1.\" 2.\" Copyright (c) 2013 by Turbo Fredriksson <turbo@bayour.com>. All rights reserved. 3.\" Copyright (c) 2019, 2021 by Delphix. All rights reserved. 4.\" Copyright (c) 2019 Datto Inc. 5.\" The contents of this file are subject to the terms of the Common Development 6.\" and Distribution License (the "License"). You may not use this file except 7.\" in compliance with the License. You can obtain a copy of the license at 8.\" usr/src/OPENSOLARIS.LICENSE or http://www.opensolaris.org/os/licensing. 9.\" 10.\" See the License for the specific language governing permissions and 11.\" limitations under the License. When distributing Covered Code, include this 12.\" CDDL HEADER in each file and include the License file at 13.\" usr/src/OPENSOLARIS.LICENSE. If applicable, add the following below this 14.\" CDDL HEADER, with the fields enclosed by brackets "[]" replaced with your 15.\" own identifying information: 16.\" Portions Copyright [yyyy] [name of copyright owner] 17.\" 18.Dd June 1, 2021 19.Dt ZFS 4 20.Os 21. 22.Sh NAME 23.Nm zfs 24.Nd tuning of the ZFS kernel module 25. 26.Sh DESCRIPTION 27The ZFS module supports these parameters: 28.Bl -tag -width Ds 29.It Sy dbuf_cache_max_bytes Ns = Ns Sy ULONG_MAX Ns B Pq ulong 30Maximum size in bytes of the dbuf cache. 31The target size is determined by the MIN versus 32.No 1/2^ Ns Sy dbuf_cache_shift Pq 1/32nd 33of the target ARC size. 34The behavior of the dbuf cache and its associated settings 35can be observed via the 36.Pa /proc/spl/kstat/zfs/dbufstats 37kstat. 38. 39.It Sy dbuf_metadata_cache_max_bytes Ns = Ns Sy ULONG_MAX Ns B Pq ulong 40Maximum size in bytes of the metadata dbuf cache. 41The target size is determined by the MIN versus 42.No 1/2^ Ns Sy dbuf_metadata_cache_shift Pq 1/64th 43of the target ARC size. 44The behavior of the metadata dbuf cache and its associated settings 45can be observed via the 46.Pa /proc/spl/kstat/zfs/dbufstats 47kstat. 48. 49.It Sy dbuf_cache_hiwater_pct Ns = Ns Sy 10 Ns % Pq uint 50The percentage over 51.Sy dbuf_cache_max_bytes 52when dbufs must be evicted directly. 53. 54.It Sy dbuf_cache_lowater_pct Ns = Ns Sy 10 Ns % Pq uint 55The percentage below 56.Sy dbuf_cache_max_bytes 57when the evict thread stops evicting dbufs. 58. 59.It Sy dbuf_cache_shift Ns = Ns Sy 5 Pq int 60Set the size of the dbuf cache 61.Pq Sy dbuf_cache_max_bytes 62to a log2 fraction of the target ARC size. 63. 64.It Sy dbuf_metadata_cache_shift Ns = Ns Sy 6 Pq int 65Set the size of the dbuf metadata cache 66.Pq Sy dbuf_metadata_cache_max_bytes 67to a log2 fraction of the target ARC size. 68. 69.It Sy dmu_object_alloc_chunk_shift Ns = Ns Sy 7 Po 128 Pc Pq int 70dnode slots allocated in a single operation as a power of 2. 71The default value minimizes lock contention for the bulk operation performed. 72. 73.It Sy dmu_prefetch_max Ns = Ns Sy 134217728 Ns B Po 128MB Pc Pq int 74Limit the amount we can prefetch with one call to this amount in bytes. 75This helps to limit the amount of memory that can be used by prefetching. 76. 77.It Sy ignore_hole_birth Pq int 78Alias for 79.Sy send_holes_without_birth_time . 80. 81.It Sy l2arc_feed_again Ns = Ns Sy 1 Ns | Ns 0 Pq int 82Turbo L2ARC warm-up. 83When the L2ARC is cold the fill interval will be set as fast as possible. 84. 85.It Sy l2arc_feed_min_ms Ns = Ns Sy 200 Pq ulong 86Min feed interval in milliseconds. 87Requires 88.Sy l2arc_feed_again Ns = Ns Ar 1 89and only applicable in related situations. 90. 91.It Sy l2arc_feed_secs Ns = Ns Sy 1 Pq ulong 92Seconds between L2ARC writing. 93. 94.It Sy l2arc_headroom Ns = Ns Sy 2 Pq ulong 95How far through the ARC lists to search for L2ARC cacheable content, 96expressed as a multiplier of 97.Sy l2arc_write_max . 98ARC persistence across reboots can be achieved with persistent L2ARC 99by setting this parameter to 100.Sy 0 , 101allowing the full length of ARC lists to be searched for cacheable content. 102. 103.It Sy l2arc_headroom_boost Ns = Ns Sy 200 Ns % Pq ulong 104Scales 105.Sy l2arc_headroom 106by this percentage when L2ARC contents are being successfully compressed 107before writing. 108A value of 109.Sy 100 110disables this feature. 111. 112.It Sy l2arc_mfuonly Ns = Ns Sy 0 Ns | Ns 1 Pq int 113Controls whether only MFU metadata and data are cached from ARC into L2ARC. 114This may be desired to avoid wasting space on L2ARC when reading/writing large 115amounts of data that are not expected to be accessed more than once. 116.Pp 117The default is off, 118meaning both MRU and MFU data and metadata are cached. 119When turning off this feature, some MRU buffers will still be present 120in ARC and eventually cached on L2ARC. 121.No If Sy l2arc_noprefetch Ns = Ns Sy 0 , 122some prefetched buffers will be cached to L2ARC, and those might later 123transition to MRU, in which case the 124.Sy l2arc_mru_asize No arcstat will not be Sy 0 . 125.Pp 126Regardless of 127.Sy l2arc_noprefetch , 128some MFU buffers might be evicted from ARC, 129accessed later on as prefetches and transition to MRU as prefetches. 130If accessed again they are counted as MRU and the 131.Sy l2arc_mru_asize No arcstat will not be Sy 0 . 132.Pp 133The ARC status of L2ARC buffers when they were first cached in 134L2ARC can be seen in the 135.Sy l2arc_mru_asize , Sy l2arc_mfu_asize , No and Sy l2arc_prefetch_asize 136arcstats when importing the pool or onlining a cache 137device if persistent L2ARC is enabled. 138.Pp 139The 140.Sy evict_l2_eligible_mru 141arcstat does not take into account if this option is enabled as the information 142provided by the 143.Sy evict_l2_eligible_m[rf]u 144arcstats can be used to decide if toggling this option is appropriate 145for the current workload. 146. 147.It Sy l2arc_meta_percent Ns = Ns Sy 33 Ns % Pq int 148Percent of ARC size allowed for L2ARC-only headers. 149Since L2ARC buffers are not evicted on memory pressure, 150too many headers on a system with an irrationally large L2ARC 151can render it slow or unusable. 152This parameter limits L2ARC writes and rebuilds to achieve the target. 153. 154.It Sy l2arc_trim_ahead Ns = Ns Sy 0 Ns % Pq ulong 155Trims ahead of the current write size 156.Pq Sy l2arc_write_max 157on L2ARC devices by this percentage of write size if we have filled the device. 158If set to 159.Sy 100 160we TRIM twice the space required to accommodate upcoming writes. 161A minimum of 162.Sy 64MB 163will be trimmed. 164It also enables TRIM of the whole L2ARC device upon creation 165or addition to an existing pool or if the header of the device is 166invalid upon importing a pool or onlining a cache device. 167A value of 168.Sy 0 169disables TRIM on L2ARC altogether and is the default as it can put significant 170stress on the underlying storage devices. 171This will vary depending of how well the specific device handles these commands. 172. 173.It Sy l2arc_noprefetch Ns = Ns Sy 1 Ns | Ns 0 Pq int 174Do not write buffers to L2ARC if they were prefetched but not used by 175applications. 176In case there are prefetched buffers in L2ARC and this option 177is later set, we do not read the prefetched buffers from L2ARC. 178Unsetting this option is useful for caching sequential reads from the 179disks to L2ARC and serve those reads from L2ARC later on. 180This may be beneficial in case the L2ARC device is significantly faster 181in sequential reads than the disks of the pool. 182.Pp 183Use 184.Sy 1 185to disable and 186.Sy 0 187to enable caching/reading prefetches to/from L2ARC. 188. 189.It Sy l2arc_norw Ns = Ns Sy 0 Ns | Ns 1 Pq int 190No reads during writes. 191. 192.It Sy l2arc_write_boost Ns = Ns Sy 8388608 Ns B Po 8MB Pc Pq ulong 193Cold L2ARC devices will have 194.Sy l2arc_write_max 195increased by this amount while they remain cold. 196. 197.It Sy l2arc_write_max Ns = Ns Sy 8388608 Ns B Po 8MB Pc Pq ulong 198Max write bytes per interval. 199. 200.It Sy l2arc_rebuild_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int 201Rebuild the L2ARC when importing a pool (persistent L2ARC). 202This can be disabled if there are problems importing a pool 203or attaching an L2ARC device (e.g. the L2ARC device is slow 204in reading stored log metadata, or the metadata 205has become somehow fragmented/unusable). 206. 207.It Sy l2arc_rebuild_blocks_min_l2size Ns = Ns Sy 1073741824 Ns B Po 1GB Pc Pq ulong 208Mininum size of an L2ARC device required in order to write log blocks in it. 209The log blocks are used upon importing the pool to rebuild the persistent L2ARC. 210.Pp 211For L2ARC devices less than 1GB, the amount of data 212.Fn l2arc_evict 213evicts is significant compared to the amount of restored L2ARC data. 214In this case, do not write log blocks in L2ARC in order not to waste space. 215. 216.It Sy metaslab_aliquot Ns = Ns Sy 524288 Ns B Po 512kB Pc Pq ulong 217Metaslab granularity, in bytes. 218This is roughly similar to what would be referred to as the "stripe size" 219in traditional RAID arrays. 220In normal operation, ZFS will try to write this amount of data 221to a top-level vdev before moving on to the next one. 222. 223.It Sy metaslab_bias_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int 224Enable metaslab group biasing based on their vdevs' over- or under-utilization 225relative to the pool. 226. 227.It Sy metaslab_force_ganging Ns = Ns Sy 16777217 Ns B Ns B Po 16MB + 1B Pc Pq ulong 228Make some blocks above a certain size be gang blocks. 229This option is used by the test suite to facilitate testing. 230. 231.It Sy zfs_history_output_max Ns = Ns Sy 1048576 Ns B Ns B Po 1MB Pc Pq int 232When attempting to log an output nvlist of an ioctl in the on-disk history, 233the output will not be stored if it is larger than this size (in bytes). 234This must be less than 235.Sy DMU_MAX_ACCESS Pq 64MB . 236This applies primarily to 237.Fn zfs_ioc_channel_program Pq cf. Xr zfs-program 8 . 238. 239.It Sy zfs_keep_log_spacemaps_at_export Ns = Ns Sy 0 Ns | Ns 1 Pq int 240Prevent log spacemaps from being destroyed during pool exports and destroys. 241. 242.It Sy zfs_metaslab_segment_weight_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int 243Enable/disable segment-based metaslab selection. 244. 245.It Sy zfs_metaslab_switch_threshold Ns = Ns Sy 2 Pq int 246When using segment-based metaslab selection, continue allocating 247from the active metaslab until this option's 248worth of buckets have been exhausted. 249. 250.It Sy metaslab_debug_load Ns = Ns Sy 0 Ns | Ns 1 Pq int 251Load all metaslabs during pool import. 252. 253.It Sy metaslab_debug_unload Ns = Ns Sy 0 Ns | Ns 1 Pq int 254Prevent metaslabs from being unloaded. 255. 256.It Sy metaslab_fragmentation_factor_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int 257Enable use of the fragmentation metric in computing metaslab weights. 258. 259.It Sy metaslab_df_max_search Ns = Ns Sy 16777216 Ns B Po 16MB Pc Pq int 260Maximum distance to search forward from the last offset. 261Without this limit, fragmented pools can see 262.Em >100`000 263iterations and 264.Fn metaslab_block_picker 265becomes the performance limiting factor on high-performance storage. 266.Pp 267With the default setting of 268.Sy 16MB , 269we typically see less than 270.Em 500 271iterations, even with very fragmented 272.Sy ashift Ns = Ns Sy 9 273pools. 274The maximum number of iterations possible is 275.Sy metaslab_df_max_search / 2^(ashift+1) . 276With the default setting of 277.Sy 16MB 278this is 279.Em 16*1024 Pq with Sy ashift Ns = Ns Sy 9 280or 281.Em 2*1024 Pq with Sy ashift Ns = Ns Sy 12 . 282. 283.It Sy metaslab_df_use_largest_segment Ns = Ns Sy 0 Ns | Ns 1 Pq int 284If not searching forward (due to 285.Sy metaslab_df_max_search , metaslab_df_free_pct , 286.No or Sy metaslab_df_alloc_threshold ) , 287this tunable controls which segment is used. 288If set, we will use the largest free segment. 289If unset, we will use a segment of at least the requested size. 290. 291.It Sy zfs_metaslab_max_size_cache_sec Ns = Ns Sy 3600 Ns s Po 1h Pc Pq ulong 292When we unload a metaslab, we cache the size of the largest free chunk. 293We use that cached size to determine whether or not to load a metaslab 294for a given allocation. 295As more frees accumulate in that metaslab while it's unloaded, 296the cached max size becomes less and less accurate. 297After a number of seconds controlled by this tunable, 298we stop considering the cached max size and start 299considering only the histogram instead. 300. 301.It Sy zfs_metaslab_mem_limit Ns = Ns Sy 25 Ns % Pq int 302When we are loading a new metaslab, we check the amount of memory being used 303to store metaslab range trees. 304If it is over a threshold, we attempt to unload the least recently used metaslab 305to prevent the system from clogging all of its memory with range trees. 306This tunable sets the percentage of total system memory that is the threshold. 307. 308.It Sy zfs_metaslab_try_hard_before_gang Ns = Ns Sy 0 Ns | Ns 1 Pq int 309.Bl -item -compact 310.It 311If unset, we will first try normal allocation. 312.It 313If that fails then we will do a gang allocation. 314.It 315If that fails then we will do a "try hard" gang allocation. 316.It 317If that fails then we will have a multi-layer gang block. 318.El 319.Pp 320.Bl -item -compact 321.It 322If set, we will first try normal allocation. 323.It 324If that fails then we will do a "try hard" allocation. 325.It 326If that fails we will do a gang allocation. 327.It 328If that fails we will do a "try hard" gang allocation. 329.It 330If that fails then we will have a multi-layer gang block. 331.El 332. 333.It Sy zfs_metaslab_find_max_tries Ns = Ns Sy 100 Pq int 334When not trying hard, we only consider this number of the best metaslabs. 335This improves performance, especially when there are many metaslabs per vdev 336and the allocation can't actually be satisfied 337(so we would otherwise iterate all metaslabs). 338. 339.It Sy zfs_vdev_default_ms_count Ns = Ns Sy 200 Pq int 340When a vdev is added, target this number of metaslabs per top-level vdev. 341. 342.It Sy zfs_vdev_default_ms_shift Ns = Ns Sy 29 Po 512MB Pc Pq int 343Default limit for metaslab size. 344. 345.It Sy zfs_vdev_max_auto_ashift Ns = Ns Sy ASHIFT_MAX Po 16 Pc Pq ulong 346Maximum ashift used when optimizing for logical -> physical sector size on new 347top-level vdevs. 348. 349.It Sy zfs_vdev_min_auto_ashift Ns = Ns Sy ASHIFT_MIN Po 9 Pc Pq ulong 350Minimum ashift used when creating new top-level vdevs. 351. 352.It Sy zfs_vdev_min_ms_count Ns = Ns Sy 16 Pq int 353Minimum number of metaslabs to create in a top-level vdev. 354. 355.It Sy vdev_validate_skip Ns = Ns Sy 0 Ns | Ns 1 Pq int 356Skip label validation steps during pool import. 357Changing is not recommended unless you know what you're doing 358and are recovering a damaged label. 359. 360.It Sy zfs_vdev_ms_count_limit Ns = Ns Sy 131072 Po 128k Pc Pq int 361Practical upper limit of total metaslabs per top-level vdev. 362. 363.It Sy metaslab_preload_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int 364Enable metaslab group preloading. 365. 366.It Sy metaslab_lba_weighting_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int 367Give more weight to metaslabs with lower LBAs, 368assuming they have greater bandwidth, 369as is typically the case on a modern constant angular velocity disk drive. 370. 371.It Sy metaslab_unload_delay Ns = Ns Sy 32 Pq int 372After a metaslab is used, we keep it loaded for this many TXGs, to attempt to 373reduce unnecessary reloading. 374Note that both this many TXGs and 375.Sy metaslab_unload_delay_ms 376milliseconds must pass before unloading will occur. 377. 378.It Sy metaslab_unload_delay_ms Ns = Ns Sy 600000 Ns ms Po 10min Pc Pq int 379After a metaslab is used, we keep it loaded for this many milliseconds, 380to attempt to reduce unnecessary reloading. 381Note, that both this many milliseconds and 382.Sy metaslab_unload_delay 383TXGs must pass before unloading will occur. 384. 385.It Sy reference_history Ns = Ns Sy 3 Pq int 386Maximum reference holders being tracked when reference_tracking_enable is active. 387. 388.It Sy reference_tracking_enable Ns = Ns Sy 0 Ns | Ns 1 Pq int 389Track reference holders to 390.Sy refcount_t 391objects (debug builds only). 392. 393.It Sy send_holes_without_birth_time Ns = Ns Sy 1 Ns | Ns 0 Pq int 394When set, the 395.Sy hole_birth 396optimization will not be used, and all holes will always be sent during a 397.Nm zfs Cm send . 398This is useful if you suspect your datasets are affected by a bug in 399.Sy hole_birth . 400. 401.It Sy spa_config_path Ns = Ns Pa /etc/zfs/zpool.cache Pq charp 402SPA config file. 403. 404.It Sy spa_asize_inflation Ns = Ns Sy 24 Pq int 405Multiplication factor used to estimate actual disk consumption from the 406size of data being written. 407The default value is a worst case estimate, 408but lower values may be valid for a given pool depending on its configuration. 409Pool administrators who understand the factors involved 410may wish to specify a more realistic inflation factor, 411particularly if they operate close to quota or capacity limits. 412. 413.It Sy spa_load_print_vdev_tree Ns = Ns Sy 0 Ns | Ns 1 Pq int 414Whether to print the vdev tree in the debugging message buffer during pool import. 415. 416.It Sy spa_load_verify_data Ns = Ns Sy 1 Ns | Ns 0 Pq int 417Whether to traverse data blocks during an "extreme rewind" 418.Pq Fl X 419import. 420.Pp 421An extreme rewind import normally performs a full traversal of all 422blocks in the pool for verification. 423If this parameter is unset, the traversal skips non-metadata blocks. 424It can be toggled once the 425import has started to stop or start the traversal of non-metadata blocks. 426. 427.It Sy spa_load_verify_metadata Ns = Ns Sy 1 Ns | Ns 0 Pq int 428Whether to traverse blocks during an "extreme rewind" 429.Pq Fl X 430pool import. 431.Pp 432An extreme rewind import normally performs a full traversal of all 433blocks in the pool for verification. 434If this parameter is unset, the traversal is not performed. 435It can be toggled once the import has started to stop or start the traversal. 436. 437.It Sy spa_load_verify_shift Ns = Ns Sy 4 Po 1/16th Pc Pq int 438Sets the maximum number of bytes to consume during pool import to the log2 439fraction of the target ARC size. 440. 441.It Sy spa_slop_shift Ns = Ns Sy 5 Po 1/32nd Pc Pq int 442Normally, we don't allow the last 443.Sy 3.2% Pq Sy 1/2^spa_slop_shift 444of space in the pool to be consumed. 445This ensures that we don't run the pool completely out of space, 446due to unaccounted changes (e.g. to the MOS). 447It also limits the worst-case time to allocate space. 448If we have less than this amount of free space, 449most ZPL operations (e.g. write, create) will return 450.Sy ENOSPC . 451. 452.It Sy vdev_removal_max_span Ns = Ns Sy 32768 Ns B Po 32kB Pc Pq int 453During top-level vdev removal, chunks of data are copied from the vdev 454which may include free space in order to trade bandwidth for IOPS. 455This parameter determines the maximum span of free space, in bytes, 456which will be included as "unnecessary" data in a chunk of copied data. 457.Pp 458The default value here was chosen to align with 459.Sy zfs_vdev_read_gap_limit , 460which is a similar concept when doing 461regular reads (but there's no reason it has to be the same). 462. 463.It Sy vdev_file_logical_ashift Ns = Ns Sy 9 Po 512B Pc Pq ulong 464Logical ashift for file-based devices. 465. 466.It Sy vdev_file_physical_ashift Ns = Ns Sy 9 Po 512B Pc Pq ulong 467Physical ashift for file-based devices. 468. 469.It Sy zap_iterate_prefetch Ns = Ns Sy 1 Ns | Ns 0 Pq int 470If set, when we start iterating over a ZAP object, 471prefetch the entire object (all leaf blocks). 472However, this is limited by 473.Sy dmu_prefetch_max . 474. 475.It Sy zfetch_array_rd_sz Ns = Ns Sy 1048576 Ns B Po 1MB Pc Pq ulong 476If prefetching is enabled, disable prefetching for reads larger than this size. 477. 478.It Sy zfetch_max_distance Ns = Ns Sy 8388608 Ns B Po 8MB Pc Pq uint 479Max bytes to prefetch per stream. 480. 481.It Sy zfetch_max_idistance Ns = Ns Sy 67108864 Ns B Po 64MB Pc Pq uint 482Max bytes to prefetch indirects for per stream. 483. 484.It Sy zfetch_max_streams Ns = Ns Sy 8 Pq uint 485Max number of streams per zfetch (prefetch streams per file). 486. 487.It Sy zfetch_min_sec_reap Ns = Ns Sy 2 Pq uint 488Min time before an active prefetch stream can be reclaimed 489. 490.It Sy zfs_abd_scatter_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int 491Enables ARC from using scatter/gather lists and forces all allocations to be 492linear in kernel memory. 493Disabling can improve performance in some code paths 494at the expense of fragmented kernel memory. 495. 496.It Sy zfs_abd_scatter_max_order Ns = Ns Sy MAX_ORDER-1 Pq uint 497Maximum number of consecutive memory pages allocated in a single block for 498scatter/gather lists. 499.Pp 500The value of 501.Sy MAX_ORDER 502depends on kernel configuration. 503. 504.It Sy zfs_abd_scatter_min_size Ns = Ns Sy 1536 Ns B Po 1.5kB Pc Pq uint 505This is the minimum allocation size that will use scatter (page-based) ABDs. 506Smaller allocations will use linear ABDs. 507. 508.It Sy zfs_arc_dnode_limit Ns = Ns Sy 0 Ns B Pq ulong 509When the number of bytes consumed by dnodes in the ARC exceeds this number of 510bytes, try to unpin some of it in response to demand for non-metadata. 511This value acts as a ceiling to the amount of dnode metadata, and defaults to 512.Sy 0 , 513which indicates that a percent which is based on 514.Sy zfs_arc_dnode_limit_percent 515of the ARC meta buffers that may be used for dnodes. 516.Pp 517Also see 518.Sy zfs_arc_meta_prune 519which serves a similar purpose but is used 520when the amount of metadata in the ARC exceeds 521.Sy zfs_arc_meta_limit 522rather than in response to overall demand for non-metadata. 523. 524.It Sy zfs_arc_dnode_limit_percent Ns = Ns Sy 10 Ns % Pq ulong 525Percentage that can be consumed by dnodes of ARC meta buffers. 526.Pp 527See also 528.Sy zfs_arc_dnode_limit , 529which serves a similar purpose but has a higher priority if nonzero. 530. 531.It Sy zfs_arc_dnode_reduce_percent Ns = Ns Sy 10 Ns % Pq ulong 532Percentage of ARC dnodes to try to scan in response to demand for non-metadata 533when the number of bytes consumed by dnodes exceeds 534.Sy zfs_arc_dnode_limit . 535. 536.It Sy zfs_arc_average_blocksize Ns = Ns Sy 8192 Ns B Po 8kB Pc Pq int 537The ARC's buffer hash table is sized based on the assumption of an average 538block size of this value. 539This works out to roughly 1MB of hash table per 1GB of physical memory 540with 8-byte pointers. 541For configurations with a known larger average block size, 542this value can be increased to reduce the memory footprint. 543. 544.It Sy zfs_arc_eviction_pct Ns = Ns Sy 200 Ns % Pq int 545When 546.Fn arc_is_overflowing , 547.Fn arc_get_data_impl 548waits for this percent of the requested amount of data to be evicted. 549For example, by default, for every 550.Em 2kB 551that's evicted, 552.Em 1kB 553of it may be "reused" by a new allocation. 554Since this is above 555.Sy 100 Ns % , 556it ensures that progress is made towards getting 557.Sy arc_size No under Sy arc_c . 558Since this is finite, it ensures that allocations can still happen, 559even during the potentially long time that 560.Sy arc_size No is more than Sy arc_c . 561. 562.It Sy zfs_arc_evict_batch_limit Ns = Ns Sy 10 Pq int 563Number ARC headers to evict per sub-list before proceeding to another sub-list. 564This batch-style operation prevents entire sub-lists from being evicted at once 565but comes at a cost of additional unlocking and locking. 566. 567.It Sy zfs_arc_grow_retry Ns = Ns Sy 0 Ns s Pq int 568If set to a non zero value, it will replace the 569.Sy arc_grow_retry 570value with this value. 571The 572.Sy arc_grow_retry 573.No value Pq default Sy 5 Ns s 574is the number of seconds the ARC will wait before 575trying to resume growth after a memory pressure event. 576. 577.It Sy zfs_arc_lotsfree_percent Ns = Ns Sy 10 Ns % Pq int 578Throttle I/O when free system memory drops below this percentage of total 579system memory. 580Setting this value to 581.Sy 0 582will disable the throttle. 583. 584.It Sy zfs_arc_max Ns = Ns Sy 0 Ns B Pq ulong 585Max size of ARC in bytes. 586If 587.Sy 0 , 588then the max size of ARC is determined by the amount of system memory installed. 589Under Linux, half of system memory will be used as the limit. 590Under 591.Fx , 592the larger of 593.Sy all_system_memory - 1GB No and Sy 5/8 * all_system_memory 594will be used as the limit. 595This value must be at least 596.Sy 67108864 Ns B Pq 64MB . 597.Pp 598This value can be changed dynamically, with some caveats. 599It cannot be set back to 600.Sy 0 601while running, and reducing it below the current ARC size will not cause 602the ARC to shrink without memory pressure to induce shrinking. 603. 604.It Sy zfs_arc_meta_adjust_restarts Ns = Ns Sy 4096 Pq ulong 605The number of restart passes to make while scanning the ARC attempting 606the free buffers in order to stay below the 607.Sy fs_arc_meta_limit . 608This value should not need to be tuned but is available to facilitate 609performance analysis. 610. 611.It Sy zfs_arc_meta_limit Ns = Ns Sy 0 Ns B Pq ulong 612The maximum allowed size in bytes that metadata buffers are allowed to 613consume in the ARC. 614When this limit is reached, metadata buffers will be reclaimed, 615even if the overall 616.Sy arc_c_max 617has not been reached. 618It defaults to 619.Sy 0 , 620which indicates that a percentage based on 621.Sy zfs_arc_meta_limit_percent 622of the ARC may be used for metadata. 623.Pp 624This value my be changed dynamically, except that must be set to an explicit value 625.Pq cannot be set back to Sy 0 . 626. 627.It Sy zfs_arc_meta_limit_percent Ns = Ns Sy 75 Ns % Pq ulong 628Percentage of ARC buffers that can be used for metadata. 629.Pp 630See also 631.Sy zfs_arc_meta_limit , 632which serves a similar purpose but has a higher priority if nonzero. 633. 634.It Sy zfs_arc_meta_min Ns = Ns Sy 0 Ns B Pq ulong 635The minimum allowed size in bytes that metadata buffers may consume in 636the ARC. 637. 638.It Sy zfs_arc_meta_prune Ns = Ns Sy 10000 Pq int 639The number of dentries and inodes to be scanned looking for entries 640which can be dropped. 641This may be required when the ARC reaches the 642.Sy zfs_arc_meta_limit 643because dentries and inodes can pin buffers in the ARC. 644Increasing this value will cause to dentry and inode caches 645to be pruned more aggressively. 646Setting this value to 647.Sy 0 648will disable pruning the inode and dentry caches. 649. 650.It Sy zfs_arc_meta_strategy Ns = Ns Sy 1 Ns | Ns 0 Pq int 651Define the strategy for ARC metadata buffer eviction (meta reclaim strategy): 652.Bl -tag -compact -offset 4n -width "0 (META_ONLY)" 653.It Sy 0 Pq META_ONLY 654evict only the ARC metadata buffers 655.It Sy 1 Pq BALANCED 656additional data buffers may be evicted if required 657to evict the required number of metadata buffers. 658.El 659. 660.It Sy zfs_arc_min Ns = Ns Sy 0 Ns B Pq ulong 661Min size of ARC in bytes. 662.No If set to Sy 0 , arc_c_min 663will default to consuming the larger of 664.Sy 32MB No or Sy all_system_memory/32 . 665. 666.It Sy zfs_arc_min_prefetch_ms Ns = Ns Sy 0 Ns ms Ns Po Ns ≡ Ns 1s Pc Pq int 667Minimum time prefetched blocks are locked in the ARC. 668. 669.It Sy zfs_arc_min_prescient_prefetch_ms Ns = Ns Sy 0 Ns ms Ns Po Ns ≡ Ns 6s Pc Pq int 670Minimum time "prescient prefetched" blocks are locked in the ARC. 671These blocks are meant to be prefetched fairly aggressively ahead of 672the code that may use them. 673. 674.It Sy zfs_max_missing_tvds Ns = Ns Sy 0 Pq int 675Number of missing top-level vdevs which will be allowed during 676pool import (only in read-only mode). 677. 678.It Sy zfs_max_nvlist_src_size Ns = Sy 0 Pq ulong 679Maximum size in bytes allowed to be passed as 680.Sy zc_nvlist_src_size 681for ioctls on 682.Pa /dev/zfs . 683This prevents a user from causing the kernel to allocate 684an excessive amount of memory. 685When the limit is exceeded, the ioctl fails with 686.Sy EINVAL 687and a description of the error is sent to the 688.Pa zfs-dbgmsg 689log. 690This parameter should not need to be touched under normal circumstances. 691If 692.Sy 0 , 693equivalent to a quarter of the user-wired memory limit under 694.Fx 695and to 696.Sy 134217728 Ns B Pq 128MB 697under Linux. 698. 699.It Sy zfs_multilist_num_sublists Ns = Ns Sy 0 Pq int 700To allow more fine-grained locking, each ARC state contains a series 701of lists for both data and metadata objects. 702Locking is performed at the level of these "sub-lists". 703This parameters controls the number of sub-lists per ARC state, 704and also applies to other uses of the multilist data structure. 705.Pp 706If 707.Sy 0 , 708equivalent to the greater of the number of online CPUs and 709.Sy 4 . 710. 711.It Sy zfs_arc_overflow_shift Ns = Ns Sy 8 Pq int 712The ARC size is considered to be overflowing if it exceeds the current 713ARC target size 714.Pq Sy arc_c 715by thresholds determined by this parameter. 716Exceeding by 717.Sy ( arc_c >> zfs_arc_overflow_shift ) * 0.5 718starts ARC reclamation process. 719If that appears insufficient, exceeding by 720.Sy ( arc_c >> zfs_arc_overflow_shift ) * 1.5 721blocks new buffer allocation until the reclaim thread catches up. 722Started reclamation process continues till ARC size returns below the 723target size. 724.Pp 725The default value of 726.Sy 8 727causes the ARC to start reclamation if it exceeds the target size by 728.Em 0.2% 729of the target size, and block allocations by 730.Em 0.6% . 731. 732.It Sy zfs_arc_p_min_shift Ns = Ns Sy 0 Pq int 733If nonzero, this will update 734.Sy arc_p_min_shift Pq default Sy 4 735with the new value. 736.Sy arc_p_min_shift No is used as a shift of Sy arc_c 737when calculating the minumum 738.Sy arc_p No size. 739. 740.It Sy zfs_arc_p_dampener_disable Ns = Ns Sy 1 Ns | Ns 0 Pq int 741Disable 742.Sy arc_p 743adapt dampener, which reduces the maximum single adjustment to 744.Sy arc_p . 745. 746.It Sy zfs_arc_shrink_shift Ns = Ns Sy 0 Pq int 747If nonzero, this will update 748.Sy arc_shrink_shift Pq default Sy 7 749with the new value. 750. 751.It Sy zfs_arc_pc_percent Ns = Ns Sy 0 Ns % Po off Pc Pq uint 752Percent of pagecache to reclaim ARC to. 753.Pp 754This tunable allows the ZFS ARC to play more nicely 755with the kernel's LRU pagecache. 756It can guarantee that the ARC size won't collapse under scanning 757pressure on the pagecache, yet still allows the ARC to be reclaimed down to 758.Sy zfs_arc_min 759if necessary. 760This value is specified as percent of pagecache size (as measured by 761.Sy NR_FILE_PAGES ) , 762where that percent may exceed 763.Sy 100 . 764This 765only operates during memory pressure/reclaim. 766. 767.It Sy zfs_arc_shrinker_limit Ns = Ns Sy 10000 Pq int 768This is a limit on how many pages the ARC shrinker makes available for 769eviction in response to one page allocation attempt. 770Note that in practice, the kernel's shrinker can ask us to evict 771up to about four times this for one allocation attempt. 772.Pp 773The default limit of 774.Sy 10000 Pq in practice, Em 160MB No per allocation attempt with 4kB pages 775limits the amount of time spent attempting to reclaim ARC memory to 776less than 100ms per allocation attempt, 777even with a small average compressed block size of ~8kB. 778.Pp 779The parameter can be set to 0 (zero) to disable the limit, 780and only applies on Linux. 781. 782.It Sy zfs_arc_sys_free Ns = Ns Sy 0 Ns B Pq ulong 783The target number of bytes the ARC should leave as free memory on the system. 784If zero, equivalent to the bigger of 785.Sy 512kB No and Sy all_system_memory/64 . 786. 787.It Sy zfs_autoimport_disable Ns = Ns Sy 1 Ns | Ns 0 Pq int 788Disable pool import at module load by ignoring the cache file 789.Pq Sy spa_config_path . 790. 791.It Sy zfs_checksum_events_per_second Ns = Ns Sy 20 Ns /s Pq uint 792Rate limit checksum events to this many per second. 793Note that this should not be set below the ZED thresholds 794(currently 10 checksums over 10 seconds) 795or else the daemon may not trigger any action. 796. 797.It Sy zfs_commit_timeout_pct Ns = Ns Sy 5 Ns % Pq int 798This controls the amount of time that a ZIL block (lwb) will remain "open" 799when it isn't "full", and it has a thread waiting for it to be committed to 800stable storage. 801The timeout is scaled based on a percentage of the last lwb 802latency to avoid significantly impacting the latency of each individual 803transaction record (itx). 804. 805.It Sy zfs_condense_indirect_commit_entry_delay_ms Ns = Ns Sy 0 Ns ms Pq int 806Vdev indirection layer (used for device removal) sleeps for this many 807milliseconds during mapping generation. 808Intended for use with the test suite to throttle vdev removal speed. 809. 810.It Sy zfs_condense_indirect_obsolete_pct Ns = Ns Sy 25 Ns % Pq int 811Minimum percent of obsolete bytes in vdev mapping required to attempt to condense 812.Pq see Sy zfs_condense_indirect_vdevs_enable . 813Intended for use with the test suite 814to facilitate triggering condensing as needed. 815. 816.It Sy zfs_condense_indirect_vdevs_enable Ns = Ns Sy 1 Ns | Ns 0 Pq int 817Enable condensing indirect vdev mappings. 818When set, attempt to condense indirect vdev mappings 819if the mapping uses more than 820.Sy zfs_condense_min_mapping_bytes 821bytes of memory and if the obsolete space map object uses more than 822.Sy zfs_condense_max_obsolete_bytes 823bytes on-disk. 824The condensing process is an attempt to save memory by removing obsolete mappings. 825. 826.It Sy zfs_condense_max_obsolete_bytes Ns = Ns Sy 1073741824 Ns B Po 1GB Pc Pq ulong 827Only attempt to condense indirect vdev mappings if the on-disk size 828of the obsolete space map object is greater than this number of bytes 829.Pq see Sy zfs_condense_indirect_vdevs_enable . 830. 831.It Sy zfs_condense_min_mapping_bytes Ns = Ns Sy 131072 Ns B Po 128kB Pc Pq ulong 832Minimum size vdev mapping to attempt to condense 833.Pq see Sy zfs_condense_indirect_vdevs_enable . 834. 835.It Sy zfs_dbgmsg_enable Ns = Ns Sy 1 Ns | Ns 0 Pq int 836Internally ZFS keeps a small log to facilitate debugging. 837The log is enabled by default, and can be disabled by unsetting this option. 838The contents of the log can be accessed by reading 839.Pa /proc/spl/kstat/zfs/dbgmsg . 840Writing 841.Sy 0 842to the file clears the log. 843.Pp 844This setting does not influence debug prints due to 845.Sy zfs_flags . 846. 847.It Sy zfs_dbgmsg_maxsize Ns = Ns Sy 4194304 Ns B Po 4MB Pc Pq int 848Maximum size of the internal ZFS debug log. 849. 850.It Sy zfs_dbuf_state_index Ns = Ns Sy 0 Pq int 851Historically used for controlling what reporting was available under 852.Pa /proc/spl/kstat/zfs . 853No effect. 854. 855.It Sy zfs_deadman_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int 856When a pool sync operation takes longer than 857.Sy zfs_deadman_synctime_ms , 858or when an individual I/O operation takes longer than 859.Sy zfs_deadman_ziotime_ms , 860then the operation is considered to be "hung". 861If 862.Sy zfs_deadman_enabled 863is set, then the deadman behavior is invoked as described by 864.Sy zfs_deadman_failmode . 865By default, the deadman is enabled and set to 866.Sy wait 867which results in "hung" I/Os only being logged. 868The deadman is automatically disabled when a pool gets suspended. 869. 870.It Sy zfs_deadman_failmode Ns = Ns Sy wait Pq charp 871Controls the failure behavior when the deadman detects a "hung" I/O operation. 872Valid values are: 873.Bl -tag -compact -offset 4n -width "continue" 874.It Sy wait 875Wait for a "hung" operation to complete. 876For each "hung" operation a "deadman" event will be posted 877describing that operation. 878.It Sy continue 879Attempt to recover from a "hung" operation by re-dispatching it 880to the I/O pipeline if possible. 881.It Sy panic 882Panic the system. 883This can be used to facilitate automatic fail-over 884to a properly configured fail-over partner. 885.El 886. 887.It Sy zfs_deadman_checktime_ms Ns = Ns Sy 60000 Ns ms Po 1min Pc Pq int 888Check time in milliseconds. 889This defines the frequency at which we check for hung I/O requests 890and potentially invoke the 891.Sy zfs_deadman_failmode 892behavior. 893. 894.It Sy zfs_deadman_synctime_ms Ns = Ns Sy 600000 Ns ms Po 10min Pc Pq ulong 895Interval in milliseconds after which the deadman is triggered and also 896the interval after which a pool sync operation is considered to be "hung". 897Once this limit is exceeded the deadman will be invoked every 898.Sy zfs_deadman_checktime_ms 899milliseconds until the pool sync completes. 900. 901.It Sy zfs_deadman_ziotime_ms Ns = Ns Sy 300000 Ns ms Po 5min Pc Pq ulong 902Interval in milliseconds after which the deadman is triggered and an 903individual I/O operation is considered to be "hung". 904As long as the operation remains "hung", 905the deadman will be invoked every 906.Sy zfs_deadman_checktime_ms 907milliseconds until the operation completes. 908. 909.It Sy zfs_dedup_prefetch Ns = Ns Sy 0 Ns | Ns 1 Pq int 910Enable prefetching dedup-ed blocks which are going to be freed. 911. 912.It Sy zfs_delay_min_dirty_percent Ns = Ns Sy 60 Ns % Pq int 913Start to delay each transaction once there is this amount of dirty data, 914expressed as a percentage of 915.Sy zfs_dirty_data_max . 916This value should be at least 917.Sy zfs_vdev_async_write_active_max_dirty_percent . 918.No See Sx ZFS TRANSACTION DELAY . 919. 920.It Sy zfs_delay_scale Ns = Ns Sy 500000 Pq int 921This controls how quickly the transaction delay approaches infinity. 922Larger values cause longer delays for a given amount of dirty data. 923.Pp 924For the smoothest delay, this value should be about 1 billion divided 925by the maximum number of operations per second. 926This will smoothly handle between ten times and a tenth of this number. 927.No See Sx ZFS TRANSACTION DELAY . 928.Pp 929.Sy zfs_delay_scale * zfs_dirty_data_max Em must be smaller than Sy 2^64 . 930. 931.It Sy zfs_disable_ivset_guid_check Ns = Ns Sy 0 Ns | Ns 1 Pq int 932Disables requirement for IVset GUIDs to be present and match when doing a raw 933receive of encrypted datasets. 934Intended for users whose pools were created with 935OpenZFS pre-release versions and now have compatibility issues. 936. 937.It Sy zfs_key_max_salt_uses Ns = Ns Sy 400000000 Po 4*10^8 Pc Pq ulong 938Maximum number of uses of a single salt value before generating a new one for 939encrypted datasets. 940The default value is also the maximum. 941. 942.It Sy zfs_object_mutex_size Ns = Ns Sy 64 Pq uint 943Size of the znode hashtable used for holds. 944.Pp 945Due to the need to hold locks on objects that may not exist yet, kernel mutexes 946are not created per-object and instead a hashtable is used where collisions 947will result in objects waiting when there is not actually contention on the 948same object. 949. 950.It Sy zfs_slow_io_events_per_second Ns = Ns Sy 20 Ns /s Pq int 951Rate limit delay and deadman zevents (which report slow I/Os) to this many per 952second. 953. 954.It Sy zfs_unflushed_max_mem_amt Ns = Ns Sy 1073741824 Ns B Po 1GB Pc Pq ulong 955Upper-bound limit for unflushed metadata changes to be held by the 956log spacemap in memory, in bytes. 957. 958.It Sy zfs_unflushed_max_mem_ppm Ns = Ns Sy 1000 Ns ppm Po 0.1% Pc Pq ulong 959Part of overall system memory that ZFS allows to be used 960for unflushed metadata changes by the log spacemap, in millionths. 961. 962.It Sy zfs_unflushed_log_block_max Ns = Ns Sy 262144 Po 256k Pc Pq ulong 963Describes the maximum number of log spacemap blocks allowed for each pool. 964The default value means that the space in all the log spacemaps 965can add up to no more than 966.Sy 262144 967blocks (which means 968.Em 32GB 969of logical space before compression and ditto blocks, 970assuming that blocksize is 971.Em 128kB ) . 972.Pp 973This tunable is important because it involves a trade-off between import 974time after an unclean export and the frequency of flushing metaslabs. 975The higher this number is, the more log blocks we allow when the pool is 976active which means that we flush metaslabs less often and thus decrease 977the number of I/Os for spacemap updates per TXG. 978At the same time though, that means that in the event of an unclean export, 979there will be more log spacemap blocks for us to read, inducing overhead 980in the import time of the pool. 981The lower the number, the amount of flushing increases, destroying log 982blocks quicker as they become obsolete faster, which leaves less blocks 983to be read during import time after a crash. 984.Pp 985Each log spacemap block existing during pool import leads to approximately 986one extra logical I/O issued. 987This is the reason why this tunable is exposed in terms of blocks rather 988than space used. 989. 990.It Sy zfs_unflushed_log_block_min Ns = Ns Sy 1000 Pq ulong 991If the number of metaslabs is small and our incoming rate is high, 992we could get into a situation that we are flushing all our metaslabs every TXG. 993Thus we always allow at least this many log blocks. 994. 995.It Sy zfs_unflushed_log_block_pct Ns = Ns Sy 400 Ns % Pq ulong 996Tunable used to determine the number of blocks that can be used for 997the spacemap log, expressed as a percentage of the total number of 998metaslabs in the pool. 999. 1000.It Sy zfs_unlink_suspend_progress Ns = Ns Sy 0 Ns | Ns 1 Pq uint 1001When enabled, files will not be asynchronously removed from the list of pending 1002unlinks and the space they consume will be leaked. 1003Once this option has been disabled and the dataset is remounted, 1004the pending unlinks will be processed and the freed space returned to the pool. 1005This option is used by the test suite. 1006. 1007.It Sy zfs_delete_blocks Ns = Ns Sy 20480 Pq ulong 1008This is the used to define a large file for the purposes of deletion. 1009Files containing more than 1010.Sy zfs_delete_blocks 1011will be deleted asynchronously, while smaller files are deleted synchronously. 1012Decreasing this value will reduce the time spent in an 1013.Xr unlink 2 1014system call, at the expense of a longer delay before the freed space is available. 1015. 1016.It Sy zfs_dirty_data_max Ns = Pq int 1017Determines the dirty space limit in bytes. 1018Once this limit is exceeded, new writes are halted until space frees up. 1019This parameter takes precedence over 1020.Sy zfs_dirty_data_max_percent . 1021.No See Sx ZFS TRANSACTION DELAY . 1022.Pp 1023Defaults to 1024.Sy physical_ram/10 , 1025capped at 1026.Sy zfs_dirty_data_max_max . 1027. 1028.It Sy zfs_dirty_data_max_max Ns = Pq int 1029Maximum allowable value of 1030.Sy zfs_dirty_data_max , 1031expressed in bytes. 1032This limit is only enforced at module load time, and will be ignored if 1033.Sy zfs_dirty_data_max 1034is later changed. 1035This parameter takes precedence over 1036.Sy zfs_dirty_data_max_max_percent . 1037.No See Sx ZFS TRANSACTION DELAY . 1038.Pp 1039Defaults to 1040.Sy physical_ram/4 , 1041. 1042.It Sy zfs_dirty_data_max_max_percent Ns = Ns Sy 25 Ns % Pq int 1043Maximum allowable value of 1044.Sy zfs_dirty_data_max , 1045expressed as a percentage of physical RAM. 1046This limit is only enforced at module load time, and will be ignored if 1047.Sy zfs_dirty_data_max 1048is later changed. 1049The parameter 1050.Sy zfs_dirty_data_max_max 1051takes precedence over this one. 1052.No See Sx ZFS TRANSACTION DELAY . 1053. 1054.It Sy zfs_dirty_data_max_percent Ns = Ns Sy 10 Ns % Pq int 1055Determines the dirty space limit, expressed as a percentage of all memory. 1056Once this limit is exceeded, new writes are halted until space frees up. 1057The parameter 1058.Sy zfs_dirty_data_max 1059takes precedence over this one. 1060.No See Sx ZFS TRANSACTION DELAY . 1061.Pp 1062Subject to 1063.Sy zfs_dirty_data_max_max . 1064. 1065.It Sy zfs_dirty_data_sync_percent Ns = Ns Sy 20 Ns % Pq int 1066Start syncing out a transaction group if there's at least this much dirty data 1067.Pq as a percentage of Sy zfs_dirty_data_max . 1068This should be less than 1069.Sy zfs_vdev_async_write_active_min_dirty_percent . 1070. 1071.It Sy zfs_wrlog_data_max Ns = Pq int 1072The upper limit of write-transaction zil log data size in bytes. 1073Once it is reached, write operation is blocked, until log data is cleared out 1074after transaction group sync. Because of some overhead, it should be set 1075at least 2 times the size of 1076.Sy zfs_dirty_data_max 1077.No to prevent harming normal write throughput. 1078It also should be smaller than the size of the slog device if slog is present. 1079.Pp 1080Defaults to 1081.Sy zfs_dirty_data_max*2 1082. 1083.It Sy zfs_fallocate_reserve_percent Ns = Ns Sy 110 Ns % Pq uint 1084Since ZFS is a copy-on-write filesystem with snapshots, blocks cannot be 1085preallocated for a file in order to guarantee that later writes will not 1086run out of space. 1087Instead, 1088.Xr fallocate 2 1089space preallocation only checks that sufficient space is currently available 1090in the pool or the user's project quota allocation, 1091and then creates a sparse file of the requested size. 1092The requested space is multiplied by 1093.Sy zfs_fallocate_reserve_percent 1094to allow additional space for indirect blocks and other internal metadata. 1095Setting this to 1096.Sy 0 1097disables support for 1098.Xr fallocate 2 1099and causes it to return 1100.Sy EOPNOTSUPP . 1101. 1102.It Sy zfs_fletcher_4_impl Ns = Ns Sy fastest Pq string 1103Select a fletcher 4 implementation. 1104.Pp 1105Supported selectors are: 1106.Sy fastest , scalar , sse2 , ssse3 , avx2 , avx512f , avx512bw , 1107.No and Sy aarch64_neon . 1108All except 1109.Sy fastest No and Sy scalar 1110require instruction set extensions to be available, 1111and will only appear if ZFS detects that they are present at runtime. 1112If multiple implementations of fletcher 4 are available, the 1113.Sy fastest 1114will be chosen using a micro benchmark. 1115Selecting 1116.Sy scalar 1117results in the original CPU-based calculation being used. 1118Selecting any option other than 1119.Sy fastest No or Sy scalar 1120results in vector instructions 1121from the respective CPU instruction set being used. 1122. 1123.It Sy zfs_free_bpobj_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int 1124Enable/disable the processing of the free_bpobj object. 1125. 1126.It Sy zfs_async_block_max_blocks Ns = Ns Sy ULONG_MAX Po unlimited Pc Pq ulong 1127Maximum number of blocks freed in a single TXG. 1128. 1129.It Sy zfs_max_async_dedup_frees Ns = Ns Sy 100000 Po 10^5 Pc Pq ulong 1130Maximum number of dedup blocks freed in a single TXG. 1131. 1132.It Sy zfs_override_estimate_recordsize Ns = Ns Sy 0 Pq ulong 1133If nonzer, override record size calculation for 1134.Nm zfs Cm send 1135estimates. 1136. 1137.It Sy zfs_vdev_async_read_max_active Ns = Ns Sy 3 Pq int 1138Maximum asynchronous read I/O operations active to each device. 1139.No See Sx ZFS I/O SCHEDULER . 1140. 1141.It Sy zfs_vdev_async_read_min_active Ns = Ns Sy 1 Pq int 1142Minimum asynchronous read I/O operation active to each device. 1143.No See Sx ZFS I/O SCHEDULER . 1144. 1145.It Sy zfs_vdev_async_write_active_max_dirty_percent Ns = Ns Sy 60 Ns % Pq int 1146When the pool has more than this much dirty data, use 1147.Sy zfs_vdev_async_write_max_active 1148to limit active async writes. 1149If the dirty data is between the minimum and maximum, 1150the active I/O limit is linearly interpolated. 1151.No See Sx ZFS I/O SCHEDULER . 1152. 1153.It Sy zfs_vdev_async_write_active_min_dirty_percent Ns = Ns Sy 30 Ns % Pq int 1154When the pool has less than this much dirty data, use 1155.Sy zfs_vdev_async_write_min_active 1156to limit active async writes. 1157If the dirty data is between the minimum and maximum, 1158the active I/O limit is linearly 1159interpolated. 1160.No See Sx ZFS I/O SCHEDULER . 1161. 1162.It Sy zfs_vdev_async_write_max_active Ns = Ns Sy 30 Pq int 1163Maximum asynchronous write I/O operations active to each device. 1164.No See Sx ZFS I/O SCHEDULER . 1165. 1166.It Sy zfs_vdev_async_write_min_active Ns = Ns Sy 2 Pq int 1167Minimum asynchronous write I/O operations active to each device. 1168.No See Sx ZFS I/O SCHEDULER . 1169.Pp 1170Lower values are associated with better latency on rotational media but poorer 1171resilver performance. 1172The default value of 1173.Sy 2 1174was chosen as a compromise. 1175A value of 1176.Sy 3 1177has been shown to improve resilver performance further at a cost of 1178further increasing latency. 1179. 1180.It Sy zfs_vdev_initializing_max_active Ns = Ns Sy 1 Pq int 1181Maximum initializing I/O operations active to each device. 1182.No See Sx ZFS I/O SCHEDULER . 1183. 1184.It Sy zfs_vdev_initializing_min_active Ns = Ns Sy 1 Pq int 1185Minimum initializing I/O operations active to each device. 1186.No See Sx ZFS I/O SCHEDULER . 1187. 1188.It Sy zfs_vdev_max_active Ns = Ns Sy 1000 Pq int 1189The maximum number of I/O operations active to each device. 1190Ideally, this will be at least the sum of each queue's 1191.Sy max_active . 1192.No See Sx ZFS I/O SCHEDULER . 1193. 1194.It Sy zfs_vdev_rebuild_max_active Ns = Ns Sy 3 Pq int 1195Maximum sequential resilver I/O operations active to each device. 1196.No See Sx ZFS I/O SCHEDULER . 1197. 1198.It Sy zfs_vdev_rebuild_min_active Ns = Ns Sy 1 Pq int 1199Minimum sequential resilver I/O operations active to each device. 1200.No See Sx ZFS I/O SCHEDULER . 1201. 1202.It Sy zfs_vdev_removal_max_active Ns = Ns Sy 2 Pq int 1203Maximum removal I/O operations active to each device. 1204.No See Sx ZFS I/O SCHEDULER . 1205. 1206.It Sy zfs_vdev_removal_min_active Ns = Ns Sy 1 Pq int 1207Minimum removal I/O operations active to each device. 1208.No See Sx ZFS I/O SCHEDULER . 1209. 1210.It Sy zfs_vdev_scrub_max_active Ns = Ns Sy 2 Pq int 1211Maximum scrub I/O operations active to each device. 1212.No See Sx ZFS I/O SCHEDULER . 1213. 1214.It Sy zfs_vdev_scrub_min_active Ns = Ns Sy 1 Pq int 1215Minimum scrub I/O operations active to each device. 1216.No See Sx ZFS I/O SCHEDULER . 1217. 1218.It Sy zfs_vdev_sync_read_max_active Ns = Ns Sy 10 Pq int 1219Maximum synchronous read I/O operations active to each device. 1220.No See Sx ZFS I/O SCHEDULER . 1221. 1222.It Sy zfs_vdev_sync_read_min_active Ns = Ns Sy 10 Pq int 1223Minimum synchronous read I/O operations active to each device. 1224.No See Sx ZFS I/O SCHEDULER . 1225. 1226.It Sy zfs_vdev_sync_write_max_active Ns = Ns Sy 10 Pq int 1227Maximum synchronous write I/O operations active to each device. 1228.No See Sx ZFS I/O SCHEDULER . 1229. 1230.It Sy zfs_vdev_sync_write_min_active Ns = Ns Sy 10 Pq int 1231Minimum synchronous write I/O operations active to each device. 1232.No See Sx ZFS I/O SCHEDULER . 1233. 1234.It Sy zfs_vdev_trim_max_active Ns = Ns Sy 2 Pq int 1235Maximum trim/discard I/O operations active to each device. 1236.No See Sx ZFS I/O SCHEDULER . 1237. 1238.It Sy zfs_vdev_trim_min_active Ns = Ns Sy 1 Pq int 1239Minimum trim/discard I/O operations active to each device. 1240.No See Sx ZFS I/O SCHEDULER . 1241. 1242.It Sy zfs_vdev_nia_delay Ns = Ns Sy 5 Pq int 1243For non-interactive I/O (scrub, resilver, removal, initialize and rebuild), 1244the number of concurrently-active I/O operations is limited to 1245.Sy zfs_*_min_active , 1246unless the vdev is "idle". 1247When there are no interactive I/O operatinons active (synchronous or otherwise), 1248and 1249.Sy zfs_vdev_nia_delay 1250operations have completed since the last interactive operation, 1251then the vdev is considered to be "idle", 1252and the number of concurrently-active non-interactive operations is increased to 1253.Sy zfs_*_max_active . 1254.No See Sx ZFS I/O SCHEDULER . 1255. 1256.It Sy zfs_vdev_nia_credit Ns = Ns Sy 5 Pq int 1257Some HDDs tend to prioritize sequential I/O so strongly, that concurrent 1258random I/O latency reaches several seconds. 1259On some HDDs this happens even if sequential I/O operations 1260are submitted one at a time, and so setting 1261.Sy zfs_*_max_active Ns = Sy 1 1262does not help. 1263To prevent non-interactive I/O, like scrub, 1264from monopolizing the device, no more than 1265.Sy zfs_vdev_nia_credit operations can be sent 1266while there are outstanding incomplete interactive operations. 1267This enforced wait ensures the HDD services the interactive I/O 1268within a reasonable amount of time. 1269.No See Sx ZFS I/O SCHEDULER . 1270. 1271.It Sy zfs_vdev_queue_depth_pct Ns = Ns Sy 1000 Ns % Pq int 1272Maximum number of queued allocations per top-level vdev expressed as 1273a percentage of 1274.Sy zfs_vdev_async_write_max_active , 1275which allows the system to detect devices that are more capable 1276of handling allocations and to allocate more blocks to those devices. 1277This allows for dynamic allocation distribution when devices are imbalanced, 1278as fuller devices will tend to be slower than empty devices. 1279.Pp 1280Also see 1281.Sy zio_dva_throttle_enabled . 1282. 1283.It Sy zfs_expire_snapshot Ns = Ns Sy 300 Ns s Pq int 1284Time before expiring 1285.Pa .zfs/snapshot . 1286. 1287.It Sy zfs_admin_snapshot Ns = Ns Sy 0 Ns | Ns 1 Pq int 1288Allow the creation, removal, or renaming of entries in the 1289.Sy .zfs/snapshot 1290directory to cause the creation, destruction, or renaming of snapshots. 1291When enabled, this functionality works both locally and over NFS exports 1292which have the 1293.Em no_root_squash 1294option set. 1295. 1296.It Sy zfs_flags Ns = Ns Sy 0 Pq int 1297Set additional debugging flags. 1298The following flags may be bitwise-ored together: 1299.TS 1300box; 1301lbz r l l . 1302 Value Symbolic Name Description 1303_ 1304 1 ZFS_DEBUG_DPRINTF Enable dprintf entries in the debug log. 1305* 2 ZFS_DEBUG_DBUF_VERIFY Enable extra dbuf verifications. 1306* 4 ZFS_DEBUG_DNODE_VERIFY Enable extra dnode verifications. 1307 8 ZFS_DEBUG_SNAPNAMES Enable snapshot name verification. 1308 16 ZFS_DEBUG_MODIFY Check for illegally modified ARC buffers. 1309 64 ZFS_DEBUG_ZIO_FREE Enable verification of block frees. 1310 128 ZFS_DEBUG_HISTOGRAM_VERIFY Enable extra spacemap histogram verifications. 1311 256 ZFS_DEBUG_METASLAB_VERIFY Verify space accounting on disk matches in-memory \fBrange_trees\fP. 1312 512 ZFS_DEBUG_SET_ERROR Enable \fBSET_ERROR\fP and dprintf entries in the debug log. 1313 1024 ZFS_DEBUG_INDIRECT_REMAP Verify split blocks created by device removal. 1314 2048 ZFS_DEBUG_TRIM Verify TRIM ranges are always within the allocatable range tree. 1315 4096 ZFS_DEBUG_LOG_SPACEMAP Verify that the log summary is consistent with the spacemap log 1316 and enable \fBzfs_dbgmsgs\fP for metaslab loading and flushing. 1317.TE 1318.Sy \& * No Requires debug build. 1319. 1320.It Sy zfs_free_leak_on_eio Ns = Ns Sy 0 Ns | Ns 1 Pq int 1321If destroy encounters an 1322.Sy EIO 1323while reading metadata (e.g. indirect blocks), 1324space referenced by the missing metadata can not be freed. 1325Normally this causes the background destroy to become "stalled", 1326as it is unable to make forward progress. 1327While in this stalled state, all remaining space to free 1328from the error-encountering filesystem is "temporarily leaked". 1329Set this flag to cause it to ignore the 1330.Sy EIO , 1331permanently leak the space from indirect blocks that can not be read, 1332and continue to free everything else that it can. 1333.Pp 1334The default "stalling" behavior is useful if the storage partially 1335fails (i.e. some but not all I/O operations fail), and then later recovers. 1336In this case, we will be able to continue pool operations while it is 1337partially failed, and when it recovers, we can continue to free the 1338space, with no leaks. 1339Note, however, that this case is actually fairly rare. 1340.Pp 1341Typically pools either 1342.Bl -enum -compact -offset 4n -width "1." 1343.It 1344fail completely (but perhaps temporarily, 1345e.g. due to a top-level vdev going offline), or 1346.It 1347have localized, permanent errors (e.g. disk returns the wrong data 1348due to bit flip or firmware bug). 1349.El 1350In the former case, this setting does not matter because the 1351pool will be suspended and the sync thread will not be able to make 1352forward progress regardless. 1353In the latter, because the error is permanent, the best we can do 1354is leak the minimum amount of space, 1355which is what setting this flag will do. 1356It is therefore reasonable for this flag to normally be set, 1357but we chose the more conservative approach of not setting it, 1358so that there is no possibility of 1359leaking space in the "partial temporary" failure case. 1360. 1361.It Sy zfs_free_min_time_ms Ns = Ns Sy 1000 Ns ms Po 1s Pc Pq int 1362During a 1363.Nm zfs Cm destroy 1364operation using the 1365.Sy async_destroy 1366feature, 1367a minimum of this much time will be spent working on freeing blocks per TXG. 1368. 1369.It Sy zfs_obsolete_min_time_ms Ns = Ns Sy 500 Ns ms Pq int 1370Similar to 1371.Sy zfs_free_min_time_ms , 1372but for cleanup of old indirection records for removed vdevs. 1373. 1374.It Sy zfs_immediate_write_sz Ns = Ns Sy 32768 Ns B Po 32kB Pc Pq long 1375Largest data block to write to the ZIL. 1376Larger blocks will be treated as if the dataset being written to had the 1377.Sy logbias Ns = Ns Sy throughput 1378property set. 1379. 1380.It Sy zfs_initialize_value Ns = Ns Sy 16045690984833335022 Po 0xDEADBEEFDEADBEEE Pc Pq ulong 1381Pattern written to vdev free space by 1382.Xr zpool-initialize 8 . 1383. 1384.It Sy zfs_initialize_chunk_size Ns = Ns Sy 1048576 Ns B Po 1MB Pc Pq ulong 1385Size of writes used by 1386.Xr zpool-initialize 8 . 1387This option is used by the test suite. 1388. 1389.It Sy zfs_livelist_max_entries Ns = Ns Sy 500000 Po 5*10^5 Pc Pq ulong 1390The threshold size (in block pointers) at which we create a new sub-livelist. 1391Larger sublists are more costly from a memory perspective but the fewer 1392sublists there are, the lower the cost of insertion. 1393. 1394.It Sy zfs_livelist_min_percent_shared Ns = Ns Sy 75 Ns % Pq int 1395If the amount of shared space between a snapshot and its clone drops below 1396this threshold, the clone turns off the livelist and reverts to the old 1397deletion method. 1398This is in place because livelists no long give us a benefit 1399once a clone has been overwritten enough. 1400. 1401.It Sy zfs_livelist_condense_new_alloc Ns = Ns Sy 0 Pq int 1402Incremented each time an extra ALLOC blkptr is added to a livelist entry while 1403it is being condensed. 1404This option is used by the test suite to track race conditions. 1405. 1406.It Sy zfs_livelist_condense_sync_cancel Ns = Ns Sy 0 Pq int 1407Incremented each time livelist condensing is canceled while in 1408.Fn spa_livelist_condense_sync . 1409This option is used by the test suite to track race conditions. 1410. 1411.It Sy zfs_livelist_condense_sync_pause Ns = Ns Sy 0 Ns | Ns 1 Pq int 1412When set, the livelist condense process pauses indefinitely before 1413executing the synctask - 1414.Fn spa_livelist_condense_sync . 1415This option is used by the test suite to trigger race conditions. 1416. 1417.It Sy zfs_livelist_condense_zthr_cancel Ns = Ns Sy 0 Pq int 1418Incremented each time livelist condensing is canceled while in 1419.Fn spa_livelist_condense_cb . 1420This option is used by the test suite to track race conditions. 1421. 1422.It Sy zfs_livelist_condense_zthr_pause Ns = Ns Sy 0 Ns | Ns 1 Pq int 1423When set, the livelist condense process pauses indefinitely before 1424executing the open context condensing work in 1425.Fn spa_livelist_condense_cb . 1426This option is used by the test suite to trigger race conditions. 1427. 1428.It Sy zfs_lua_max_instrlimit Ns = Ns Sy 100000000 Po 10^8 Pc Pq ulong 1429The maximum execution time limit that can be set for a ZFS channel program, 1430specified as a number of Lua instructions. 1431. 1432.It Sy zfs_lua_max_memlimit Ns = Ns Sy 104857600 Po 100MB Pc Pq ulong 1433The maximum memory limit that can be set for a ZFS channel program, specified 1434in bytes. 1435. 1436.It Sy zfs_max_dataset_nesting Ns = Ns Sy 50 Pq int 1437The maximum depth of nested datasets. 1438This value can be tuned temporarily to 1439fix existing datasets that exceed the predefined limit. 1440. 1441.It Sy zfs_max_log_walking Ns = Ns Sy 5 Pq ulong 1442The number of past TXGs that the flushing algorithm of the log spacemap 1443feature uses to estimate incoming log blocks. 1444. 1445.It Sy zfs_max_logsm_summary_length Ns = Ns Sy 10 Pq ulong 1446Maximum number of rows allowed in the summary of the spacemap log. 1447. 1448.It Sy zfs_max_recordsize Ns = Ns Sy 1048576 Po 1MB Pc Pq int 1449We currently support block sizes from 1450.Em 512B No to Em 16MB . 1451The benefits of larger blocks, and thus larger I/O, 1452need to be weighed against the cost of COWing a giant block to modify one byte. 1453Additionally, very large blocks can have an impact on I/O latency, 1454and also potentially on the memory allocator. 1455Therefore, we do not allow the recordsize to be set larger than this tunable. 1456Larger blocks can be created by changing it, 1457and pools with larger blocks can always be imported and used, 1458regardless of this setting. 1459. 1460.It Sy zfs_allow_redacted_dataset_mount Ns = Ns Sy 0 Ns | Ns 1 Pq int 1461Allow datasets received with redacted send/receive to be mounted. 1462Normally disabled because these datasets may be missing key data. 1463. 1464.It Sy zfs_min_metaslabs_to_flush Ns = Ns Sy 1 Pq ulong 1465Minimum number of metaslabs to flush per dirty TXG. 1466. 1467.It Sy zfs_metaslab_fragmentation_threshold Ns = Ns Sy 70 Ns % Pq int 1468Allow metaslabs to keep their active state as long as their fragmentation 1469percentage is no more than this value. 1470An active metaslab that exceeds this threshold 1471will no longer keep its active status allowing better metaslabs to be selected. 1472. 1473.It Sy zfs_mg_fragmentation_threshold Ns = Ns Sy 95 Ns % Pq int 1474Metaslab groups are considered eligible for allocations if their 1475fragmentation metric (measured as a percentage) is less than or equal to 1476this value. 1477If a metaslab group exceeds this threshold then it will be 1478skipped unless all metaslab groups within the metaslab class have also 1479crossed this threshold. 1480. 1481.It Sy zfs_mg_noalloc_threshold Ns = Ns Sy 0 Ns % Pq int 1482Defines a threshold at which metaslab groups should be eligible for allocations. 1483The value is expressed as a percentage of free space 1484beyond which a metaslab group is always eligible for allocations. 1485If a metaslab group's free space is less than or equal to the 1486threshold, the allocator will avoid allocating to that group 1487unless all groups in the pool have reached the threshold. 1488Once all groups have reached the threshold, all groups are allowed to accept 1489allocations. 1490The default value of 1491.Sy 0 1492disables the feature and causes all metaslab groups to be eligible for allocations. 1493.Pp 1494This parameter allows one to deal with pools having heavily imbalanced 1495vdevs such as would be the case when a new vdev has been added. 1496Setting the threshold to a non-zero percentage will stop allocations 1497from being made to vdevs that aren't filled to the specified percentage 1498and allow lesser filled vdevs to acquire more allocations than they 1499otherwise would under the old 1500.Sy zfs_mg_alloc_failures 1501facility. 1502. 1503.It Sy zfs_ddt_data_is_special Ns = Ns Sy 1 Ns | Ns 0 Pq int 1504If enabled, ZFS will place DDT data into the special allocation class. 1505. 1506.It Sy zfs_user_indirect_is_special Ns = Ns Sy 1 Ns | Ns 0 Pq int 1507If enabled, ZFS will place user data indirect blocks 1508into the special allocation class. 1509. 1510.It Sy zfs_multihost_history Ns = Ns Sy 0 Pq int 1511Historical statistics for this many latest multihost updates will be available in 1512.Pa /proc/spl/kstat/zfs/ Ns Ao Ar pool Ac Ns Pa /multihost . 1513. 1514.It Sy zfs_multihost_interval Ns = Ns Sy 1000 Ns ms Po 1s Pc Pq ulong 1515Used to control the frequency of multihost writes which are performed when the 1516.Sy multihost 1517pool property is on. 1518This is one of the factors used to determine the 1519length of the activity check during import. 1520.Pp 1521The multihost write period is 1522.Sy zfs_multihost_interval / leaf-vdevs . 1523On average a multihost write will be issued for each leaf vdev 1524every 1525.Sy zfs_multihost_interval 1526milliseconds. 1527In practice, the observed period can vary with the I/O load 1528and this observed value is the delay which is stored in the uberblock. 1529. 1530.It Sy zfs_multihost_import_intervals Ns = Ns Sy 20 Pq uint 1531Used to control the duration of the activity test on import. 1532Smaller values of 1533.Sy zfs_multihost_import_intervals 1534will reduce the import time but increase 1535the risk of failing to detect an active pool. 1536The total activity check time is never allowed to drop below one second. 1537.Pp 1538On import the activity check waits a minimum amount of time determined by 1539.Sy zfs_multihost_interval * zfs_multihost_import_intervals , 1540or the same product computed on the host which last had the pool imported, 1541whichever is greater. 1542The activity check time may be further extended if the value of MMP 1543delay found in the best uberblock indicates actual multihost updates happened 1544at longer intervals than 1545.Sy zfs_multihost_interval . 1546A minimum of 1547.Em 100ms 1548is enforced. 1549.Pp 1550.Sy 0 No is equivalent to Sy 1 . 1551. 1552.It Sy zfs_multihost_fail_intervals Ns = Ns Sy 10 Pq uint 1553Controls the behavior of the pool when multihost write failures or delays are 1554detected. 1555.Pp 1556When 1557.Sy 0 , 1558multihost write failures or delays are ignored. 1559The failures will still be reported to the ZED which depending on 1560its configuration may take action such as suspending the pool or offlining a 1561device. 1562.Pp 1563Otherwise, the pool will be suspended if 1564.Sy zfs_multihost_fail_intervals * zfs_multihost_interval 1565milliseconds pass without a successful MMP write. 1566This guarantees the activity test will see MMP writes if the pool is imported. 1567.Sy 1 No is equivalent to Sy 2 ; 1568this is necessary to prevent the pool from being suspended 1569due to normal, small I/O latency variations. 1570. 1571.It Sy zfs_no_scrub_io Ns = Ns Sy 0 Ns | Ns 1 Pq int 1572Set to disable scrub I/O. 1573This results in scrubs not actually scrubbing data and 1574simply doing a metadata crawl of the pool instead. 1575. 1576.It Sy zfs_no_scrub_prefetch Ns = Ns Sy 0 Ns | Ns 1 Pq int 1577Set to disable block prefetching for scrubs. 1578. 1579.It Sy zfs_nocacheflush Ns = Ns Sy 0 Ns | Ns 1 Pq int 1580Disable cache flush operations on disks when writing. 1581Setting this will cause pool corruption on power loss 1582if a volatile out-of-order write cache is enabled. 1583. 1584.It Sy zfs_nopwrite_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int 1585Allow no-operation writes. 1586The occurrence of nopwrites will further depend on other pool properties 1587.Pq i.a. the checksumming and compression algorithms . 1588. 1589.It Sy zfs_dmu_offset_next_sync Ns = Ns Sy 0 Ns | Ns 1 Pq int 1590Enable forcing TXG sync to find holes. 1591When enabled forces ZFS to act like prior versions when 1592.Sy SEEK_HOLE No or Sy SEEK_DATA 1593flags are used, which, when a dnode is dirty, 1594causes TXGs to be synced so that this data can be found. 1595. 1596.It Sy zfs_pd_bytes_max Ns = Ns Sy 52428800 Ns B Po 50MB Pc Pq int 1597The number of bytes which should be prefetched during a pool traversal, like 1598.Nm zfs Cm send 1599or other data crawling operations. 1600. 1601.It Sy zfs_traverse_indirect_prefetch_limit Ns = Ns Sy 32 Pq int 1602The number of blocks pointed by indirect (non-L0) block which should be 1603prefetched during a pool traversal, like 1604.Nm zfs Cm send 1605or other data crawling operations. 1606. 1607.It Sy zfs_per_txg_dirty_frees_percent Ns = Ns Sy 5 Ns % Pq ulong 1608Control percentage of dirtied indirect blocks from frees allowed into one TXG. 1609After this threshold is crossed, additional frees will wait until the next TXG. 1610.Sy 0 No disables this throttle. 1611. 1612.It Sy zfs_prefetch_disable Ns = Ns Sy 0 Ns | Ns 1 Pq int 1613Disable predictive prefetch. 1614Note that it leaves "prescient" prefetch (for. e.g.\& 1615.Nm zfs Cm send ) 1616intact. 1617Unlike predictive prefetch, prescient prefetch never issues I/O 1618that ends up not being needed, so it can't hurt performance. 1619. 1620.It Sy zfs_qat_checksum_disable Ns = Ns Sy 0 Ns | Ns 1 Pq int 1621Disable QAT hardware acceleration for SHA256 checksums. 1622May be unset after the ZFS modules have been loaded to initialize the QAT 1623hardware as long as support is compiled in and the QAT driver is present. 1624. 1625.It Sy zfs_qat_compress_disable Ns = Ns Sy 0 Ns | Ns 1 Pq int 1626Disable QAT hardware acceleration for gzip compression. 1627May be unset after the ZFS modules have been loaded to initialize the QAT 1628hardware as long as support is compiled in and the QAT driver is present. 1629. 1630.It Sy zfs_qat_encrypt_disable Ns = Ns Sy 0 Ns | Ns 1 Pq int 1631Disable QAT hardware acceleration for AES-GCM encryption. 1632May be unset after the ZFS modules have been loaded to initialize the QAT 1633hardware as long as support is compiled in and the QAT driver is present. 1634. 1635.It Sy zfs_vnops_read_chunk_size Ns = Ns Sy 1048576 Ns B Po 1MB Pc Pq long 1636Bytes to read per chunk. 1637. 1638.It Sy zfs_read_history Ns = Ns Sy 0 Pq int 1639Historical statistics for this many latest reads will be available in 1640.Pa /proc/spl/kstat/zfs/ Ns Ao Ar pool Ac Ns Pa /reads . 1641. 1642.It Sy zfs_read_history_hits Ns = Ns Sy 0 Ns | Ns 1 Pq int 1643Include cache hits in read history 1644. 1645.It Sy zfs_rebuild_max_segment Ns = Ns Sy 1048576 Ns B Po 1MB Pc Pq ulong 1646Maximum read segment size to issue when sequentially resilvering a 1647top-level vdev. 1648. 1649.It Sy zfs_rebuild_scrub_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int 1650Automatically start a pool scrub when the last active sequential resilver 1651completes in order to verify the checksums of all blocks which have been 1652resilvered. 1653This is enabled by default and strongly recommended. 1654. 1655.It Sy zfs_rebuild_vdev_limit Ns = Ns Sy 33554432 Ns B Po 32MB Pc Pq ulong 1656Maximum amount of I/O that can be concurrently issued for a sequential 1657resilver per leaf device, given in bytes. 1658. 1659.It Sy zfs_reconstruct_indirect_combinations_max Ns = Ns Sy 4096 Pq int 1660If an indirect split block contains more than this many possible unique 1661combinations when being reconstructed, consider it too computationally 1662expensive to check them all. 1663Instead, try at most this many randomly selected 1664combinations each time the block is accessed. 1665This allows all segment copies to participate fairly 1666in the reconstruction when all combinations 1667cannot be checked and prevents repeated use of one bad copy. 1668. 1669.It Sy zfs_recover Ns = Ns Sy 0 Ns | Ns 1 Pq int 1670Set to attempt to recover from fatal errors. 1671This should only be used as a last resort, 1672as it typically results in leaked space, or worse. 1673. 1674.It Sy zfs_removal_ignore_errors Ns = Ns Sy 0 Ns | Ns 1 Pq int 1675Ignore hard IO errors during device removal. 1676When set, if a device encounters a hard IO error during the removal process 1677the removal will not be cancelled. 1678This can result in a normally recoverable block becoming permanently damaged 1679and is hence not recommended. 1680This should only be used as a last resort when the 1681pool cannot be returned to a healthy state prior to removing the device. 1682. 1683.It Sy zfs_removal_suspend_progress Ns = Ns Sy 0 Ns | Ns 1 Pq int 1684This is used by the test suite so that it can ensure that certain actions 1685happen while in the middle of a removal. 1686. 1687.It Sy zfs_remove_max_segment Ns = Ns Sy 16777216 Ns B Po 16MB Pc Pq int 1688The largest contiguous segment that we will attempt to allocate when removing 1689a device. 1690If there is a performance problem with attempting to allocate large blocks, 1691consider decreasing this. 1692The default value is also the maximum. 1693. 1694.It Sy zfs_resilver_disable_defer Ns = Ns Sy 0 Ns | Ns 1 Pq int 1695Ignore the 1696.Sy resilver_defer 1697feature, causing an operation that would start a resilver to 1698immediately restart the one in progress. 1699. 1700.It Sy zfs_resilver_min_time_ms Ns = Ns Sy 3000 Ns ms Po 3s Pc Pq int 1701Resilvers are processed by the sync thread. 1702While resilvering, it will spend at least this much time 1703working on a resilver between TXG flushes. 1704. 1705.It Sy zfs_scan_ignore_errors Ns = Ns Sy 0 Ns | Ns 1 Pq int 1706If set, remove the DTL (dirty time list) upon completion of a pool scan (scrub), 1707even if there were unrepairable errors. 1708Intended to be used during pool repair or recovery to 1709stop resilvering when the pool is next imported. 1710. 1711.It Sy zfs_scrub_min_time_ms Ns = Ns Sy 1000 Ns ms Po 1s Pc Pq int 1712Scrubs are processed by the sync thread. 1713While scrubbing, it will spend at least this much time 1714working on a scrub between TXG flushes. 1715. 1716.It Sy zfs_scan_checkpoint_intval Ns = Ns Sy 7200 Ns s Po 2h Pc Pq int 1717To preserve progress across reboots, the sequential scan algorithm periodically 1718needs to stop metadata scanning and issue all the verification I/O to disk. 1719The frequency of this flushing is determined by this tunable. 1720. 1721.It Sy zfs_scan_fill_weight Ns = Ns Sy 3 Pq int 1722This tunable affects how scrub and resilver I/O segments are ordered. 1723A higher number indicates that we care more about how filled in a segment is, 1724while a lower number indicates we care more about the size of the extent without 1725considering the gaps within a segment. 1726This value is only tunable upon module insertion. 1727Changing the value afterwards will have no affect on scrub or resilver performance. 1728. 1729.It Sy zfs_scan_issue_strategy Ns = Ns Sy 0 Pq int 1730Determines the order that data will be verified while scrubbing or resilvering: 1731.Bl -tag -compact -offset 4n -width "a" 1732.It Sy 1 1733Data will be verified as sequentially as possible, given the 1734amount of memory reserved for scrubbing 1735.Pq see Sy zfs_scan_mem_lim_fact . 1736This may improve scrub performance if the pool's data is very fragmented. 1737.It Sy 2 1738The largest mostly-contiguous chunk of found data will be verified first. 1739By deferring scrubbing of small segments, we may later find adjacent data 1740to coalesce and increase the segment size. 1741.It Sy 0 1742.No Use strategy Sy 1 No during normal verification 1743.No and strategy Sy 2 No while taking a checkpoint. 1744.El 1745. 1746.It Sy zfs_scan_legacy Ns = Ns Sy 0 Ns | Ns 1 Pq int 1747If unset, indicates that scrubs and resilvers will gather metadata in 1748memory before issuing sequential I/O. 1749Otherwise indicates that the legacy algorithm will be used, 1750where I/O is initiated as soon as it is discovered. 1751Unsetting will not affect scrubs or resilvers that are already in progress. 1752. 1753.It Sy zfs_scan_max_ext_gap Ns = Ns Sy 2097152 Ns B Po 2MB Pc Pq int 1754Sets the largest gap in bytes between scrub/resilver I/O operations 1755that will still be considered sequential for sorting purposes. 1756Changing this value will not 1757affect scrubs or resilvers that are already in progress. 1758. 1759.It Sy zfs_scan_mem_lim_fact Ns = Ns Sy 20 Ns ^-1 Pq int 1760Maximum fraction of RAM used for I/O sorting by sequential scan algorithm. 1761This tunable determines the hard limit for I/O sorting memory usage. 1762When the hard limit is reached we stop scanning metadata and start issuing 1763data verification I/O. 1764This is done until we get below the soft limit. 1765. 1766.It Sy zfs_scan_mem_lim_soft_fact Ns = Ns Sy 20 Ns ^-1 Pq int 1767The fraction of the hard limit used to determined the soft limit for I/O sorting 1768by the sequential scan algorithm. 1769When we cross this limit from below no action is taken. 1770When we cross this limit from above it is because we are issuing verification I/O. 1771In this case (unless the metadata scan is done) we stop issuing verification I/O 1772and start scanning metadata again until we get to the hard limit. 1773. 1774.It Sy zfs_scan_strict_mem_lim Ns = Ns Sy 0 Ns | Ns 1 Pq int 1775Enforce tight memory limits on pool scans when a sequential scan is in progress. 1776When disabled, the memory limit may be exceeded by fast disks. 1777. 1778.It Sy zfs_scan_suspend_progress Ns = Ns Sy 0 Ns | Ns 1 Pq int 1779Freezes a scrub/resilver in progress without actually pausing it. 1780Intended for testing/debugging. 1781. 1782.It Sy zfs_scan_vdev_limit Ns = Ns Sy 4194304 Ns B Po 4MB Pc Pq int 1783Maximum amount of data that can be concurrently issued at once for scrubs and 1784resilvers per leaf device, given in bytes. 1785. 1786.It Sy zfs_send_corrupt_data Ns = Ns Sy 0 Ns | Ns 1 Pq int 1787Allow sending of corrupt data (ignore read/checksum errors when sending). 1788. 1789.It Sy zfs_send_unmodified_spill_blocks Ns = Ns Sy 1 Ns | Ns 0 Pq int 1790Include unmodified spill blocks in the send stream. 1791Under certain circumstances, previous versions of ZFS could incorrectly 1792remove the spill block from an existing object. 1793Including unmodified copies of the spill blocks creates a backwards-compatible 1794stream which will recreate a spill block if it was incorrectly removed. 1795. 1796.It Sy zfs_send_no_prefetch_queue_ff Ns = Ns Sy 20 Ns ^-1 Pq int 1797The fill fraction of the 1798.Nm zfs Cm send 1799internal queues. 1800The fill fraction controls the timing with which internal threads are woken up. 1801. 1802.It Sy zfs_send_no_prefetch_queue_length Ns = Ns Sy 1048576 Ns B Po 1MB Pc Pq int 1803The maximum number of bytes allowed in 1804.Nm zfs Cm send Ns 's 1805internal queues. 1806. 1807.It Sy zfs_send_queue_ff Ns = Ns Sy 20 Ns ^-1 Pq int 1808The fill fraction of the 1809.Nm zfs Cm send 1810prefetch queue. 1811The fill fraction controls the timing with which internal threads are woken up. 1812. 1813.It Sy zfs_send_queue_length Ns = Ns Sy 16777216 Ns B Po 16MB Pc Pq int 1814The maximum number of bytes allowed that will be prefetched by 1815.Nm zfs Cm send . 1816This value must be at least twice the maximum block size in use. 1817. 1818.It Sy zfs_recv_queue_ff Ns = Ns Sy 20 Ns ^-1 Pq int 1819The fill fraction of the 1820.Nm zfs Cm receive 1821queue. 1822The fill fraction controls the timing with which internal threads are woken up. 1823. 1824.It Sy zfs_recv_queue_length Ns = Ns Sy 16777216 Ns B Po 16MB Pc Pq int 1825The maximum number of bytes allowed in the 1826.Nm zfs Cm receive 1827queue. 1828This value must be at least twice the maximum block size in use. 1829. 1830.It Sy zfs_recv_write_batch_size Ns = Ns Sy 1048576 Ns B Po 1MB Pc Pq int 1831The maximum amount of data, in bytes, that 1832.Nm zfs Cm receive 1833will write in one DMU transaction. 1834This is the uncompressed size, even when receiving a compressed send stream. 1835This setting will not reduce the write size below a single block. 1836Capped at a maximum of 1837.Sy 32MB . 1838. 1839.It Sy zfs_override_estimate_recordsize Ns = Ns Sy 0 Ns | Ns 1 Pq ulong 1840Setting this variable overrides the default logic for estimating block 1841sizes when doing a 1842.Nm zfs Cm send . 1843The default heuristic is that the average block size 1844will be the current recordsize. 1845Override this value if most data in your dataset is not of that size 1846and you require accurate zfs send size estimates. 1847. 1848.It Sy zfs_sync_pass_deferred_free Ns = Ns Sy 2 Pq int 1849Flushing of data to disk is done in passes. 1850Defer frees starting in this pass. 1851. 1852.It Sy zfs_spa_discard_memory_limit Ns = Ns Sy 16777216 Ns B Po 16MB Pc Pq int 1853Maximum memory used for prefetching a checkpoint's space map on each 1854vdev while discarding the checkpoint. 1855. 1856.It Sy zfs_special_class_metadata_reserve_pct Ns = Ns Sy 25 Ns % Pq int 1857Only allow small data blocks to be allocated on the special and dedup vdev 1858types when the available free space percentage on these vdevs exceeds this value. 1859This ensures reserved space is available for pool metadata as the 1860special vdevs approach capacity. 1861. 1862.It Sy zfs_sync_pass_dont_compress Ns = Ns Sy 8 Pq int 1863Starting in this sync pass, disable compression (including of metadata). 1864With the default setting, in practice, we don't have this many sync passes, 1865so this has no effect. 1866.Pp 1867The original intent was that disabling compression would help the sync passes 1868to converge. 1869However, in practice, disabling compression increases 1870the average number of sync passes; because when we turn compression off, 1871many blocks' size will change, and thus we have to re-allocate 1872(not overwrite) them. 1873It also increases the number of 1874.Em 128kB 1875allocations (e.g. for indirect blocks and spacemaps) 1876because these will not be compressed. 1877The 1878.Em 128kB 1879allocations are especially detrimental to performance 1880on highly fragmented systems, which may have very few free segments of this size, 1881and may need to load new metaslabs to satisfy these allocations. 1882. 1883.It Sy zfs_sync_pass_rewrite Ns = Ns Sy 2 Pq int 1884Rewrite new block pointers starting in this pass. 1885. 1886.It Sy zfs_sync_taskq_batch_pct Ns = Ns Sy 75 Ns % Pq int 1887This controls the number of threads used by 1888.Sy dp_sync_taskq . 1889The default value of 1890.Sy 75% 1891will create a maximum of one thread per CPU. 1892. 1893.It Sy zfs_trim_extent_bytes_max Ns = Ns Sy 134217728 Ns B Po 128MB Pc Pq uint 1894Maximum size of TRIM command. 1895Larger ranges will be split into chunks no larger than this value before issuing. 1896. 1897.It Sy zfs_trim_extent_bytes_min Ns = Ns Sy 32768 Ns B Po 32kB Pc Pq uint 1898Minimum size of TRIM commands. 1899TRIM ranges smaller than this will be skipped, 1900unless they're part of a larger range which was chunked. 1901This is done because it's common for these small TRIMs 1902to negatively impact overall performance. 1903. 1904.It Sy zfs_trim_metaslab_skip Ns = Ns Sy 0 Ns | Ns 1 Pq uint 1905Skip uninitialized metaslabs during the TRIM process. 1906This option is useful for pools constructed from large thinly-provisioned devices 1907where TRIM operations are slow. 1908As a pool ages, an increasing fraction of the pool's metaslabs 1909will be initialized, progressively degrading the usefulness of this option. 1910This setting is stored when starting a manual TRIM and will 1911persist for the duration of the requested TRIM. 1912. 1913.It Sy zfs_trim_queue_limit Ns = Ns Sy 10 Pq uint 1914Maximum number of queued TRIMs outstanding per leaf vdev. 1915The number of concurrent TRIM commands issued to the device is controlled by 1916.Sy zfs_vdev_trim_min_active No and Sy zfs_vdev_trim_max_active . 1917. 1918.It Sy zfs_trim_txg_batch Ns = Ns Sy 32 Pq uint 1919The number of transaction groups' worth of frees which should be aggregated 1920before TRIM operations are issued to the device. 1921This setting represents a trade-off between issuing larger, 1922more efficient TRIM operations and the delay 1923before the recently trimmed space is available for use by the device. 1924.Pp 1925Increasing this value will allow frees to be aggregated for a longer time. 1926This will result is larger TRIM operations and potentially increased memory usage. 1927Decreasing this value will have the opposite effect. 1928The default of 1929.Sy 32 1930was determined to be a reasonable compromise. 1931. 1932.It Sy zfs_txg_history Ns = Ns Sy 0 Pq int 1933Historical statistics for this many latest TXGs will be available in 1934.Pa /proc/spl/kstat/zfs/ Ns Ao Ar pool Ac Ns Pa /TXGs . 1935. 1936.It Sy zfs_txg_timeout Ns = Ns Sy 5 Ns s Pq int 1937Flush dirty data to disk at least every this many seconds (maximum TXG duration). 1938. 1939.It Sy zfs_vdev_aggregate_trim Ns = Ns Sy 0 Ns | Ns 1 Pq int 1940Allow TRIM I/Os to be aggregated. 1941This is normally not helpful because the extents to be trimmed 1942will have been already been aggregated by the metaslab. 1943This option is provided for debugging and performance analysis. 1944. 1945.It Sy zfs_vdev_aggregation_limit Ns = Ns Sy 1048576 Ns B Po 1MB Pc Pq int 1946Max vdev I/O aggregation size. 1947. 1948.It Sy zfs_vdev_aggregation_limit_non_rotating Ns = Ns Sy 131072 Ns B Po 128kB Pc Pq int 1949Max vdev I/O aggregation size for non-rotating media. 1950. 1951.It Sy zfs_vdev_cache_bshift Ns = Ns Sy 16 Po 64kB Pc Pq int 1952Shift size to inflate reads to. 1953. 1954.It Sy zfs_vdev_cache_max Ns = Ns Sy 16384 Ns B Po 16kB Pc Pq int 1955Inflate reads smaller than this value to meet the 1956.Sy zfs_vdev_cache_bshift 1957size 1958.Pq default Sy 64kB . 1959. 1960.It Sy zfs_vdev_cache_size Ns = Ns Sy 0 Pq int 1961Total size of the per-disk cache in bytes. 1962.Pp 1963Currently this feature is disabled, as it has been found to not be helpful 1964for performance and in some cases harmful. 1965. 1966.It Sy zfs_vdev_mirror_rotating_inc Ns = Ns Sy 0 Pq int 1967A number by which the balancing algorithm increments the load calculation for 1968the purpose of selecting the least busy mirror member when an I/O operation 1969immediately follows its predecessor on rotational vdevs 1970for the purpose of making decisions based on load. 1971. 1972.It Sy zfs_vdev_mirror_rotating_seek_inc Ns = Ns Sy 5 Pq int 1973A number by which the balancing algorithm increments the load calculation for 1974the purpose of selecting the least busy mirror member when an I/O operation 1975lacks locality as defined by 1976.Sy zfs_vdev_mirror_rotating_seek_offset . 1977Operations within this that are not immediately following the previous operation 1978are incremented by half. 1979. 1980.It Sy zfs_vdev_mirror_rotating_seek_offset Ns = Ns Sy 1048576 Ns B Po 1MB Pc Pq int 1981The maximum distance for the last queued I/O operation in which 1982the balancing algorithm considers an operation to have locality. 1983.No See Sx ZFS I/O SCHEDULER . 1984. 1985.It Sy zfs_vdev_mirror_non_rotating_inc Ns = Ns Sy 0 Pq int 1986A number by which the balancing algorithm increments the load calculation for 1987the purpose of selecting the least busy mirror member on non-rotational vdevs 1988when I/O operations do not immediately follow one another. 1989. 1990.It Sy zfs_vdev_mirror_non_rotating_seek_inc Ns = Ns Sy 1 Pq int 1991A number by which the balancing algorithm increments the load calculation for 1992the purpose of selecting the least busy mirror member when an I/O operation lacks 1993locality as defined by the 1994.Sy zfs_vdev_mirror_rotating_seek_offset . 1995Operations within this that are not immediately following the previous operation 1996are incremented by half. 1997. 1998.It Sy zfs_vdev_read_gap_limit Ns = Ns Sy 32768 Ns B Po 32kB Pc Pq int 1999Aggregate read I/O operations if the on-disk gap between them is within this 2000threshold. 2001. 2002.It Sy zfs_vdev_write_gap_limit Ns = Ns Sy 4096 Ns B Po 4kB Pc Pq int 2003Aggregate write I/O operations if the on-disk gap between them is within this 2004threshold. 2005. 2006.It Sy zfs_vdev_raidz_impl Ns = Ns Sy fastest Pq string 2007Select the raidz parity implementation to use. 2008.Pp 2009Variants that don't depend on CPU-specific features 2010may be selected on module load, as they are supported on all systems. 2011The remaining options may only be set after the module is loaded, 2012as they are available only if the implementations are compiled in 2013and supported on the running system. 2014.Pp 2015Once the module is loaded, 2016.Pa /sys/module/zfs/parameters/zfs_vdev_raidz_impl 2017will show the available options, 2018with the currently selected one enclosed in square brackets. 2019.Pp 2020.TS 2021lb l l . 2022fastest selected by built-in benchmark 2023original original implementation 2024scalar scalar implementation 2025sse2 SSE2 instruction set 64-bit x86 2026ssse3 SSSE3 instruction set 64-bit x86 2027avx2 AVX2 instruction set 64-bit x86 2028avx512f AVX512F instruction set 64-bit x86 2029avx512bw AVX512F & AVX512BW instruction sets 64-bit x86 2030aarch64_neon NEON Aarch64/64-bit ARMv8 2031aarch64_neonx2 NEON with more unrolling Aarch64/64-bit ARMv8 2032powerpc_altivec Altivec PowerPC 2033.TE 2034. 2035.It Sy zfs_vdev_scheduler Pq charp 2036.Sy DEPRECATED . 2037Prints warning to kernel log for compatibility. 2038. 2039.It Sy zfs_zevent_len_max Ns = Ns Sy 512 Pq int 2040Max event queue length. 2041Events in the queue can be viewed with 2042.Xr zpool-events 8 . 2043. 2044.It Sy zfs_zevent_retain_max Ns = Ns Sy 2000 Pq int 2045Maximum recent zevent records to retain for duplicate checking. 2046Setting this to 2047.Sy 0 2048disables duplicate detection. 2049. 2050.It Sy zfs_zevent_retain_expire_secs Ns = Ns Sy 900 Ns s Po 15min Pc Pq int 2051Lifespan for a recent ereport that was retained for duplicate checking. 2052. 2053.It Sy zfs_zil_clean_taskq_maxalloc Ns = Ns Sy 1048576 Pq int 2054The maximum number of taskq entries that are allowed to be cached. 2055When this limit is exceeded transaction records (itxs) 2056will be cleaned synchronously. 2057. 2058.It Sy zfs_zil_clean_taskq_minalloc Ns = Ns Sy 1024 Pq int 2059The number of taskq entries that are pre-populated when the taskq is first 2060created and are immediately available for use. 2061. 2062.It Sy zfs_zil_clean_taskq_nthr_pct Ns = Ns Sy 100 Ns % Pq int 2063This controls the number of threads used by 2064.Sy dp_zil_clean_taskq . 2065The default value of 2066.Sy 100% 2067will create a maximum of one thread per cpu. 2068. 2069.It Sy zil_maxblocksize Ns = Ns Sy 131072 Ns B Po 128kB Pc Pq int 2070This sets the maximum block size used by the ZIL. 2071On very fragmented pools, lowering this 2072.Pq typically to Sy 36kB 2073can improve performance. 2074. 2075.It Sy zil_nocacheflush Ns = Ns Sy 0 Ns | Ns 1 Pq int 2076Disable the cache flush commands that are normally sent to disk by 2077the ZIL after an LWB write has completed. 2078Setting this will cause ZIL corruption on power loss 2079if a volatile out-of-order write cache is enabled. 2080. 2081.It Sy zil_replay_disable Ns = Ns Sy 0 Ns | Ns 1 Pq int 2082Disable intent logging replay. 2083Can be disabled for recovery from corrupted ZIL. 2084. 2085.It Sy zil_slog_bulk Ns = Ns Sy 786432 Ns B Po 768kB Pc Pq ulong 2086Limit SLOG write size per commit executed with synchronous priority. 2087Any writes above that will be executed with lower (asynchronous) priority 2088to limit potential SLOG device abuse by single active ZIL writer. 2089. 2090.It Sy zfs_embedded_slog_min_ms Ns = Ns Sy 64 Pq int 2091Usually, one metaslab from each normal-class vdev is dedicated for use by 2092the ZIL to log synchronous writes. 2093However, if there are fewer than 2094.Sy zfs_embedded_slog_min_ms 2095metaslabs in the vdev, this functionality is disabled. 2096This ensures that we don't set aside an unreasonable amount of space for the ZIL. 2097. 2098.It Sy zio_deadman_log_all Ns = Ns Sy 0 Ns | Ns 1 Pq int 2099If non-zero, the zio deadman will produce debugging messages 2100.Pq see Sy zfs_dbgmsg_enable 2101for all zios, rather than only for leaf zios possessing a vdev. 2102This is meant to be used by developers to gain 2103diagnostic information for hang conditions which don't involve a mutex 2104or other locking primitive: typically conditions in which a thread in 2105the zio pipeline is looping indefinitely. 2106. 2107.It Sy zio_slow_io_ms Ns = Ns Sy 30000 Ns ms Po 30s Pc Pq int 2108When an I/O operation takes more than this much time to complete, 2109it's marked as slow. 2110Each slow operation causes a delay zevent. 2111Slow I/O counters can be seen with 2112.Nm zpool Cm status Fl s . 2113. 2114.It Sy zio_dva_throttle_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int 2115Throttle block allocations in the I/O pipeline. 2116This allows for dynamic allocation distribution when devices are imbalanced. 2117When enabled, the maximum number of pending allocations per top-level vdev 2118is limited by 2119.Sy zfs_vdev_queue_depth_pct . 2120. 2121.It Sy zio_requeue_io_start_cut_in_line Ns = Ns Sy 0 Ns | Ns 1 Pq int 2122Prioritize requeued I/O. 2123. 2124.It Sy zio_taskq_batch_pct Ns = Ns Sy 80 Ns % Pq uint 2125Percentage of online CPUs which will run a worker thread for I/O. 2126These workers are responsible for I/O work such as compression and 2127checksum calculations. 2128Fractional number of CPUs will be rounded down. 2129.Pp 2130The default value of 2131.Sy 80% 2132was chosen to avoid using all CPUs which can result in 2133latency issues and inconsistent application performance, 2134especially when slower compression and/or checksumming is enabled. 2135. 2136.It Sy zio_taskq_batch_tpq Ns = Ns Sy 0 Pq uint 2137Number of worker threads per taskq. 2138Lower values improve I/O ordering and CPU utilization, 2139while higher reduces lock contention. 2140.Pp 2141If 2142.Sy 0 , 2143generate a system-dependent value close to 6 threads per taskq. 2144. 2145.It Sy zvol_inhibit_dev Ns = Ns Sy 0 Ns | Ns 1 Pq uint 2146Do not create zvol device nodes. 2147This may slightly improve startup time on 2148systems with a very large number of zvols. 2149. 2150.It Sy zvol_major Ns = Ns Sy 230 Pq uint 2151Major number for zvol block devices. 2152. 2153.It Sy zvol_max_discard_blocks Ns = Ns Sy 16384 Pq ulong 2154Discard (TRIM) operations done on zvols will be done in batches of this 2155many blocks, where block size is determined by the 2156.Sy volblocksize 2157property of a zvol. 2158. 2159.It Sy zvol_prefetch_bytes Ns = Ns Sy 131072 Ns B Po 128kB Pc Pq uint 2160When adding a zvol to the system, prefetch this many bytes 2161from the start and end of the volume. 2162Prefetching these regions of the volume is desirable, 2163because they are likely to be accessed immediately by 2164.Xr blkid 8 2165or the kernel partitioner. 2166. 2167.It Sy zvol_request_sync Ns = Ns Sy 0 Ns | Ns 1 Pq uint 2168When processing I/O requests for a zvol, submit them synchronously. 2169This effectively limits the queue depth to 2170.Em 1 2171for each I/O submitter. 2172When unset, requests are handled asynchronously by a thread pool. 2173The number of requests which can be handled concurrently is controlled by 2174.Sy zvol_threads . 2175. 2176.It Sy zvol_threads Ns = Ns Sy 32 Pq uint 2177Max number of threads which can handle zvol I/O requests concurrently. 2178. 2179.It Sy zvol_volmode Ns = Ns Sy 1 Pq uint 2180Defines zvol block devices behaviour when 2181.Sy volmode Ns = Ns Sy default : 2182.Bl -tag -compact -offset 4n -width "a" 2183.It Sy 1 2184.No equivalent to Sy full 2185.It Sy 2 2186.No equivalent to Sy dev 2187.It Sy 3 2188.No equivalent to Sy none 2189.El 2190.El 2191. 2192.Sh ZFS I/O SCHEDULER 2193ZFS issues I/O operations to leaf vdevs to satisfy and complete I/O operations. 2194The scheduler determines when and in what order those operations are issued. 2195The scheduler divides operations into five I/O classes, 2196prioritized in the following order: sync read, sync write, async read, 2197async write, and scrub/resilver. 2198Each queue defines the minimum and maximum number of concurrent operations 2199that may be issued to the device. 2200In addition, the device has an aggregate maximum, 2201.Sy zfs_vdev_max_active . 2202Note that the sum of the per-queue minima must not exceed the aggregate maximum. 2203If the sum of the per-queue maxima exceeds the aggregate maximum, 2204then the number of active operations may reach 2205.Sy zfs_vdev_max_active , 2206in which case no further operations will be issued, 2207regardless of whether all per-queue minima have been met. 2208.Pp 2209For many physical devices, throughput increases with the number of 2210concurrent operations, but latency typically suffers. 2211Furthermore, physical devices typically have a limit 2212at which more concurrent operations have no 2213effect on throughput or can actually cause it to decrease. 2214.Pp 2215The scheduler selects the next operation to issue by first looking for an 2216I/O class whose minimum has not been satisfied. 2217Once all are satisfied and the aggregate maximum has not been hit, 2218the scheduler looks for classes whose maximum has not been satisfied. 2219Iteration through the I/O classes is done in the order specified above. 2220No further operations are issued 2221if the aggregate maximum number of concurrent operations has been hit, 2222or if there are no operations queued for an I/O class that has not hit its maximum. 2223Every time an I/O operation is queued or an operation completes, 2224the scheduler looks for new operations to issue. 2225.Pp 2226In general, smaller 2227.Sy max_active Ns s 2228will lead to lower latency of synchronous operations. 2229Larger 2230.Sy max_active Ns s 2231may lead to higher overall throughput, depending on underlying storage. 2232.Pp 2233The ratio of the queues' 2234.Sy max_active Ns s 2235determines the balance of performance between reads, writes, and scrubs. 2236For example, increasing 2237.Sy zfs_vdev_scrub_max_active 2238will cause the scrub or resilver to complete more quickly, 2239but reads and writes to have higher latency and lower throughput. 2240.Pp 2241All I/O classes have a fixed maximum number of outstanding operations, 2242except for the async write class. 2243Asynchronous writes represent the data that is committed to stable storage 2244during the syncing stage for transaction groups. 2245Transaction groups enter the syncing state periodically, 2246so the number of queued async writes will quickly burst up 2247and then bleed down to zero. 2248Rather than servicing them as quickly as possible, 2249the I/O scheduler changes the maximum number of active async write operations 2250according to the amount of dirty data in the pool. 2251Since both throughput and latency typically increase with the number of 2252concurrent operations issued to physical devices, reducing the 2253burstiness in the number of concurrent operations also stabilizes the 2254response time of operations from other – and in particular synchronous – queues. 2255In broad strokes, the I/O scheduler will issue more concurrent operations 2256from the async write queue as there's more dirty data in the pool. 2257. 2258.Ss Async Writes 2259The number of concurrent operations issued for the async write I/O class 2260follows a piece-wise linear function defined by a few adjustable points: 2261.Bd -literal 2262 | o---------| <-- \fBzfs_vdev_async_write_max_active\fP 2263 ^ | /^ | 2264 | | / | | 2265active | / | | 2266 I/O | / | | 2267count | / | | 2268 | / | | 2269 |-------o | | <-- \fBzfs_vdev_async_write_min_active\fP 2270 0|_______^______|_________| 2271 0% | | 100% of \fBzfs_dirty_data_max\fP 2272 | | 2273 | `-- \fBzfs_vdev_async_write_active_max_dirty_percent\fP 2274 `--------- \fBzfs_vdev_async_write_active_min_dirty_percent\fP 2275.Ed 2276.Pp 2277Until the amount of dirty data exceeds a minimum percentage of the dirty 2278data allowed in the pool, the I/O scheduler will limit the number of 2279concurrent operations to the minimum. 2280As that threshold is crossed, the number of concurrent operations issued 2281increases linearly to the maximum at the specified maximum percentage 2282of the dirty data allowed in the pool. 2283.Pp 2284Ideally, the amount of dirty data on a busy pool will stay in the sloped 2285part of the function between 2286.Sy zfs_vdev_async_write_active_min_dirty_percent 2287and 2288.Sy zfs_vdev_async_write_active_max_dirty_percent . 2289If it exceeds the maximum percentage, 2290this indicates that the rate of incoming data is 2291greater than the rate that the backend storage can handle. 2292In this case, we must further throttle incoming writes, 2293as described in the next section. 2294. 2295.Sh ZFS TRANSACTION DELAY 2296We delay transactions when we've determined that the backend storage 2297isn't able to accommodate the rate of incoming writes. 2298.Pp 2299If there is already a transaction waiting, we delay relative to when 2300that transaction will finish waiting. 2301This way the calculated delay time 2302is independent of the number of threads concurrently executing transactions. 2303.Pp 2304If we are the only waiter, wait relative to when the transaction started, 2305rather than the current time. 2306This credits the transaction for "time already served", 2307e.g. reading indirect blocks. 2308.Pp 2309The minimum time for a transaction to take is calculated as 2310.Dl min_time = min( Ns Sy zfs_delay_scale No * (dirty - min) / (max - dirty), 100ms) 2311.Pp 2312The delay has two degrees of freedom that can be adjusted via tunables. 2313The percentage of dirty data at which we start to delay is defined by 2314.Sy zfs_delay_min_dirty_percent . 2315This should typically be at or above 2316.Sy zfs_vdev_async_write_active_max_dirty_percent , 2317so that we only start to delay after writing at full speed 2318has failed to keep up with the incoming write rate. 2319The scale of the curve is defined by 2320.Sy zfs_delay_scale . 2321Roughly speaking, this variable determines the amount of delay at the midpoint of the curve. 2322.Bd -literal 2323delay 2324 10ms +-------------------------------------------------------------*+ 2325 | *| 2326 9ms + *+ 2327 | *| 2328 8ms + *+ 2329 | * | 2330 7ms + * + 2331 | * | 2332 6ms + * + 2333 | * | 2334 5ms + * + 2335 | * | 2336 4ms + * + 2337 | * | 2338 3ms + * + 2339 | * | 2340 2ms + (midpoint) * + 2341 | | ** | 2342 1ms + v *** + 2343 | \fBzfs_delay_scale\fP ----------> ******** | 2344 0 +-------------------------------------*********----------------+ 2345 0% <- \fBzfs_dirty_data_max\fP -> 100% 2346.Ed 2347.Pp 2348Note, that since the delay is added to the outstanding time remaining on the 2349most recent transaction it's effectively the inverse of IOPS. 2350Here, the midpoint of 2351.Em 500us 2352translates to 2353.Em 2000 IOPS . 2354The shape of the curve 2355was chosen such that small changes in the amount of accumulated dirty data 2356in the first three quarters of the curve yield relatively small differences 2357in the amount of delay. 2358.Pp 2359The effects can be easier to understand when the amount of delay is 2360represented on a logarithmic scale: 2361.Bd -literal 2362delay 2363100ms +-------------------------------------------------------------++ 2364 + + 2365 | | 2366 + *+ 2367 10ms + *+ 2368 + ** + 2369 | (midpoint) ** | 2370 + | ** + 2371 1ms + v **** + 2372 + \fBzfs_delay_scale\fP ----------> ***** + 2373 | **** | 2374 + **** + 2375100us + ** + 2376 + * + 2377 | * | 2378 + * + 2379 10us + * + 2380 + + 2381 | | 2382 + + 2383 +--------------------------------------------------------------+ 2384 0% <- \fBzfs_dirty_data_max\fP -> 100% 2385.Ed 2386.Pp 2387Note here that only as the amount of dirty data approaches its limit does 2388the delay start to increase rapidly. 2389The goal of a properly tuned system should be to keep the amount of dirty data 2390out of that range by first ensuring that the appropriate limits are set 2391for the I/O scheduler to reach optimal throughput on the back-end storage, 2392and then by changing the value of 2393.Sy zfs_delay_scale 2394to increase the steepness of the curve. 2395