kernel/drivers/staging/lustre/sysfs-fs-lustre

   1 What:           /sys/fs/lustre/version
   2 Date:           May 2015
   3 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
   4 Description:
   5                 Shows current running lustre version.
   6
   7 What:           /sys/fs/lustre/pinger
   8 Date:           May 2015
   9 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
  10 Description:
  11                 Shows if the lustre module has pinger support.
  12                 "on" means yes and "off" means no.
  13
  14 What:           /sys/fs/lustre/health
  15 Date:           May 2015
  16 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
  17 Description:
  18                 Shows whenever current system state believed to be "healthy",
  19                 "NOT HEALTHY", or "LBUG" whenever lustre has experienced
  20                 an internal assertion failure
  21
  22 What:           /sys/fs/lustre/jobid_name
  23 Date:           May 2015
  24 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
  25 Description:
  26                 Currently running job "name" for this node to be transferred
  27                 to Lustre servers for purposes of QoS and statistics gathering.
  28                 Writing into this file will change the name, reading outputs
  29                 currently set value.
  30
  31 What:           /sys/fs/lustre/jobid_var
  32 Date:           May 2015
  33 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
  34 Description:
  35                 Control file for lustre "jobstats" functionality, write new
  36                 value from the list below to change the mode:
  37                 disable - disable job name reporting to the servers (default)
  38                 procname_uid - form the job name as the current running
  39                                command name and pid with a dot in between
  40                                e.g. dd.1253
  41                 nodelocal - use jobid_name value from above.
  42
  43 What:           /sys/fs/lustre/timeout
  44 Date:           June 2015
  45 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
  46 Description:
  47                 Controls "lustre timeout" variable, also known as obd_timeout
  48                 in some old manual. In the past obd_timeout was of paramount
  49                 importance as the timeout value used everywhere and where
  50                 other timeouts were derived from. These days it's much less
  51                 important as network timeouts are mostly determined by
  52                 AT (adaptive timeouts).
  53                 Unit: seconds, default: 100
  54
  55 What:           /sys/fs/lustre/max_dirty_mb
  56 Date:           June 2015
  57 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
  58 Description:
  59                 Controls total number of dirty cache (in megabytes) allowed
  60                 across all mounted lustre filesystems.
  61                 Since writeout of dirty pages in Lustre is somewhat expensive,
  62                 when you allow to many dirty pages, this might lead to
  63                 performance degradations as kernel tries to desperately
  64                 find some pages to free/writeout.
  65                 Default 1/2 RAM. Min value 4, max value 9/10 of RAM.
  66
  67 What:           /sys/fs/lustre/debug_peer_on_timeout
  68 Date:           June 2015
  69 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
  70 Description:
  71                 Control if lnet debug information should be printed when
  72                 an RPC timeout occurs.
  73                 0 disabled (default)
  74                 1 enabled
  75
  76 What:           /sys/fs/lustre/dump_on_timeout
  77 Date:           June 2015
  78 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
  79 Description:
  80                 Controls if Lustre debug log should be dumped when an RPC
  81                 timeout occurs. This is useful if yout debug buffer typically
  82                 rolls over by the time you notice RPC timeouts.
  83
  84 What:           /sys/fs/lustre/dump_on_eviction
  85 Date:           June 2015
  86 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
  87 Description:
  88                 Controls if Lustre debug log should be dumped when an this
  89                 client is evicted from one of the servers.
  90                 This is useful if yout debug buffer typically rolls over
  91                  by the time you notice the eviction event.
  92
  93 What:           /sys/fs/lustre/at_min
  94 Date:           July 2015
  95 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
  96 Description:
  97                 Controls minimum adaptive timeout in seconds. If you encounter
  98                 a case where clients timeout due to server-reported processing
  99                 time being too short, you might consider increasing this value.
 100                 One common case of this if the underlying network has
 101                 unpredictable long delays.
 102                 Default: 0
 103
 104 What:           /sys/fs/lustre/at_max
 105 Date:           July 2015
 106 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 107 Description:
 108                 Controls maximum adaptive timeout in seconds. If at_max timeout
 109                 is reached for an RPC, the RPC will time out.
 110                 Some genuinuely slow network hardware might warrant increasing
 111                 this value.
 112                 Setting this value to 0 disables Adaptive Timeouts
 113                 functionality and old-style obd_timeout value is then used.
 114                 Default: 600
 115
 116 What:           /sys/fs/lustre/at_extra
 117 Date:           July 2015
 118 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 119 Description:
 120                 Controls how much extra time to request for unfinished requests
 121                 in processing in seconds. Normally a server-side parameter, it
 122                 is also used on the client for responses to various LDLM ASTs
 123                 that are handled with a special server thread on the client.
 124                 This is a way for the servers to ask the clients not to time
 125                 out the request that reached current servicing time estimate
 126                 yet and give it some more time.
 127                 Default: 30
 128
 129 What:           /sys/fs/lustre/at_early_margin
 130 Date:           July 2015
 131 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 132 Description:
 133                 Controls when to send the early reply for requests that are
 134                 about to timeout as an offset to the estimated service time in
 135                 seconds..
 136                 Default: 5
 137
 138 What:           /sys/fs/lustre/at_history
 139 Date:           July 2015
 140 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 141 Description:
 142                 Controls for how many seconds to remember slowest events
 143                 encountered by adaptive timeouts code.
 144                 Default: 600
 145
 146 What:           /sys/fs/lustre/llite/<fsname>-<uuid>/blocksize
 147 Date:           May 2015
 148 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 149 Description:
 150                 Biggest blocksize on object storage server for this filesystem.
 151
 152 What:           /sys/fs/lustre/llite/<fsname>-<uuid>/kbytestotal
 153 Date:           May 2015
 154 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 155 Description:
 156                 Shows total number of kilobytes of space on this filesystem
 157
 158 What:           /sys/fs/lustre/llite/<fsname>-<uuid>/kbytesfree
 159 Date:           May 2015
 160 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 161 Description:
 162                 Shows total number of free kilobytes of space on this filesystem
 163
 164 What:           /sys/fs/lustre/llite/<fsname>-<uuid>/kbytesavail
 165 Date:           May 2015
 166 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 167 Description:
 168                 Shows total number of free kilobytes of space on this filesystem
 169                 actually available for use (taking into account per-client
 170                 grants and filesystem reservations).
 171
 172 What:           /sys/fs/lustre/llite/<fsname>-<uuid>/filestotal
 173 Date:           May 2015
 174 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 175 Description:
 176                 Shows total number of inodes on the filesystem.
 177
 178 What:           /sys/fs/lustre/llite/<fsname>-<uuid>/filesfree
 179 Date:           May 2015
 180 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 181 Description:
 182                 Shows estimated number of free inodes on the filesystem
 183
 184 What:           /sys/fs/lustre/llite/<fsname>-<uuid>/client_type
 185 Date:           May 2015
 186 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 187 Description:
 188                 Shows whenever this filesystem considers this client to be
 189                 compute cluster-local or remote. Remote clients have
 190                 additional uid/gid convrting logic applied.
 191
 192 What:           /sys/fs/lustre/llite/<fsname>-<uuid>/fstype
 193 Date:           May 2015
 194 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 195 Description:
 196                 Shows filesystem type of the filesystem
 197
 198 What:           /sys/fs/lustre/llite/<fsname>-<uuid>/uuid
 199 Date:           May 2015
 200 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 201 Description:
 202                 Shows this filesystem superblock uuid
 203
 204 What:           /sys/fs/lustre/llite/<fsname>-<uuid>/max_read_ahead_mb
 205 Date:           May 2015
 206 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 207 Description:
 208                 Sets maximum number of megabytes in system memory to be
 209                 given to read-ahead cache.
 210
 211 What:           /sys/fs/lustre/llite/<fsname>-<uuid>/max_read_ahead_per_file_mb
 212 Date:           May 2015
 213 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 214 Description:
 215                 Sets maximum number of megabytes to read-ahead for a single file
 216
 217 What:           /sys/fs/lustre/llite/<fsname>-<uuid>/max_read_ahead_whole_mb
 218 Date:           May 2015
 219 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 220 Description:
 221                 For small reads, how many megabytes to actually request from
 222                 the server as initial read-ahead.
 223
 224 What:           /sys/fs/lustre/llite/<fsname>-<uuid>/checksum_pages
 225 Date:           May 2015
 226 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 227 Description:
 228                 Enables or disables per-page checksum at llite layer, before
 229                 the pages are actually given to lower level for network transfer
 230
 231 What:           /sys/fs/lustre/llite/<fsname>-<uuid>/stats_track_pid
 232 Date:           May 2015
 233 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 234 Description:
 235                 Limit Lustre vfs operations gathering to just a single pid.
 236                 0 to track everything.
 237
 238 What:           /sys/fs/lustre/llite/<fsname>-<uuid>/stats_track_ppid
 239 Date:           May 2015
 240 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 241 Description:
 242                 Limit Lustre vfs operations gathering to just a single ppid.
 243                 0 to track everything.
 244
 245 What:           /sys/fs/lustre/llite/<fsname>-<uuid>/stats_track_gid
 246 Date:           May 2015
 247 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 248 Description:
 249                 Limit Lustre vfs operations gathering to just a single gid.
 250                 0 to track everything.
 251
 252 What:           /sys/fs/lustre/llite/<fsname>-<uuid>/statahead_max
 253 Date:           May 2015
 254 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 255 Description:
 256                 Controls maximum number of statahead requests to send when
 257                 sequential readdir+stat pattern is detected.
 258
 259 What:           /sys/fs/lustre/llite/<fsname>-<uuid>/statahead_agl
 260 Date:           May 2015
 261 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 262 Description:
 263                 Controls if AGL (async glimpse ahead - obtain object information
 264                 from OSTs in parallel with MDS during statahead) should be
 265                 enabled or disabled.
 266                 0 to disable, 1 to enable.
 267
 268 What:           /sys/fs/lustre/llite/<fsname>-<uuid>/lazystatfs
 269 Date:           May 2015
 270 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 271 Description:
 272                 Controls statfs(2) behaviour in the face of down servers.
 273                 If 0, always wait for all servers to come online,
 274                 if 1, ignote inactive servers.
 275
 276 What:           /sys/fs/lustre/llite/<fsname>-<uuid>/max_easize
 277 Date:           May 2015
 278 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 279 Description:
 280                 Shows maximum number of bytes file striping data could be
 281                 in current configuration of storage.
 282
 283 What:           /sys/fs/lustre/llite/<fsname>-<uuid>/default_easize
 284 Date:           May 2015
 285 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 286 Description:
 287                 Shows maximum observed file striping data seen by this
 288                 filesystem client instance.
 289
 290 What:           /sys/fs/lustre/llite/<fsname>-<uuid>/xattr_cache
 291 Date:           May 2015
 292 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 293 Description:
 294                 Controls extended attributes client-side cache.
 295                 1 to enable, 0 to disable.
 296
 297 What:           /sys/fs/lustre/ldlm/cancel_unused_locks_before_replay
 298 Date:           May 2015
 299 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 300 Description:
 301                 Controls if client should replay unused locks during recovery
 302                 If a client tends to have a lot of unused locks in LRU,
 303                 recovery times might become prolonged.
 304                 1 - just locally cancel unused locks (default)
 305                 0 - replay unused locks.
 306
 307 What:           /sys/fs/lustre/ldlm/namespaces/<name>/resource_count
 308 Date:           May 2015
 309 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 310 Description:
 311                 Displays number of lock resources (objects on which individual
 312                 locks are taken) currently allocated in this namespace.
 313
 314 What:           /sys/fs/lustre/ldlm/namespaces/<name>/lock_count
 315 Date:           May 2015
 316 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 317 Description:
 318                 Displays number or locks allocated in this namespace.
 319
 320 What:           /sys/fs/lustre/ldlm/namespaces/<name>/lru_size
 321 Date:           May 2015
 322 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 323 Description:
 324                 Controls and displays LRU size limit for unused locks for this
 325                 namespace.
 326                 0 - LRU size is unlimited, controlled by server resources
 327                 positive number - number of locks to allow in lock LRU list
 328
 329 What:           /sys/fs/lustre/ldlm/namespaces/<name>/lock_unused_count
 330 Date:           May 2015
 331 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 332 Description:
 333                 Display number of locks currently sitting in the LRU list
 334                 of this namespace
 335
 336 What:           /sys/fs/lustre/ldlm/namespaces/<name>/lru_max_age
 337 Date:           May 2015
 338 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 339 Description:
 340                 Maximum number of milliseconds a lock could sit in LRU list
 341                 before client would voluntarily cancel it as unused.
 342
 343 What:           /sys/fs/lustre/ldlm/namespaces/<name>/early_lock_cancel
 344 Date:           May 2015
 345 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 346 Description:
 347                 Controls "early lock cancellation" feature on this namespace
 348                 if supported by the server.
 349                 When enabled, tries to preemtively cancel locks that would be
 350                 cancelled by verious operations and bundle the cancellation
 351                 requests in the same RPC as the main operation, which results
 352                 in significant speedups due to reduced lock-pingpong RPCs.
 353                 0 - disabled
 354                 1 - enabled (default)
 355
 356 What:           /sys/fs/lustre/ldlm/namespaces/<name>/pool/granted
 357 Date:           May 2015
 358 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 359 Description:
 360                 Displays number of granted locks in this namespace
 361
 362 What:           /sys/fs/lustre/ldlm/namespaces/<name>/pool/grant_rate
 363 Date:           May 2015
 364 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 365 Description:
 366                 Number of granted locks in this namespace during last
 367                 time interval
 368
 369 What:           /sys/fs/lustre/ldlm/namespaces/<name>/pool/cancel_rate
 370 Date:           May 2015
 371 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 372 Description:
 373                 Number of lock cancellations in this namespace during
 374                 last time interval
 375
 376 What:           /sys/fs/lustre/ldlm/namespaces/<name>/pool/grant_speed
 377 Date:           May 2015
 378 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 379 Description:
 380                 Calculated speed of lock granting (grant_rate - cancel_rate)
 381                 in this namespace
 382
 383 What:           /sys/fs/lustre/ldlm/namespaces/<name>/pool/grant_plan
 384 Date:           May 2015
 385 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 386 Description:
 387                 Estimated number of locks to be granted in the next time
 388                 interval in this namespace
 389
 390 What:           /sys/fs/lustre/ldlm/namespaces/<name>/pool/limit
 391 Date:           May 2015
 392 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 393 Description:
 394                 Controls number of allowed locks in this pool.
 395                 When lru_size is 0, this is the actual limit then.
 396
 397 What:           /sys/fs/lustre/ldlm/namespaces/<name>/pool/lock_volume_factor
 398 Date:           May 2015
 399 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 400 Description:
 401                 Multiplier for all lock volume calculations above.
 402                 Default is 1. Increase to make the client to more agressively
 403                 clean it's lock LRU list for this namespace.
 404
 405 What:           /sys/fs/lustre/ldlm/namespaces/<name>/pool/server_lock_volume
 406 Date:           May 2015
 407 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 408 Description:
 409                 Calculated server lock volume.
 410
 411 What:           /sys/fs/lustre/ldlm/namespaces/<name>/pool/recalc_period
 412 Date:           May 2015
 413 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 414 Description:
 415                 Controls length of time between recalculation of above
 416                 values (in seconds).
 417
 418 What:           /sys/fs/lustre/ldlm/services/ldlm_cbd/threads_min
 419 Date:           May 2015
 420 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 421 Description:
 422                 Controls minimum number of ldlm callback threads to start.
 423
 424 What:           /sys/fs/lustre/ldlm/services/ldlm_cbd/threads_max
 425 Date:           May 2015
 426 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 427 Description:
 428                 Controls maximum number of ldlm callback threads to start.
 429
 430 What:           /sys/fs/lustre/ldlm/services/ldlm_cbd/threads_started
 431 Date:           May 2015
 432 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 433 Description:
 434                 Shows actual number of ldlm callback threads running.
 435
 436 What:           /sys/fs/lustre/ldlm/services/ldlm_cbd/high_priority_ratio
 437 Date:           May 2015
 438 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 439 Description:
 440                 Controls what percentage of ldlm callback threads is dedicated
 441                 to "high priority" incoming requests.
 442
 443 What:           /sys/fs/lustre/{obdtype}/{connection_name}/blocksize
 444 Date:           May 2015
 445 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 446 Description:
 447                 Blocksize on backend filesystem for service behind this obd
 448                 device (or biggest blocksize for compound devices like lov
 449                 and lmv)
 450
 451 What:           /sys/fs/lustre/{obdtype}/{connection_name}/kbytestotal
 452 Date:           May 2015
 453 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 454 Description:
 455                 Total number of kilobytes of space on backend filesystem
 456                 for service behind this obd (or total amount for compound
 457                 devices like lov lmv)
 458
 459 What:           /sys/fs/lustre/{obdtype}/{connection_name}/kbytesfree
 460 Date:           May 2015
 461 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 462 Description:
 463                 Number of free kilobytes on backend filesystem for service
 464                 behind this obd (or total amount for compound devices
 465                 like lov lmv)
 466
 467 What:           /sys/fs/lustre/{obdtype}/{connection_name}/kbytesavail
 468 Date:           May 2015
 469 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 470 Description:
 471                 Number of kilobytes of free space on backend filesystem
 472                 for service behind this obd (or total amount for compound
 473                 devices like lov lmv) that is actually available for use
 474                 (taking into account per-client and filesystem reservations).
 475
 476 What:           /sys/fs/lustre/{obdtype}/{connection_name}/filestotal
 477 Date:           May 2015
 478 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 479 Description:
 480                 Number of inodes on backend filesystem for service behind this
 481                 obd.
 482
 483 What:           /sys/fs/lustre/{obdtype}/{connection_name}/filesfree
 484 Date:           May 2015
 485 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 486 Description:
 487                 Number of free inodes on backend filesystem for service
 488                 behind this obd.
 489
 490 What:           /sys/fs/lustre/mdc/{connection_name}/max_pages_per_rpc
 491 Date:           May 2015
 492 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 493 Description:
 494                 Maximum number of readdir pages to fit into a single readdir
 495                 RPC.
 496
 497 What:           /sys/fs/lustre/{mdc,osc}/{connection_name}/max_rpcs_in_flight
 498 Date:           May 2015
 499 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 500 Description:
 501                 Maximum number of parallel RPCs on the wire to allow on
 502                 this connection. Increasing this number would help on higher
 503                 latency links, but has a chance of overloading a server
 504                 if you have too many clients like this.
 505                 Default: 8
 506
 507 What:           /sys/fs/lustre/osc/{connection_name}/max_pages_per_rpc
 508 Date:           May 2015
 509 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 510 Description:
 511                 Maximum number of pages to fit into a single RPC.
 512                 Typically bigger RPCs allow for better performance.
 513                 Default: however many pages to form 1M of data (256 pages
 514                 for 4K page sized platforms)
 515
 516 What:           /sys/fs/lustre/osc/{connection_name}/active
 517 Date:           May 2015
 518 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 519 Description:
 520                 Controls accessibility of this connection. If set to 0,
 521                 fail all accesses immediately.
 522
 523 What:           /sys/fs/lustre/osc/{connection_name}/checksums
 524 Date:           May 2015
 525 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 526 Description:
 527                 Controls whenever to checksum bulk RPC data over the wire
 528                 to this target.
 529                 1: enable (default) ; 0: disable
 530
 531 What:           /sys/fs/lustre/osc/{connection_name}/contention_seconds
 532 Date:           May 2015
 533 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 534 Description:
 535                 Controls for how long to consider a file contended once
 536                 indicated as such by the server.
 537                 When a file is considered contended, all operations switch to
 538                 synchronous lockless mode to avoid cache and lock pingpong.
 539
 540 What:           /sys/fs/lustre/osc/{connection_name}/cur_dirty_bytes
 541 Date:           May 2015
 542 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 543 Description:
 544                 Displays how many dirty bytes is presently in the cache for this
 545                 target.
 546
 547 What:           /sys/fs/lustre/osc/{connection_name}/cur_grant_bytes
 548 Date:           May 2015
 549 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 550 Description:
 551                 Shows how many bytes we have as a "dirty cache" grant from the
 552                 server. Writing a value smaller than shown allows to release
 553                 some grant back to the server.
 554                 Dirty cache grant is a way Lustre ensures that cached successful
 555                 writes on client do not end up discarded by the server due to
 556                 lack of space later on.
 557
 558 What:           /sys/fs/lustre/osc/{connection_name}/cur_lost_grant_bytes
 559 Date:           May 2015
 560 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 561 Description:
 562                 Shows how many granted bytes were released to the server due
 563                 to lack of write activity on this client.
 564
 565 What:           /sys/fs/lustre/osc/{connection_name}/grant_shrink_interval
 566 Date:           May 2015
 567 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 568 Description:
 569                 Number of seconds with no write activity for this target
 570                 to start releasing dirty grant back to the server.
 571
 572 What:           /sys/fs/lustre/osc/{connection_name}/destroys_in_flight
 573 Date:           May 2015
 574 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 575 Description:
 576                 Number of DESTROY RPCs currently in flight to this target.
 577
 578 What:           /sys/fs/lustre/osc/{connection_name}/lockless_truncate
 579 Date:           May 2015
 580 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 581 Description:
 582                 Controls whether lockless truncate RPCs are allowed to this
 583                 target.
 584                 Lockless truncate causes server to perform the locking which
 585                 is beneficial if the truncate is not followed by a write
 586                 immediately.
 587                 1: enable ; 0: disable (default)
 588
 589 What:           /sys/fs/lustre/osc/{connection_name}/max_dirty_mb
 590 Date:           May 2015
 591 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 592 Description:
 593                 Controls how much dirty data this client can accumulate
 594                 for this target. This is orthogonal to dirty grant and is
 595                 a hard limit even if the server would allow a bigger dirty
 596                 cache.
 597                 While allowing higher dirty cache is beneficial for write
 598                 performance, flushing write cache takes longer and as such
 599                 the node might be more prone to OOMs.
 600                 Having this value set too low might result in not being able
 601                 to sent too many parallel WRITE RPCs.
 602                 Default: 32
 603
 604 What:           /sys/fs/lustre/osc/{connection_name}/resend_count
 605 Date:           May 2015
 606 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 607 Description:
 608                 Controls how many times to try and resend RPCs to this target
 609                 that failed with "recoverable" status, such as EAGAIN,
 610                 ENOMEM.
 611
 612 What:           /sys/fs/lustre/lov/{connection_name}/numobd
 613 Date:           May 2015
 614 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 615 Description:
 616                 Number of OSC targets managed by this LOV instance.
 617
 618 What:           /sys/fs/lustre/lov/{connection_name}/activeobd
 619 Date:           May 2015
 620 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 621 Description:
 622                 Number of OSC targets managed by this LOV instance that are
 623                 actually active.
 624
 625 What:           /sys/fs/lustre/lmv/{connection_name}/numobd
 626 Date:           May 2015
 627 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 628 Description:
 629                 Number of MDC targets managed by this LMV instance.
 630
 631 What:           /sys/fs/lustre/lmv/{connection_name}/activeobd
 632 Date:           May 2015
 633 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 634 Description:
 635                 Number of MDC targets managed by this LMV instance that are
 636                 actually active.
 637
 638 What:           /sys/fs/lustre/lmv/{connection_name}/placement
 639 Date:           May 2015
 640 Contact:        "Oleg Drokin" <oleg.drokin@intel.com>
 641 Description:
 642                 Determines policy of inode placement in case of multiple
 643                 metadata servers:
 644                 CHAR - based on a hash of the file name used at creation time
 645                        (Default)
 646                 NID - based on a hash of creating client network id.