows the number of bytes encoded by the helper',
 332                                      ]),
 333           'graph_config': '\n'.join(['%(name)s.label %(name)s',
 334                                      '%(name)s.type DERIVE',
 335                                      '%(name)s.min 0',
 336                                      '%(name)s.draw LINE1',
 337                                      ]),
 338           'graph_render': '\n'.join(['%(name)s.value %(value)s',
 339                                      ]),
 340         },
 341     'tahoe_helper_fetched_bytes':
 342         { 'statid': 'chk_upload_helper.fetched_bytes',
 343           'category': 'counters',
 344           'configheader': '\n'.join(['graph_title Tahoe Upload Helper Fetched Bytes',
 345                                      'graph_vlabel bytes',
 346                                      'graph_category tahoe_helper',
 347                                      'graph_info This graph shows the number of bytes fetched by the helper',
 348                                      ]),
 349           'graph_config': '\n'.join(['%(name)s.label %(name)s',
 350                                      '%(name)s.type DERIVE',
 351                                      '%(name)s.min 0',
 352                                      '%(name)s.draw LINE1',
 353                                      ]),
 354           'graph_render': '\n'.join(['%(name)s.value %(value)s',
 355                                      ]),
 356         },
 357
 358     # WEBAPI
 359     'tahoe_uploader_bytes_uploaded':
 360         { 'statid': 'uploader.bytes_uploaded',
 361           'category': 'counters',
 362           'configheader': '\n'.join(['graph_title Tahoe Uploader Bytes Uploaded',
 363                                      'graph_vlabel bytes',
 364                                      'graph_category tahoe_traffic',
 365                                      'graph_info This graph shows the number of bytes uploaded',
 366                                      ]),
 367           'graph_config': '\n'.join(['%(name)s.label %(name)s',
 368                                      '%(name)s.type DERIVE',
 369                                      '%(name)s.min 0',
 370                                      '%(name)s.draw LINE1',
 371                                      ]),
 372           'graph_render': '\n'.join(['%(name)s.value %(value)s',
 373                                      ]),
 374         },
 375     'tahoe_uploader_files_uploaded':
 376         { 'statid': 'uploader.files_uploaded',
 377           'category': 'counters',
 378           'configheader': '\n'.join(['graph_title Tahoe Uploader Bytes Uploaded',
 379                                      'graph_vlabel files',
 380                                      'graph_category tahoe_traffic',
 381                                      'graph_info This graph shows the number of files uploaded',
 382                                      ]),
 383           'graph_config': '\n'.join(['%(name)s.label %(name)s',
 384                                      '%(name)s.type DERIVE',
 385                                      '%(name)s.min 0',
 386                                      '%(name)s.draw LINE1',
 387                                      ]),
 388           'graph_render': '\n'.join(['%(name)s.value %(value)s',
 389                                      ]),
 390         },
 391     'tahoe_mutable_files_published':
 392         { 'statid': 'mutable.files_published',
 393           'category': 'counters',
 394           'configheader': '\n'.join(['graph_title Tahoe Mutable Files Published',
 395                                      'graph_vlabel files',
 396                                      'graph_category tahoe_traffic',
 397                                      'graph_info This graph shows the number of mutable files published',
 398                                      ]),
 399           'graph_config': '\n'.join(['%(name)s.label %(name)s',
 400                                      '%(name)s.type DERIVE',
 401                                      '%(name)s.min 0',
 402                                      '%(name)s.draw LINE1',
 403                                      ]),
 404           'graph_render': '\n'.join(['%(name)s.value %(value)s',
 405                                      ]),
 406         },
 407     'tahoe_mutable_files_retrieved':
 408         { 'statid': 'mutable.files_retrieved',
 409           'category': 'counters',
 410           'configheader': '\n'.join(['graph_title Tahoe Mutable Files Retrieved',
 411                                      'graph_vlabel files',
 412                                      'graph_category tahoe_traffic',
 413                                      'graph_info This graph shows the number of files retrieved',
 414                                      ]),
 415           'graph_config': '\n'.join(['%(name)s.label %(name)s',
 416                                      '%(name)s.type DERIVE',
 417                                      '%(name)s.min 0',
 418                                      '%(name)s.draw LINE1',
 419                                      ]),
 420           'graph_render': '\n'.join(['%(name)s.value %(value)s',
 421                                      ]),
 422         },
 423
 424     }
 425
 426 def smash_name(name):
 427     return re.sub('[^a-zA-Z0-9]', '_', name)
 428
 429 def open_stats(fname):
 430     f = open(fname, 'rb')
 431     stats = pickle.load(f)
 432     f.close()
 433     return stats
 434
 435 def main(argv):
 436     graph_name = os.path.basename(argv[0])
 437     if graph_name.endswith('.py'):
 438         graph_name = graph_name[:-3]
 439
 440     plugin_conf = PLUGINS.get(graph_name)
 441
 442     for k,v in os.environ.items():
 443         if k.startswith('statsfile'):
 444             stats_file = v
 445             break
 446     else:
 447         raise RuntimeError("No 'statsfile' env var found")
 448
 449     stats = open_stats(stats_file)
 450
 451     now = time.time()
 452     def output_nodes(output_section, check_time):
 453         for tubid, nodestats in stats.items():
 454             if check_time and (now - nodestats.get('timestamp', 0)) > STAT_VALIDITY:
 455                 continue
 456             name = smash_name("%s_%s" % (nodestats['nickname'], tubid[:4]))
 457             #value = nodestats['stats'][plugin_conf['category']].get(plugin_conf['statid'])
 458             category = plugin_conf['category']
 459             statid = plugin_conf['statid']
 460             value = nodestats['stats'][category].get(statid)
 461             if value is not None:
 462                 args = { 'name': name, 'value': value }
 463                 print plugin_conf[output_section] % args
 464
 465     if len(argv) > 1:
 466         if sys.argv[1] == 'config':
 467             print plugin_conf['configheader']
 468             output_nodes('graph_config', False)
 469             sys.exit(0)
 470
 471     output_nodes('graph_render', True)
 472
 473 if __name__ == '__main__':
 474     main(sys.argv)