1.. _cgroup-v2: 2 3================ 4Control Group v2 5================ 6 7:Date: October, 2015 8:Author: Tejun Heo <tj@kernel.org> 9 10This is the authoritative documentation on the design, interface and 11conventions of cgroup v2. It describes all userland-visible aspects 12of cgroup including core and specific controller behaviors. All 13future changes must be reflected in this document. Documentation for 14v1 is available under :ref:`Documentation/admin-guide/cgroup-v1/index.rst <cgroup-v1>`. 15 16.. CONTENTS 17 18 1. Introduction 19 1-1. Terminology 20 1-2. What is cgroup? 21 2. Basic Operations 22 2-1. Mounting 23 2-2. Organizing Processes and Threads 24 2-2-1. Processes 25 2-2-2. Threads 26 2-3. [Un]populated Notification 27 2-4. Controlling Controllers 28 2-4-1. Enabling and Disabling 29 2-4-2. Top-down Constraint 30 2-4-3. No Internal Process Constraint 31 2-5. Delegation 32 2-5-1. Model of Delegation 33 2-5-2. Delegation Containment 34 2-6. Guidelines 35 2-6-1. Organize Once and Control 36 2-6-2. Avoid Name Collisions 37 3. Resource Distribution Models 38 3-1. Weights 39 3-2. Limits 40 3-3. Protections 41 3-4. Allocations 42 4. Interface Files 43 4-1. Format 44 4-2. Conventions 45 4-3. Core Interface Files 46 5. Controllers 47 5-1. CPU 48 5-1-1. CPU Interface Files 49 5-2. Memory 50 5-2-1. Memory Interface Files 51 5-2-2. Usage Guidelines 52 5-2-3. Memory Ownership 53 5-3. IO 54 5-3-1. IO Interface Files 55 5-3-2. Writeback 56 5-3-3. IO Latency 57 5-3-3-1. How IO Latency Throttling Works 58 5-3-3-2. IO Latency Interface Files 59 5-3-4. IO Priority 60 5-4. PID 61 5-4-1. PID Interface Files 62 5-5. Cpuset 63 5.5-1. Cpuset Interface Files 64 5-6. Device 65 5-7. RDMA 66 5-7-1. RDMA Interface Files 67 5-8. HugeTLB 68 5.8-1. HugeTLB Interface Files 69 5-9. Misc 70 5.9-1 Miscellaneous cgroup Interface Files 71 5.9-2 Migration and Ownership 72 5-10. Others 73 5-10-1. perf_event 74 5-N. Non-normative information 75 5-N-1. CPU controller root cgroup process behaviour 76 5-N-2. IO controller root cgroup process behaviour 77 6. Namespace 78 6-1. Basics 79 6-2. The Root and Views 80 6-3. Migration and setns(2) 81 6-4. Interaction with Other Namespaces 82 P. Information on Kernel Programming 83 P-1. Filesystem Support for Writeback 84 D. Deprecated v1 Core Features 85 R. Issues with v1 and Rationales for v2 86 R-1. Multiple Hierarchies 87 R-2. Thread Granularity 88 R-3. Competition Between Inner Nodes and Threads 89 R-4. Other Interface Issues 90 R-5. Controller Issues and Remedies 91 R-5-1. Memory 92 93 94Introduction 95============ 96 97Terminology 98----------- 99 100"cgroup" stands for "control group" and is never capitalized. The 101singular form is used to designate the whole feature and also as a 102qualifier as in "cgroup controllers". When explicitly referring to 103multiple individual control groups, the plural form "cgroups" is used. 104 105 106What is cgroup? 107--------------- 108 109cgroup is a mechanism to organize processes hierarchically and 110distribute system resources along the hierarchy in a controlled and 111configurable manner. 112 113cgroup is largely composed of two parts - the core and controllers. 114cgroup core is primarily responsible for hierarchically organizing 115processes. A cgroup controller is usually responsible for 116distributing a specific type of system resource along the hierarchy 117although there are utility controllers which serve purposes other than 118resource distribution. 119 120cgroups form a tree structure and every process in the system belongs 121to one and only one cgroup. All threads of a process belong to the 122same cgroup. On creation, all processes are put in the cgroup that 123the parent process belongs to at the time. A process can be migrated 124to another cgroup. Migration of a process doesn't affect already 125existing descendant processes. 126 127Following certain structural constraints, controllers may be enabled or 128disabled selectively on a cgroup. All controller behaviors are 129hierarchical - if a controller is enabled on a cgroup, it affects all 130processes which belong to the cgroups consisting the inclusive 131sub-hierarchy of the cgroup. When a controller is enabled on a nested 132cgroup, it always restricts the resource distribution further. The 133restrictions set closer to the root in the hierarchy can not be 134overridden from further away. 135 136 137Basic Operations 138================ 139 140Mounting 141-------- 142 143Unlike v1, cgroup v2 has only single hierarchy. The cgroup v2 144hierarchy can be mounted with the following mount command:: 145 146 # mount -t cgroup2 none $MOUNT_POINT 147 148cgroup2 filesystem has the magic number 0x63677270 ("cgrp"). All 149controllers which support v2 and are not bound to a v1 hierarchy are 150automatically bound to the v2 hierarchy and show up at the root. 151Controllers which are not in active use in the v2 hierarchy can be 152bound to other hierarchies. This allows mixing v2 hierarchy with the 153legacy v1 multiple hierarchies in a fully backward compatible way. 154 155A controller can be moved across hierarchies only after the controller 156is no longer referenced in its current hierarchy. Because per-cgroup 157controller states are destroyed asynchronously and controllers may 158have lingering references, a controller may not show up immediately on 159the v2 hierarchy after the final umount of the previous hierarchy. 160Similarly, a controller should be fully disabled to be moved out of 161the unified hierarchy and it may take some time for the disabled 162controller to become available for other hierarchies; furthermore, due 163to inter-controller dependencies, other controllers may need to be 164disabled too. 165 166While useful for development and manual configurations, moving 167controllers dynamically between the v2 and other hierarchies is 168strongly discouraged for production use. It is recommended to decide 169the hierarchies and controller associations before starting using the 170controllers after system boot. 171 172During transition to v2, system management software might still 173automount the v1 cgroup filesystem and so hijack all controllers 174during boot, before manual intervention is possible. To make testing 175and experimenting easier, the kernel parameter cgroup_no_v1= allows 176disabling controllers in v1 and make them always available in v2. 177 178cgroup v2 currently supports the following mount options. 179 180 nsdelegate 181 Consider cgroup namespaces as delegation boundaries. This 182 option is system wide and can only be set on mount or modified 183 through remount from the init namespace. The mount option is 184 ignored on non-init namespace mounts. Please refer to the 185 Delegation section for details. 186 187 favordynmods 188 Reduce the latencies of dynamic cgroup modifications such as 189 task migrations and controller on/offs at the cost of making 190 hot path operations such as forks and exits more expensive. 191 The static usage pattern of creating a cgroup, enabling 192 controllers, and then seeding it with CLONE_INTO_CGROUP is 193 not affected by this option. 194 195 memory_localevents 196 Only populate memory.events with data for the current cgroup, 197 and not any subtrees. This is legacy behaviour, the default 198 behaviour without this option is to include subtree counts. 199 This option is system wide and can only be set on mount or 200 modified through remount from the init namespace. The mount 201 option is ignored on non-init namespace mounts. 202 203 memory_recursiveprot 204 Recursively apply memory.min and memory.low protection to 205 entire subtrees, without requiring explicit downward 206 propagation into leaf cgroups. This allows protecting entire 207 subtrees from one another, while retaining free competition 208 within those subtrees. This should have been the default 209 behavior but is a mount-option to avoid regressing setups 210 relying on the original semantics (e.g. specifying bogusly 211 high 'bypass' protection values at higher tree levels). 212 213 214Organizing Processes and Threads 215-------------------------------- 216 217Processes 218~~~~~~~~~ 219 220Initially, only the root cgroup exists to which all processes belong. 221A child cgroup can be created by creating a sub-directory:: 222 223 # mkdir $CGROUP_NAME 224 225A given cgroup may have multiple child cgroups forming a tree 226structure. Each cgroup has a read-writable interface file 227"cgroup.procs". When read, it lists the PIDs of all processes which 228belong to the cgroup one-per-line. The PIDs are not ordered and the 229same PID may show up more than once if the process got moved to 230another cgroup and then back or the PID got recycled while reading. 231 232A process can be migrated into a cgroup by writing its PID to the 233target cgroup's "cgroup.procs" file. Only one process can be migrated 234on a single write(2) call. If a process is composed of multiple 235threads, writing the PID of any thread migrates all threads of the 236process. 237 238When a process forks a child process, the new process is born into the 239cgroup that the forking process belongs to at the time of the 240operation. After exit, a process stays associated with the cgroup 241that it belonged to at the time of exit until it's reaped; however, a 242zombie process does not appear in "cgroup.procs" and thus can't be 243moved to another cgroup. 244 245A cgroup which doesn't have any children or live processes can be 246destroyed by removing the directory. Note that a cgroup which doesn't 247have any children and is associated only with zombie processes is 248considered empty and can be removed:: 249 250 # rmdir $CGROUP_NAME 251 252"/proc/$PID/cgroup" lists a process's cgroup membership. If legacy 253cgroup is in use in the system, this file may contain multiple lines, 254one for each hierarchy. The entry for cgroup v2 is always in the 255format "0::$PATH":: 256 257 # cat /proc/842/cgroup 258 ... 259 0::/test-cgroup/test-cgroup-nested 260 261If the process becomes a zombie and the cgroup it was associated with 262is removed subsequently, " (deleted)" is appended to the path:: 263 264 # cat /proc/842/cgroup 265 ... 266 0::/test-cgroup/test-cgroup-nested (deleted) 267 268 269Threads 270~~~~~~~ 271 272cgroup v2 supports thread granularity for a subset of controllers to 273support use cases requiring hierarchical resource distribution across 274the threads of a group of processes. By default, all threads of a 275process belong to the same cgroup, which also serves as the resource 276domain to host resource consumptions which are not specific to a 277process or thread. The thread mode allows threads to be spread across 278a subtree while still maintaining the common resource domain for them. 279 280Controllers which support thread mode are called threaded controllers. 281The ones which don't are called domain controllers. 282 283Marking a cgroup threaded makes it join the resource domain of its 284parent as a threaded cgroup. The parent may be another threaded 285cgroup whose resource domain is further up in the hierarchy. The root 286of a threaded subtree, that is, the nearest ancestor which is not 287threaded, is called threaded domain or thread root interchangeably and 288serves as the resource domain for the entire subtree. 289 290Inside a threaded subtree, threads of a process can be put in 291different cgroups and are not subject to the no internal process 292constraint - threaded controllers can be enabled on non-leaf cgroups 293whether they have threads in them or not. 294 295As the threaded domain cgroup hosts all the domain resource 296consumptions of the subtree, it is considered to have internal 297resource consumptions whether there are processes in it or not and 298can't have populated child cgroups which aren't threaded. Because the 299root cgroup is not subject to no internal process constraint, it can 300serve both as a threaded domain and a parent to domain cgroups. 301 302The current operation mode or type of the cgroup is shown in the 303"cgroup.type" file which indicates whether the cgroup is a normal 304domain, a domain which is serving as the domain of a threaded subtree, 305or a threaded cgroup. 306 307On creation, a cgroup is always a domain cgroup and can be made 308threaded by writing "threaded" to the "cgroup.type" file. The 309operation is single direction:: 310 311 # echo threaded > cgroup.type 312 313Once threaded, the cgroup can't be made a domain again. To enable the 314thread mode, the following conditions must be met. 315 316- As the cgroup will join the parent's resource domain. The parent 317 must either be a valid (threaded) domain or a threaded cgroup. 318 319- When the parent is an unthreaded domain, it must not have any domain 320 controllers enabled or populated domain children. The root is 321 exempt from this requirement. 322 323Topology-wise, a cgroup can be in an invalid state. Please consider 324the following topology:: 325 326 A (threaded domain) - B (threaded) - C (domain, just created) 327 328C is created as a domain but isn't connected to a parent which can 329host child domains. C can't be used until it is turned into a 330threaded cgroup. "cgroup.type" file will report "domain (invalid)" in 331these cases. Operations which fail due to invalid topology use 332EOPNOTSUPP as the errno. 333 334A domain cgroup is turned into a threaded domain when one of its child 335cgroup becomes threaded or threaded controllers are enabled in the 336"cgroup.subtree_control" file while there are processes in the cgroup. 337A threaded domain reverts to a normal domain when the conditions 338clear. 339 340When read, "cgroup.threads" contains the list of the thread IDs of all 341threads in the cgroup. Except that the operations are per-thread 342instead of per-process, "cgroup.threads" has the same format and 343behaves the same way as "cgroup.procs". While "cgroup.threads" can be 344written to in any cgroup, as it can only move threads inside the same 345threaded domain, its operations are confined inside each threaded 346subtree. 347 348The threaded domain cgroup serves as the resource domain for the whole 349subtree, and, while the threads can be scattered across the subtree, 350all the processes are considered to be in the threaded domain cgroup. 351"cgroup.procs" in a threaded domain cgroup contains the PIDs of all 352processes in the subtree and is not readable in the subtree proper. 353However, "cgroup.procs" can be written to from anywhere in the subtree 354to migrate all threads of the matching process to the cgroup. 355 356Only threaded controllers can be enabled in a threaded subtree. When 357a threaded controller is enabled inside a threaded subtree, it only 358accounts for and controls resource consumptions associated with the 359threads in the cgroup and its descendants. All consumptions which 360aren't tied to a specific thread belong to the threaded domain cgroup. 361 362Because a threaded subtree is exempt from no internal process 363constraint, a threaded controller must be able to handle competition 364between threads in a non-leaf cgroup and its child cgroups. Each 365threaded controller defines how such competitions are handled. 366 367 368[Un]populated Notification 369-------------------------- 370 371Each non-root cgroup has a "cgroup.events" file which contains 372"populated" field indicating whether the cgroup's sub-hierarchy has 373live processes in it. Its value is 0 if there is no live process in 374the cgroup and its descendants; otherwise, 1. poll and [id]notify 375events are triggered when the value changes. This can be used, for 376example, to start a clean-up operation after all processes of a given 377sub-hierarchy have exited. The populated state updates and 378notifications are recursive. Consider the following sub-hierarchy 379where the numbers in the parentheses represent the numbers of processes 380in each cgroup:: 381 382 A(4) - B(0) - C(1) 383 \ D(0) 384 385A, B and C's "populated" fields would be 1 while D's 0. After the one 386process in C exits, B and C's "populated" fields would flip to "0" and 387file modified events will be generated on the "cgroup.events" files of 388both cgroups. 389 390 391Controlling Controllers 392----------------------- 393 394Enabling and Disabling 395~~~~~~~~~~~~~~~~~~~~~~ 396 397Each cgroup has a "cgroup.controllers" file which lists all 398controllers available for the cgroup to enable:: 399 400 # cat cgroup.controllers 401 cpu io memory 402 403No controller is enabled by default. Controllers can be enabled and 404disabled by writing to the "cgroup.subtree_control" file:: 405 406 # echo "+cpu +memory -io" > cgroup.subtree_control 407 408Only controllers which are listed in "cgroup.controllers" can be 409enabled. When multiple operations are specified as above, either they 410all succeed or fail. If multiple operations on the same controller 411are specified, the last one is effective. 412 413Enabling a controller in a cgroup indicates that the distribution of 414the target resource across its immediate children will be controlled. 415Consider the following sub-hierarchy. The enabled controllers are 416listed in parentheses:: 417 418 A(cpu,memory) - B(memory) - C() 419 \ D() 420 421As A has "cpu" and "memory" enabled, A will control the distribution 422of CPU cycles and memory to its children, in this case, B. As B has 423"memory" enabled but not "CPU", C and D will compete freely on CPU 424cycles but their division of memory available to B will be controlled. 425 426As a controller regulates the distribution of the target resource to 427the cgroup's children, enabling it creates the controller's interface 428files in the child cgroups. In the above example, enabling "cpu" on B 429would create the "cpu." prefixed controller interface files in C and 430D. Likewise, disabling "memory" from B would remove the "memory." 431prefixed controller interface files from C and D. This means that the 432controller interface files - anything which doesn't start with 433"cgroup." are owned by the parent rather than the cgroup itself. 434 435 436Top-down Constraint 437~~~~~~~~~~~~~~~~~~~ 438 439Resources are distributed top-down and a cgroup can further distribute 440a resource only if the resource has been distributed to it from the 441parent. This means that all non-root "cgroup.subtree_control" files 442can only contain controllers which are enabled in the parent's 443"cgroup.subtree_control" file. A controller can be enabled only if 444the parent has the controller enabled and a controller can't be 445disabled if one or more children have it enabled. 446 447 448No Internal Process Constraint 449~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 450 451Non-root cgroups can distribute domain resources to their children 452only when they don't have any processes of their own. In other words, 453only domain cgroups which don't contain any processes can have domain 454controllers enabled in their "cgroup.subtree_control" files. 455 456This guarantees that, when a domain controller is looking at the part 457of the hierarchy which has it enabled, processes are always only on 458the leaves. This rules out situations where child cgroups compete 459against internal processes of the parent. 460 461The root cgroup is exempt from this restriction. Root contains 462processes and anonymous resource consumption which can't be associated 463with any other cgroups and requires special treatment from most 464controllers. How resource consumption in the root cgroup is governed 465is up to each controller (for more information on this topic please 466refer to the Non-normative information section in the Controllers 467chapter). 468 469Note that the restriction doesn't get in the way if there is no 470enabled controller in the cgroup's "cgroup.subtree_control". This is 471important as otherwise it wouldn't be possible to create children of a 472populated cgroup. To control resource distribution of a cgroup, the 473cgroup must create children and transfer all its processes to the 474children before enabling controllers in its "cgroup.subtree_control" 475file. 476 477 478Delegation 479---------- 480 481Model of Delegation 482~~~~~~~~~~~~~~~~~~~ 483 484A cgroup can be delegated in two ways. First, to a less privileged 485user by granting write access of the directory and its "cgroup.procs", 486"cgroup.threads" and "cgroup.subtree_control" files to the user. 487Second, if the "nsdelegate" mount option is set, automatically to a 488cgroup namespace on namespace creation. 489 490Because the resource control interface files in a given directory 491control the distribution of the parent's resources, the delegatee 492shouldn't be allowed to write to them. For the first method, this is 493achieved by not granting access to these files. For the second, the 494kernel rejects writes to all files other than "cgroup.procs" and 495"cgroup.subtree_control" on a namespace root from inside the 496namespace. 497 498The end results are equivalent for both delegation types. Once 499delegated, the user can build sub-hierarchy under the directory, 500organize processes inside it as it sees fit and further distribute the 501resources it received from the parent. The limits and other settings 502of all resource controllers are hierarchical and regardless of what 503happens in the delegated sub-hierarchy, nothing can escape the 504resource restrictions imposed by the parent. 505 506Currently, cgroup doesn't impose any restrictions on the number of 507cgroups in or nesting depth of a delegated sub-hierarchy; however, 508this may be limited explicitly in the future. 509 510 511Delegation Containment 512~~~~~~~~~~~~~~~~~~~~~~ 513 514A delegated sub-hierarchy is contained in the sense that processes 515can't be moved into or out of the sub-hierarchy by the delegatee. 516 517For delegations to a less privileged user, this is achieved by 518requiring the following conditions for a process with a non-root euid 519to migrate a target process into a cgroup by writing its PID to the 520"cgroup.procs" file. 521 522- The writer must have write access to the "cgroup.procs" file. 523 524- The writer must have write access to the "cgroup.procs" file of the 525 common ancestor of the source and destination cgroups. 526 527The above two constraints ensure that while a delegatee may migrate 528processes around freely in the delegated sub-hierarchy it can't pull 529in from or push out to outside the sub-hierarchy. 530 531For an example, let's assume cgroups C0 and C1 have been delegated to 532user U0 who created C00, C01 under C0 and C10 under C1 as follows and 533all processes under C0 and C1 belong to U0:: 534 535 ~~~~~~~~~~~~~ - C0 - C00 536 ~ cgroup ~ \ C01 537 ~ hierarchy ~ 538 ~~~~~~~~~~~~~ - C1 - C10 539 540Let's also say U0 wants to write the PID of a process which is 541currently in C10 into "C00/cgroup.procs". U0 has write access to the 542file; however, the common ancestor of the source cgroup C10 and the 543destination cgroup C00 is above the points of delegation and U0 would 544not have write access to its "cgroup.procs" files and thus the write 545will be denied with -EACCES. 546 547For delegations to namespaces, containment is achieved by requiring 548that both the source and destination cgroups are reachable from the 549namespace of the process which is attempting the migration. If either 550is not reachable, the migration is rejected with -ENOENT. 551 552 553Guidelines 554---------- 555 556Organize Once and Control 557~~~~~~~~~~~~~~~~~~~~~~~~~ 558 559Migrating a process across cgroups is a relatively expensive operation 560and stateful resources such as memory are not moved together with the 561process. This is an explicit design decision as there often exist 562inherent trade-offs between migration and various hot paths in terms 563of synchronization cost. 564 565As such, migrating processes across cgroups frequently as a means to 566apply different resource restrictions is discouraged. A workload 567should be assigned to a cgroup according to the system's logical and 568resource structure once on start-up. Dynamic adjustments to resource 569distribution can be made by changing controller configuration through 570the interface files. 571 572 573Avoid Name Collisions 574~~~~~~~~~~~~~~~~~~~~~ 575 576Interface files for a cgroup and its children cgroups occupy the same 577directory and it is possible to create children cgroups which collide 578with interface files. 579 580All cgroup core interface files are prefixed with "cgroup." and each 581controller's interface files are prefixed with the controller name and 582a dot. A controller's name is composed of lower case alphabets and 583'_'s but never begins with an '_' so it can be used as the prefix 584character for collision avoidance. Also, interface file names won't 585start or end with terms which are often used in categorizing workloads 586such as job, service, slice, unit or workload. 587 588cgroup doesn't do anything to prevent name collisions and it's the 589user's responsibility to avoid them. 590 591 592Resource Distribution Models 593============================ 594 595cgroup controllers implement several resource distribution schemes 596depending on the resource type and expected use cases. This section 597describes major schemes in use along with their expected behaviors. 598 599 600Weights 601------- 602 603A parent's resource is distributed by adding up the weights of all 604active children and giving each the fraction matching the ratio of its 605weight against the sum. As only children which can make use of the 606resource at the moment participate in the distribution, this is 607work-conserving. Due to the dynamic nature, this model is usually 608used for stateless resources. 609 610All weights are in the range [1, 10000] with the default at 100. This 611allows symmetric multiplicative biases in both directions at fine 612enough granularity while staying in the intuitive range. 613 614As long as the weight is in range, all configuration combinations are 615valid and there is no reason to reject configuration changes or 616process migrations. 617 618"cpu.weight" proportionally distributes CPU cycles to active children 619and is an example of this type. 620 621 622Limits 623------ 624 625A child can only consume upto the configured amount of the resource. 626Limits can be over-committed - the sum of the limits of children can 627exceed the amount of resource available to the parent. 628 629Limits are in the range [0, max] and defaults to "max", which is noop. 630 631As limits can be over-committed, all configuration combinations are 632valid and there is no reason to reject configuration changes or 633process migrations. 634 635"io.max" limits the maximum BPS and/or IOPS that a cgroup can consume 636on an IO device and is an example of this type. 637 638 639Protections 640----------- 641 642A cgroup is protected upto the configured amount of the resource 643as long as the usages of all its ancestors are under their 644protected levels. Protections can be hard guarantees or best effort 645soft boundaries. Protections can also be over-committed in which case 646only upto the amount available to the parent is protected among 647children. 648 649Protections are in the range [0, max] and defaults to 0, which is 650noop. 651 652As protections can be over-committed, all configuration combinations 653are valid and there is no reason to reject configuration changes or 654process migrations. 655 656"memory.low" implements best-effort memory protection and is an 657example of this type. 658 659 660Allocations 661----------- 662 663A cgroup is exclusively allocated a certain amount of a finite 664resource. Allocations can't be over-committed - the sum of the 665allocations of children can not exceed the amount of resource 666available to the parent. 667 668Allocations are in the range [0, max] and defaults to 0, which is no 669resource. 670 671As allocations can't be over-committed, some configuration 672combinations are invalid and should be rejected. Also, if the 673resource is mandatory for execution of processes, process migrations 674may be rejected. 675 676"cpu.rt.max" hard-allocates realtime slices and is an example of this 677type. 678 679 680Interface Files 681=============== 682 683Format 684------ 685 686All interface files should be in one of the following formats whenever 687possible:: 688 689 New-line separated values 690 (when only one value can be written at once) 691 692 VAL0\n 693 VAL1\n 694 ... 695 696 Space separated values 697 (when read-only or multiple values can be written at once) 698 699 VAL0 VAL1 ...\n 700 701 Flat keyed 702 703 KEY0 VAL0\n 704 KEY1 VAL1\n 705 ... 706 707 Nested keyed 708 709 KEY0 SUB_KEY0=VAL00 SUB_KEY1=VAL01... 710 KEY1 SUB_KEY0=VAL10 SUB_KEY1=VAL11... 711 ... 712 713For a writable file, the format for writing should generally match 714reading; however, controllers may allow omitting later fields or 715implement restricted shortcuts for most common use cases. 716 717For both flat and nested keyed files, only the values for a single key 718can be written at a time. For nested keyed files, the sub key pairs 719may be specified in any order and not all pairs have to be specified. 720 721 722Conventions 723----------- 724 725- Settings for a single feature should be contained in a single file. 726 727- The root cgroup should be exempt from resource control and thus 728 shouldn't have resource control interface files. 729 730- The default time unit is microseconds. If a different unit is ever 731 used, an explicit unit suffix must be present. 732 733- A parts-per quantity should use a percentage decimal with at least 734 two digit fractional part - e.g. 13.40. 735 736- If a controller implements weight based resource distribution, its 737 interface file should be named "weight" and have the range [1, 738 10000] with 100 as the default. The values are chosen to allow 739 enough and symmetric bias in both directions while keeping it 740 intuitive (the default is 100%). 741 742- If a controller implements an absolute resource guarantee and/or 743 limit, the interface files should be named "min" and "max" 744 respectively. If a controller implements best effort resource 745 guarantee and/or limit, the interface files should be named "low" 746 and "high" respectively. 747 748 In the above four control files, the special token "max" should be 749 used to represent upward infinity for both reading and writing. 750 751- If a setting has a configurable default value and keyed specific 752 overrides, the default entry should be keyed with "default" and 753 appear as the first entry in the file. 754 755 The default value can be updated by writing either "default $VAL" or 756 "$VAL". 757 758 When writing to update a specific override, "default" can be used as 759 the value to indicate removal of the override. Override entries 760 with "default" as the value must not appear when read. 761 762 For example, a setting which is keyed by major:minor device numbers 763 with integer values may look like the following:: 764 765 # cat cgroup-example-interface-file 766 default 150 767 8:0 300 768 769 The default value can be updated by:: 770 771 # echo 125 > cgroup-example-interface-file 772 773 or:: 774 775 # echo "default 125" > cgroup-example-interface-file 776 777 An override can be set by:: 778 779 # echo "8:16 170" > cgroup-example-interface-file 780 781 and cleared by:: 782 783 # echo "8:0 default" > cgroup-example-interface-file 784 # cat cgroup-example-interface-file 785 default 125 786 8:16 170 787 788- For events which are not very high frequency, an interface file 789 "events" should be created which lists event key value pairs. 790 Whenever a notifiable event happens, file modified event should be 791 generated on the file. 792 793 794Core Interface Files 795-------------------- 796 797All cgroup core files are prefixed with "cgroup." 798 799 cgroup.type 800 A read-write single value file which exists on non-root 801 cgroups. 802 803 When read, it indicates the current type of the cgroup, which 804 can be one of the following values. 805 806 - "domain" : A normal valid domain cgroup. 807 808 - "domain threaded" : A threaded domain cgroup which is 809 serving as the root of a threaded subtree. 810 811 - "domain invalid" : A cgroup which is in an invalid state. 812 It can't be populated or have controllers enabled. It may 813 be allowed to become a threaded cgroup. 814 815 - "threaded" : A threaded cgroup which is a member of a 816 threaded subtree. 817 818 A cgroup can be turned into a threaded cgroup by writing 819 "threaded" to this file. 820 821 cgroup.procs 822 A read-write new-line separated values file which exists on 823 all cgroups. 824 825 When read, it lists the PIDs of all processes which belong to 826 the cgroup one-per-line. The PIDs are not ordered and the 827 same PID may show up more than once if the process got moved 828 to another cgroup and then back or the PID got recycled while 829 reading. 830 831 A PID can be written to migrate the process associated with 832 the PID to the cgroup. The writer should match all of the 833 following conditions. 834 835 - It must have write access to the "cgroup.procs" file. 836 837 - It must have write access to the "cgroup.procs" file of the 838 common ancestor of the source and destination cgroups. 839 840 When delegating a sub-hierarchy, write access to this file 841 should be granted along with the containing directory. 842 843 In a threaded cgroup, reading this file fails with EOPNOTSUPP 844 as all the processes belong to the thread root. Writing is 845 supported and moves every thread of the process to the cgroup. 846 847 cgroup.threads 848 A read-write new-line separated values file which exists on 849 all cgroups. 850 851 When read, it lists the TIDs of all threads which belong to 852 the cgroup one-per-line. The TIDs are not ordered and the 853 same TID may show up more than once if the thread got moved to 854 another cgroup and then back or the TID got recycled while 855 reading. 856 857 A TID can be written to migrate the thread associated with the 858 TID to the cgroup. The writer should match all of the 859 following conditions. 860 861 - It must have write access to the "cgroup.threads" file. 862 863 - The cgroup that the thread is currently in must be in the 864 same resource domain as the destination cgroup. 865 866 - It must have write access to the "cgroup.procs" file of the 867 common ancestor of the source and destination cgroups. 868 869 When delegating a sub-hierarchy, write access to this file 870 should be granted along with the containing directory. 871 872 cgroup.controllers 873 A read-only space separated values file which exists on all 874 cgroups. 875 876 It shows space separated list of all controllers available to 877 the cgroup. The controllers are not ordered. 878 879 cgroup.subtree_control 880 A read-write space separated values file which exists on all 881 cgroups. Starts out empty. 882 883 When read, it shows space separated list of the controllers 884 which are enabled to control resource distribution from the 885 cgroup to its children. 886 887 Space separated list of controllers prefixed with '+' or '-' 888 can be written to enable or disable controllers. A controller 889 name prefixed with '+' enables the controller and '-' 890 disables. If a controller appears more than once on the list, 891 the last one is effective. When multiple enable and disable 892 operations are specified, either all succeed or all fail. 893 894 cgroup.events 895 A read-only flat-keyed file which exists on non-root cgroups. 896 The following entries are defined. Unless specified 897 otherwise, a value change in this file generates a file 898 modified event. 899 900 populated 901 1 if the cgroup or its descendants contains any live 902 processes; otherwise, 0. 903 frozen 904 1 if the cgroup is frozen; otherwise, 0. 905 906 cgroup.max.descendants 907 A read-write single value files. The default is "max". 908 909 Maximum allowed number of descent cgroups. 910 If the actual number of descendants is equal or larger, 911 an attempt to create a new cgroup in the hierarchy will fail. 912 913 cgroup.max.depth 914 A read-write single value files. The default is "max". 915 916 Maximum allowed descent depth below the current cgroup. 917 If the actual descent depth is equal or larger, 918 an attempt to create a new child cgroup will fail. 919 920 cgroup.stat 921 A read-only flat-keyed file with the following entries: 922 923 nr_descendants 924 Total number of visible descendant cgroups. 925 926 nr_dying_descendants 927 Total number of dying descendant cgroups. A cgroup becomes 928 dying after being deleted by a user. The cgroup will remain 929 in dying state for some time undefined time (which can depend 930 on system load) before being completely destroyed. 931 932 A process can't enter a dying cgroup under any circumstances, 933 a dying cgroup can't revive. 934 935 A dying cgroup can consume system resources not exceeding 936 limits, which were active at the moment of cgroup deletion. 937 938 cgroup.freeze 939 A read-write single value file which exists on non-root cgroups. 940 Allowed values are "0" and "1". The default is "0". 941 942 Writing "1" to the file causes freezing of the cgroup and all 943 descendant cgroups. This means that all belonging processes will 944 be stopped and will not run until the cgroup will be explicitly 945 unfrozen. Freezing of the cgroup may take some time; when this action 946 is completed, the "frozen" value in the cgroup.events control file 947 will be updated to "1" and the corresponding notification will be 948 issued. 949 950 A cgroup can be frozen either by its own settings, or by settings 951 of any ancestor cgroups. If any of ancestor cgroups is frozen, the 952 cgroup will remain frozen. 953 954 Processes in the frozen cgroup can be killed by a fatal signal. 955 They also can enter and leave a frozen cgroup: either by an explicit 956 move by a user, or if freezing of the cgroup races with fork(). 957 If a process is moved to a frozen cgroup, it stops. If a process is 958 moved out of a frozen cgroup, it becomes running. 959 960 Frozen status of a cgroup doesn't affect any cgroup tree operations: 961 it's possible to delete a frozen (and empty) cgroup, as well as 962 create new sub-cgroups. 963 964 cgroup.kill 965 A write-only single value file which exists in non-root cgroups. 966 The only allowed value is "1". 967 968 Writing "1" to the file causes the cgroup and all descendant cgroups to 969 be killed. This means that all processes located in the affected cgroup 970 tree will be killed via SIGKILL. 971 972 Killing a cgroup tree will deal with concurrent forks appropriately and 973 is protected against migrations. 974 975 In a threaded cgroup, writing this file fails with EOPNOTSUPP as 976 killing cgroups is a process directed operation, i.e. it affects 977 the whole thread-group. 978 979 cgroup.pressure 980 A read-write single value file that allowed values are "0" and "1". 981 The default is "1". 982 983 Writing "0" to the file will disable the cgroup PSI accounting. 984 Writing "1" to the file will re-enable the cgroup PSI accounting. 985 986 This control attribute is not hierarchical, so disable or enable PSI 987 accounting in a cgroup does not affect PSI accounting in descendants 988 and doesn't need pass enablement via ancestors from root. 989 990 The reason this control attribute exists is that PSI accounts stalls for 991 each cgroup separately and aggregates it at each level of the hierarchy. 992 This may cause non-negligible overhead for some workloads when under 993 deep level of the hierarchy, in which case this control attribute can 994 be used to disable PSI accounting in the non-leaf cgroups. 995 996 irq.pressure 997 A read-write nested-keyed file. 998 999 Shows pressure stall information for IRQ/SOFTIRQ. See 1000 :ref:`Documentation/accounting/psi.rst <psi>` for details. 1001 1002Controllers 1003=========== 1004 1005.. _cgroup-v2-cpu: 1006 1007CPU 1008--- 1009 1010The "cpu" controllers regulates distribution of CPU cycles. This 1011controller implements weight and absolute bandwidth limit models for 1012normal scheduling policy and absolute bandwidth allocation model for 1013realtime scheduling policy. 1014 1015In all the above models, cycles distribution is defined only on a temporal 1016base and it does not account for the frequency at which tasks are executed. 1017The (optional) utilization clamping support allows to hint the schedutil 1018cpufreq governor about the minimum desired frequency which should always be 1019provided by a CPU, as well as the maximum desired frequency, which should not 1020be exceeded by a CPU. 1021 1022WARNING: cgroup2 doesn't yet support control of realtime processes and 1023the cpu controller can only be enabled when all RT processes are in 1024the root cgroup. Be aware that system management software may already 1025have placed RT processes into nonroot cgroups during the system boot 1026process, and these processes may need to be moved to the root cgroup 1027before the cpu controller can be enabled. 1028 1029 1030CPU Interface Files 1031~~~~~~~~~~~~~~~~~~~ 1032 1033All time durations are in microseconds. 1034 1035 cpu.stat 1036 A read-only flat-keyed file. 1037 This file exists whether the controller is enabled or not. 1038 1039 It always reports the following three stats: 1040 1041 - usage_usec 1042 - user_usec 1043 - system_usec 1044 1045 and the following three when the controller is enabled: 1046 1047 - nr_periods 1048 - nr_throttled 1049 - throttled_usec 1050 - nr_bursts 1051 - burst_usec 1052 1053 cpu.weight 1054 A read-write single value file which exists on non-root 1055 cgroups. The default is "100". 1056 1057 The weight in the range [1, 10000]. 1058 1059 cpu.weight.nice 1060 A read-write single value file which exists on non-root 1061 cgroups. The default is "0". 1062 1063 The nice value is in the range [-20, 19]. 1064 1065 This interface file is an alternative interface for 1066 "cpu.weight" and allows reading and setting weight using the 1067 same values used by nice(2). Because the range is smaller and 1068 granularity is coarser for the nice values, the read value is 1069 the closest approximation of the current weight. 1070 1071 cpu.max 1072 A read-write two value file which exists on non-root cgroups. 1073 The default is "max 100000". 1074 1075 The maximum bandwidth limit. It's in the following format:: 1076 1077 $MAX $PERIOD 1078 1079 which indicates that the group may consume upto $MAX in each 1080 $PERIOD duration. "max" for $MAX indicates no limit. If only 1081 one number is written, $MAX is updated. 1082 1083 cpu.max.burst 1084 A read-write single value file which exists on non-root 1085 cgroups. The default is "0". 1086 1087 The burst in the range [0, $MAX]. 1088 1089 cpu.pressure 1090 A read-write nested-keyed file. 1091 1092 Shows pressure stall information for CPU. See 1093 :ref:`Documentation/accounting/psi.rst <psi>` for details. 1094 1095 cpu.uclamp.min 1096 A read-write single value file which exists on non-root cgroups. 1097 The default is "0", i.e. no utilization boosting. 1098 1099 The requested minimum utilization (protection) as a percentage 1100 rational number, e.g. 12.34 for 12.34%. 1101 1102 This interface allows reading and setting minimum utilization clamp 1103 values similar to the sched_setattr(2). This minimum utilization 1104 value is used to clamp the task specific minimum utilization clamp. 1105 1106 The requested minimum utilization (protection) is always capped by 1107 the current value for the maximum utilization (limit), i.e. 1108 `cpu.uclamp.max`. 1109 1110 cpu.uclamp.max 1111 A read-write single value file which exists on non-root cgroups. 1112 The default is "max". i.e. no utilization capping 1113 1114 The requested maximum utilization (limit) as a percentage rational 1115 number, e.g. 98.76 for 98.76%. 1116 1117 This interface allows reading and setting maximum utilization clamp 1118 values similar to the sched_setattr(2). This maximum utilization 1119 value is used to clamp the task specific maximum utilization clamp. 1120 1121 1122 1123Memory 1124------ 1125 1126The "memory" controller regulates distribution of memory. Memory is 1127stateful and implements both limit and protection models. Due to the 1128intertwining between memory usage and reclaim pressure and the 1129stateful nature of memory, the distribution model is relatively 1130complex. 1131 1132While not completely water-tight, all major memory usages by a given 1133cgroup are tracked so that the total memory consumption can be 1134accounted and controlled to a reasonable extent. Currently, the 1135following types of memory usages are tracked. 1136 1137- Userland memory - page cache and anonymous memory. 1138 1139- Kernel data structures such as dentries and inodes. 1140 1141- TCP socket buffers. 1142 1143The above list may expand in the future for better coverage. 1144 1145 1146Memory Interface Files 1147~~~~~~~~~~~~~~~~~~~~~~ 1148 1149All memory amounts are in bytes. If a value which is not aligned to 1150PAGE_SIZE is written, the value may be rounded up to the closest 1151PAGE_SIZE multiple when read back. 1152 1153 memory.current 1154 A read-only single value file which exists on non-root 1155 cgroups. 1156 1157 The total amount of memory currently being used by the cgroup 1158 and its descendants. 1159 1160 memory.min 1161 A read-write single value file which exists on non-root 1162 cgroups. The default is "0". 1163 1164 Hard memory protection. If the memory usage of a cgroup 1165 is within its effective min boundary, the cgroup's memory 1166 won't be reclaimed under any conditions. If there is no 1167 unprotected reclaimable memory available, OOM killer 1168 is invoked. Above the effective min boundary (or 1169 effective low boundary if it is higher), pages are reclaimed 1170 proportionally to the overage, reducing reclaim pressure for 1171 smaller overages. 1172 1173 Effective min boundary is limited by memory.min values of 1174 all ancestor cgroups. If there is memory.min overcommitment 1175 (child cgroup or cgroups are requiring more protected memory 1176 than parent will allow), then each child cgroup will get 1177 the part of parent's protection proportional to its 1178 actual memory usage below memory.min. 1179 1180 Putting more memory than generally available under this 1181 protection is discouraged and may lead to constant OOMs. 1182 1183 If a memory cgroup is not populated with processes, 1184 its memory.min is ignored. 1185 1186 memory.low 1187 A read-write single value file which exists on non-root 1188 cgroups. The default is "0". 1189 1190 Best-effort memory protection. If the memory usage of a 1191 cgroup is within its effective low boundary, the cgroup's 1192 memory won't be reclaimed unless there is no reclaimable 1193 memory available in unprotected cgroups. 1194 Above the effective low boundary (or 1195 effective min boundary if it is higher), pages are reclaimed 1196 proportionally to the overage, reducing reclaim pressure for 1197 smaller overages. 1198 1199 Effective low boundary is limited by memory.low values of 1200 all ancestor cgroups. If there is memory.low overcommitment 1201 (child cgroup or cgroups are requiring more protected memory 1202 than parent will allow), then each child cgroup will get 1203 the part of parent's protection proportional to its 1204 actual memory usage below memory.low. 1205 1206 Putting more memory than generally available under this 1207 protection is discouraged. 1208 1209 memory.high 1210 A read-write single value file which exists on non-root 1211 cgroups. The default is "max". 1212 1213 Memory usage throttle limit. This is the main mechanism to 1214 control memory usage of a cgroup. If a cgroup's usage goes 1215 over the high boundary, the processes of the cgroup are 1216 throttled and put under heavy reclaim pressure. 1217 1218 Going over the high limit never invokes the OOM killer and 1219 under extreme conditions the limit may be breached. 1220 1221 memory.max 1222 A read-write single value file which exists on non-root 1223 cgroups. The default is "max". 1224 1225 Memory usage hard limit. This is the final protection 1226 mechanism. If a cgroup's memory usage reaches this limit and 1227 can't be reduced, the OOM killer is invoked in the cgroup. 1228 Under certain circumstances, the usage may go over the limit 1229 temporarily. 1230 1231 In default configuration regular 0-order allocations always 1232 succeed unless OOM killer chooses current task as a victim. 1233 1234 Some kinds of allocations don't invoke the OOM killer. 1235 Caller could retry them differently, return into userspace 1236 as -ENOMEM or silently ignore in cases like disk readahead. 1237 1238 This is the ultimate protection mechanism. As long as the 1239 high limit is used and monitored properly, this limit's 1240 utility is limited to providing the final safety net. 1241 1242 memory.reclaim 1243 A write-only nested-keyed file which exists for all cgroups. 1244 1245 This is a simple interface to trigger memory reclaim in the 1246 target cgroup. 1247 1248 This file accepts a string which contains the number of bytes to 1249 reclaim. 1250 1251 Example:: 1252 1253 echo "1G" > memory.reclaim 1254 1255 Please note that the kernel can over or under reclaim from 1256 the target cgroup. If less bytes are reclaimed than the 1257 specified amount, -EAGAIN is returned. 1258 1259 Please note that the proactive reclaim (triggered by this 1260 interface) is not meant to indicate memory pressure on the 1261 memory cgroup. Therefore socket memory balancing triggered by 1262 the memory reclaim normally is not exercised in this case. 1263 This means that the networking layer will not adapt based on 1264 reclaim induced by memory.reclaim. 1265 1266 This file also allows the user to specify the nodes to reclaim from, 1267 via the 'nodes=' key, for example:: 1268 1269 echo "1G nodes=0,1" > memory.reclaim 1270 1271 The above instructs the kernel to reclaim memory from nodes 0,1. 1272 1273 memory.peak 1274 A read-only single value file which exists on non-root 1275 cgroups. 1276 1277 The max memory usage recorded for the cgroup and its 1278 descendants since the creation of the cgroup. 1279 1280 memory.oom.group 1281 A read-write single value file which exists on non-root 1282 cgroups. The default value is "0". 1283 1284 Determines whether the cgroup should be treated as 1285 an indivisible workload by the OOM killer. If set, 1286 all tasks belonging to the cgroup or to its descendants 1287 (if the memory cgroup is not a leaf cgroup) are killed 1288 together or not at all. This can be used to avoid 1289 partial kills to guarantee workload integrity. 1290 1291 Tasks with the OOM protection (oom_score_adj set to -1000) 1292 are treated as an exception and are never killed. 1293 1294 If the OOM killer is invoked in a cgroup, it's not going 1295 to kill any tasks outside of this cgroup, regardless 1296 memory.oom.group values of ancestor cgroups. 1297 1298 memory.events 1299 A read-only flat-keyed file which exists on non-root cgroups. 1300 The following entries are defined. Unless specified 1301 otherwise, a value change in this file generates a file 1302 modified event. 1303 1304 Note that all fields in this file are hierarchical and the 1305 file modified event can be generated due to an event down the 1306 hierarchy. For the local events at the cgroup level see 1307 memory.events.local. 1308 1309 low 1310 The number of times the cgroup is reclaimed due to 1311 high memory pressure even though its usage is under 1312 the low boundary. This usually indicates that the low 1313 boundary is over-committed. 1314 1315 high 1316 The number of times processes of the cgroup are 1317 throttled and routed to perform direct memory reclaim 1318 because the high memory boundary was exceeded. For a 1319 cgroup whose memory usage is capped by the high limit 1320 rather than global memory pressure, this event's 1321 occurrences are expected. 1322 1323 max 1324 The number of times the cgroup's memory usage was 1325 about to go over the max boundary. If direct reclaim 1326 fails to bring it down, the cgroup goes to OOM state. 1327 1328 oom 1329 The number of time the cgroup's memory usage was 1330 reached the limit and allocation was about to fail. 1331 1332 This event is not raised if the OOM killer is not 1333 considered as an option, e.g. for failed high-order 1334 allocations or if caller asked to not retry attempts. 1335 1336 oom_kill 1337 The number of processes belonging to this cgroup 1338 killed by any kind of OOM killer. 1339 1340 oom_group_kill 1341 The number of times a group OOM has occurred. 1342 1343 memory.events.local 1344 Similar to memory.events but the fields in the file are local 1345 to the cgroup i.e. not hierarchical. The file modified event 1346 generated on this file reflects only the local events. 1347 1348 memory.stat 1349 A read-only flat-keyed file which exists on non-root cgroups. 1350 1351 This breaks down the cgroup's memory footprint into different 1352 types of memory, type-specific details, and other information 1353 on the state and past events of the memory management system. 1354 1355 All memory amounts are in bytes. 1356 1357 The entries are ordered to be human readable, and new entries 1358 can show up in the middle. Don't rely on items remaining in a 1359 fixed position; use the keys to look up specific values! 1360 1361 If the entry has no per-node counter (or not show in the 1362 memory.numa_stat). We use 'npn' (non-per-node) as the tag 1363 to indicate that it will not show in the memory.numa_stat. 1364 1365 anon 1366 Amount of memory used in anonymous mappings such as 1367 brk(), sbrk(), and mmap(MAP_ANONYMOUS) 1368 1369 file 1370 Amount of memory used to cache filesystem data, 1371 including tmpfs and shared memory. 1372 1373 kernel (npn) 1374 Amount of total kernel memory, including 1375 (kernel_stack, pagetables, percpu, vmalloc, slab) in 1376 addition to other kernel memory use cases. 1377 1378 kernel_stack 1379 Amount of memory allocated to kernel stacks. 1380 1381 pagetables 1382 Amount of memory allocated for page tables. 1383 1384 sec_pagetables 1385 Amount of memory allocated for secondary page tables, 1386 this currently includes KVM mmu allocations on x86 1387 and arm64. 1388 1389 percpu (npn) 1390 Amount of memory used for storing per-cpu kernel 1391 data structures. 1392 1393 sock (npn) 1394 Amount of memory used in network transmission buffers 1395 1396 vmalloc (npn) 1397 Amount of memory used for vmap backed memory. 1398 1399 shmem 1400 Amount of cached filesystem data that is swap-backed, 1401 such as tmpfs, shm segments, shared anonymous mmap()s 1402 1403 zswap 1404 Amount of memory consumed by the zswap compression backend. 1405 1406 zswapped 1407 Amount of application memory swapped out to zswap. 1408 1409 file_mapped 1410 Amount of cached filesystem data mapped with mmap() 1411 1412 file_dirty 1413 Amount of cached filesystem data that was modified but 1414 not yet written back to disk 1415 1416 file_writeback 1417 Amount of cached filesystem data that was modified and 1418 is currently being written back to disk 1419 1420 swapcached 1421 Amount of swap cached in memory. The swapcache is accounted 1422 against both memory and swap usage. 1423 1424 anon_thp 1425 Amount of memory used in anonymous mappings backed by 1426 transparent hugepages 1427 1428 file_thp 1429 Amount of cached filesystem data backed by transparent 1430 hugepages 1431 1432 shmem_thp 1433 Amount of shm, tmpfs, shared anonymous mmap()s backed by 1434 transparent hugepages 1435 1436 inactive_anon, active_anon, inactive_file, active_file, unevictable 1437 Amount of memory, swap-backed and filesystem-backed, 1438 on the internal memory management lists used by the 1439 page reclaim algorithm. 1440 1441 As these represent internal list state (eg. shmem pages are on anon 1442 memory management lists), inactive_foo + active_foo may not be equal to 1443 the value for the foo counter, since the foo counter is type-based, not 1444 list-based. 1445 1446 slab_reclaimable 1447 Part of "slab" that might be reclaimed, such as 1448 dentries and inodes. 1449 1450 slab_unreclaimable 1451 Part of "slab" that cannot be reclaimed on memory 1452 pressure. 1453 1454 slab (npn) 1455 Amount of memory used for storing in-kernel data 1456 structures. 1457 1458 workingset_refault_anon 1459 Number of refaults of previously evicted anonymous pages. 1460 1461 workingset_refault_file 1462 Number of refaults of previously evicted file pages. 1463 1464 workingset_activate_anon 1465 Number of refaulted anonymous pages that were immediately 1466 activated. 1467 1468 workingset_activate_file 1469 Number of refaulted file pages that were immediately activated. 1470 1471 workingset_restore_anon 1472 Number of restored anonymous pages which have been detected as 1473 an active workingset before they got reclaimed. 1474 1475 workingset_restore_file 1476 Number of restored file pages which have been detected as an 1477 active workingset before they got reclaimed. 1478 1479 workingset_nodereclaim 1480 Number of times a shadow node has been reclaimed 1481 1482 pgscan (npn) 1483 Amount of scanned pages (in an inactive LRU list) 1484 1485 pgsteal (npn) 1486 Amount of reclaimed pages 1487 1488 pgscan_kswapd (npn) 1489 Amount of scanned pages by kswapd (in an inactive LRU list) 1490 1491 pgscan_direct (npn) 1492 Amount of scanned pages directly (in an inactive LRU list) 1493 1494 pgscan_khugepaged (npn) 1495 Amount of scanned pages by khugepaged (in an inactive LRU list) 1496 1497 pgsteal_kswapd (npn) 1498 Amount of reclaimed pages by kswapd 1499 1500 pgsteal_direct (npn) 1501 Amount of reclaimed pages directly 1502 1503 pgsteal_khugepaged (npn) 1504 Amount of reclaimed pages by khugepaged 1505 1506 pgfault (npn) 1507 Total number of page faults incurred 1508 1509 pgmajfault (npn) 1510 Number of major page faults incurred 1511 1512 pgrefill (npn) 1513 Amount of scanned pages (in an active LRU list) 1514 1515 pgactivate (npn) 1516 Amount of pages moved to the active LRU list 1517 1518 pgdeactivate (npn) 1519 Amount of pages moved to the inactive LRU list 1520 1521 pglazyfree (npn) 1522 Amount of pages postponed to be freed under memory pressure 1523 1524 pglazyfreed (npn) 1525 Amount of reclaimed lazyfree pages 1526 1527 thp_fault_alloc (npn) 1528 Number of transparent hugepages which were allocated to satisfy 1529 a page fault. This counter is not present when CONFIG_TRANSPARENT_HUGEPAGE 1530 is not set. 1531 1532 thp_collapse_alloc (npn) 1533 Number of transparent hugepages which were allocated to allow 1534 collapsing an existing range of pages. This counter is not 1535 present when CONFIG_TRANSPARENT_HUGEPAGE is not set. 1536 1537 memory.numa_stat 1538 A read-only nested-keyed file which exists on non-root cgroups. 1539 1540 This breaks down the cgroup's memory footprint into different 1541 types of memory, type-specific details, and other information 1542 per node on the state of the memory management system. 1543 1544 This is useful for providing visibility into the NUMA locality 1545 information within an memcg since the pages are allowed to be 1546 allocated from any physical node. One of the use case is evaluating 1547 application performance by combining this information with the 1548 application's CPU allocation. 1549 1550 All memory amounts are in bytes. 1551 1552 The output format of memory.numa_stat is:: 1553 1554 type N0=<bytes in node 0> N1=<bytes in node 1> ... 1555 1556 The entries are ordered to be human readable, and new entries 1557 can show up in the middle. Don't rely on items remaining in a 1558 fixed position; use the keys to look up specific values! 1559 1560 The entries can refer to the memory.stat. 1561 1562 memory.swap.current 1563 A read-only single value file which exists on non-root 1564 cgroups. 1565 1566 The total amount of swap currently being used by the cgroup 1567 and its descendants. 1568 1569 memory.swap.high 1570 A read-write single value file which exists on non-root 1571 cgroups. The default is "max". 1572 1573 Swap usage throttle limit. If a cgroup's swap usage exceeds 1574 this limit, all its further allocations will be throttled to 1575 allow userspace to implement custom out-of-memory procedures. 1576 1577 This limit marks a point of no return for the cgroup. It is NOT 1578 designed to manage the amount of swapping a workload does 1579 during regular operation. Compare to memory.swap.max, which 1580 prohibits swapping past a set amount, but lets the cgroup 1581 continue unimpeded as long as other memory can be reclaimed. 1582 1583 Healthy workloads are not expected to reach this limit. 1584 1585 memory.swap.max 1586 A read-write single value file which exists on non-root 1587 cgroups. The default is "max". 1588 1589 Swap usage hard limit. If a cgroup's swap usage reaches this 1590 limit, anonymous memory of the cgroup will not be swapped out. 1591 1592 memory.swap.events 1593 A read-only flat-keyed file which exists on non-root cgroups. 1594 The following entries are defined. Unless specified 1595 otherwise, a value change in this file generates a file 1596 modified event. 1597 1598 high 1599 The number of times the cgroup's swap usage was over 1600 the high threshold. 1601 1602 max 1603 The number of times the cgroup's swap usage was about 1604 to go over the max boundary and swap allocation 1605 failed. 1606 1607 fail 1608 The number of times swap allocation failed either 1609 because of running out of swap system-wide or max 1610 limit. 1611 1612 When reduced under the current usage, the existing swap 1613 entries are reclaimed gradually and the swap usage may stay 1614 higher than the limit for an extended period of time. This 1615 reduces the impact on the workload and memory management. 1616 1617 memory.zswap.current 1618 A read-only single value file which exists on non-root 1619 cgroups. 1620 1621 The total amount of memory consumed by the zswap compression 1622 backend. 1623 1624 memory.zswap.max 1625 A read-write single value file which exists on non-root 1626 cgroups. The default is "max". 1627 1628 Zswap usage hard limit. If a cgroup's zswap pool reaches this 1629 limit, it will refuse to take any more stores before existing 1630 entries fault back in or are written out to disk. 1631 1632 memory.pressure 1633 A read-only nested-keyed file. 1634 1635 Shows pressure stall information for memory. See 1636 :ref:`Documentation/accounting/psi.rst <psi>` for details. 1637 1638 1639Usage Guidelines 1640~~~~~~~~~~~~~~~~ 1641 1642"memory.high" is the main mechanism to control memory usage. 1643Over-committing on high limit (sum of high limits > available memory) 1644and letting global memory pressure to distribute memory according to 1645usage is a viable strategy. 1646 1647Because breach of the high limit doesn't trigger the OOM killer but 1648throttles the offending cgroup, a management agent has ample 1649opportunities to monitor and take appropriate actions such as granting 1650more memory or terminating the workload. 1651 1652Determining whether a cgroup has enough memory is not trivial as 1653memory usage doesn't indicate whether the workload can benefit from 1654more memory. For example, a workload which writes data received from 1655network to a file can use all available memory but can also operate as 1656performant with a small amount of memory. A measure of memory 1657pressure - how much the workload is being impacted due to lack of 1658memory - is necessary to determine whether a workload needs more 1659memory; unfortunately, memory pressure monitoring mechanism isn't 1660implemented yet. 1661 1662 1663Memory Ownership 1664~~~~~~~~~~~~~~~~ 1665 1666A memory area is charged to the cgroup which instantiated it and stays 1667charged to the cgroup until the area is released. Migrating a process 1668to a different cgroup doesn't move the memory usages that it 1669instantiated while in the previous cgroup to the new cgroup. 1670 1671A memory area may be used by processes belonging to different cgroups. 1672To which cgroup the area will be charged is in-deterministic; however, 1673over time, the memory area is likely to end up in a cgroup which has 1674enough memory allowance to avoid high reclaim pressure. 1675 1676If a cgroup sweeps a considerable amount of memory which is expected 1677to be accessed repeatedly by other cgroups, it may make sense to use 1678POSIX_FADV_DONTNEED to relinquish the ownership of memory areas 1679belonging to the affected files to ensure correct memory ownership. 1680 1681 1682IO 1683-- 1684 1685The "io" controller regulates the distribution of IO resources. This 1686controller implements both weight based and absolute bandwidth or IOPS 1687limit distribution; however, weight based distribution is available 1688only if cfq-iosched is in use and neither scheme is available for 1689blk-mq devices. 1690 1691 1692IO Interface Files 1693~~~~~~~~~~~~~~~~~~ 1694 1695 io.stat 1696 A read-only nested-keyed file. 1697 1698 Lines are keyed by $MAJ:$MIN device numbers and not ordered. 1699 The following nested keys are defined. 1700 1701 ====== ===================== 1702 rbytes Bytes read 1703 wbytes Bytes written 1704 rios Number of read IOs 1705 wios Number of write IOs 1706 dbytes Bytes discarded 1707 dios Number of discard IOs 1708 ====== ===================== 1709 1710 An example read output follows:: 1711 1712 8:16 rbytes=1459200 wbytes=314773504 rios=192 wios=353 dbytes=0 dios=0 1713 8:0 rbytes=90430464 wbytes=299008000 rios=8950 wios=1252 dbytes=50331648 dios=3021 1714 1715 io.cost.qos 1716 A read-write nested-keyed file which exists only on the root 1717 cgroup. 1718 1719 This file configures the Quality of Service of the IO cost 1720 model based controller (CONFIG_BLK_CGROUP_IOCOST) which 1721 currently implements "io.weight" proportional control. Lines 1722 are keyed by $MAJ:$MIN device numbers and not ordered. The 1723 line for a given device is populated on the first write for 1724 the device on "io.cost.qos" or "io.cost.model". The following 1725 nested keys are defined. 1726 1727 ====== ===================================== 1728 enable Weight-based control enable 1729 ctrl "auto" or "user" 1730 rpct Read latency percentile [0, 100] 1731 rlat Read latency threshold 1732 wpct Write latency percentile [0, 100] 1733 wlat Write latency threshold 1734 min Minimum scaling percentage [1, 10000] 1735 max Maximum scaling percentage [1, 10000] 1736 ====== ===================================== 1737 1738 The controller is disabled by default and can be enabled by 1739 setting "enable" to 1. "rpct" and "wpct" parameters default 1740 to zero and the controller uses internal device saturation 1741 state to adjust the overall IO rate between "min" and "max". 1742 1743 When a better control quality is needed, latency QoS 1744 parameters can be configured. For example:: 1745 1746 8:16 enable=1 ctrl=auto rpct=95.00 rlat=75000 wpct=95.00 wlat=150000 min=50.00 max=150.0 1747 1748 shows that on sdb, the controller is enabled, will consider 1749 the device saturated if the 95th percentile of read completion 1750 latencies is above 75ms or write 150ms, and adjust the overall 1751 IO issue rate between 50% and 150% accordingly. 1752 1753 The lower the saturation point, the better the latency QoS at 1754 the cost of aggregate bandwidth. The narrower the allowed 1755 adjustment range between "min" and "max", the more conformant 1756 to the cost model the IO behavior. Note that the IO issue 1757 base rate may be far off from 100% and setting "min" and "max" 1758 blindly can lead to a significant loss of device capacity or 1759 control quality. "min" and "max" are useful for regulating 1760 devices which show wide temporary behavior changes - e.g. a 1761 ssd which accepts writes at the line speed for a while and 1762 then completely stalls for multiple seconds. 1763 1764 When "ctrl" is "auto", the parameters are controlled by the 1765 kernel and may change automatically. Setting "ctrl" to "user" 1766 or setting any of the percentile and latency parameters puts 1767 it into "user" mode and disables the automatic changes. The 1768 automatic mode can be restored by setting "ctrl" to "auto". 1769 1770 io.cost.model 1771 A read-write nested-keyed file which exists only on the root 1772 cgroup. 1773 1774 This file configures the cost model of the IO cost model based 1775 controller (CONFIG_BLK_CGROUP_IOCOST) which currently 1776 implements "io.weight" proportional control. Lines are keyed 1777 by $MAJ:$MIN device numbers and not ordered. The line for a 1778 given device is populated on the first write for the device on 1779 "io.cost.qos" or "io.cost.model". The following nested keys 1780 are defined. 1781 1782 ===== ================================ 1783 ctrl "auto" or "user" 1784 model The cost model in use - "linear" 1785 ===== ================================ 1786 1787 When "ctrl" is "auto", the kernel may change all parameters 1788 dynamically. When "ctrl" is set to "user" or any other 1789 parameters are written to, "ctrl" become "user" and the 1790 automatic changes are disabled. 1791 1792 When "model" is "linear", the following model parameters are 1793 defined. 1794 1795 ============= ======================================== 1796 [r|w]bps The maximum sequential IO throughput 1797 [r|w]seqiops The maximum 4k sequential IOs per second 1798 [r|w]randiops The maximum 4k random IOs per second 1799 ============= ======================================== 1800 1801 From the above, the builtin linear model determines the base 1802 costs of a sequential and random IO and the cost coefficient 1803 for the IO size. While simple, this model can cover most 1804 common device classes acceptably. 1805 1806 The IO cost model isn't expected to be accurate in absolute 1807 sense and is scaled to the device behavior dynamically. 1808 1809 If needed, tools/cgroup/iocost_coef_gen.py can be used to 1810 generate device-specific coefficients. 1811 1812 io.weight 1813 A read-write flat-keyed file which exists on non-root cgroups. 1814 The default is "default 100". 1815 1816 The first line is the default weight applied to devices 1817 without specific override. The rest are overrides keyed by 1818 $MAJ:$MIN device numbers and not ordered. The weights are in 1819 the range [1, 10000] and specifies the relative amount IO time 1820 the cgroup can use in relation to its siblings. 1821 1822 The default weight can be updated by writing either "default 1823 $WEIGHT" or simply "$WEIGHT". Overrides can be set by writing 1824 "$MAJ:$MIN $WEIGHT" and unset by writing "$MAJ:$MIN default". 1825 1826 An example read output follows:: 1827 1828 default 100 1829 8:16 200 1830 8:0 50 1831 1832 io.max 1833 A read-write nested-keyed file which exists on non-root 1834 cgroups. 1835 1836 BPS and IOPS based IO limit. Lines are keyed by $MAJ:$MIN 1837 device numbers and not ordered. The following nested keys are 1838 defined. 1839 1840 ===== ================================== 1841 rbps Max read bytes per second 1842 wbps Max write bytes per second 1843 riops Max read IO operations per second 1844 wiops Max write IO operations per second 1845 ===== ================================== 1846 1847 When writing, any number of nested key-value pairs can be 1848 specified in any order. "max" can be specified as the value 1849 to remove a specific limit. If the same key is specified 1850 multiple times, the outcome is undefined. 1851 1852 BPS and IOPS are measured in each IO direction and IOs are 1853 delayed if limit is reached. Temporary bursts are allowed. 1854 1855 Setting read limit at 2M BPS and write at 120 IOPS for 8:16:: 1856 1857 echo "8:16 rbps=2097152 wiops=120" > io.max 1858 1859 Reading returns the following:: 1860 1861 8:16 rbps=2097152 wbps=max riops=max wiops=120 1862 1863 Write IOPS limit can be removed by writing the following:: 1864 1865 echo "8:16 wiops=max" > io.max 1866 1867 Reading now returns the following:: 1868 1869 8:16 rbps=2097152 wbps=max riops=max wiops=max 1870 1871 io.pressure 1872 A read-only nested-keyed file. 1873 1874 Shows pressure stall information for IO. See 1875 :ref:`Documentation/accounting/psi.rst <psi>` for details. 1876 1877 1878Writeback 1879~~~~~~~~~ 1880 1881Page cache is dirtied through buffered writes and shared mmaps and 1882written asynchronously to the backing filesystem by the writeback 1883mechanism. Writeback sits between the memory and IO domains and 1884regulates the proportion of dirty memory by balancing dirtying and 1885write IOs. 1886 1887The io controller, in conjunction with the memory controller, 1888implements control of page cache writeback IOs. The memory controller 1889defines the memory domain that dirty memory ratio is calculated and 1890maintained for and the io controller defines the io domain which 1891writes out dirty pages for the memory domain. Both system-wide and 1892per-cgroup dirty memory states are examined and the more restrictive 1893of the two is enforced. 1894 1895cgroup writeback requires explicit support from the underlying 1896filesystem. Currently, cgroup writeback is implemented on ext2, ext4, 1897btrfs, f2fs, and xfs. On other filesystems, all writeback IOs are 1898attributed to the root cgroup. 1899 1900There are inherent differences in memory and writeback management 1901which affects how cgroup ownership is tracked. Memory is tracked per 1902page while writeback per inode. For the purpose of writeback, an 1903inode is assigned to a cgroup and all IO requests to write dirty pages 1904from the inode are attributed to that cgroup. 1905 1906As cgroup ownership for memory is tracked per page, there can be pages 1907which are associated with different cgroups than the one the inode is 1908associated with. These are called foreign pages. The writeback 1909constantly keeps track of foreign pages and, if a particular foreign 1910cgroup becomes the majority over a certain period of time, switches 1911the ownership of the inode to that cgroup. 1912 1913While this model is enough for most use cases where a given inode is 1914mostly dirtied by a single cgroup even when the main writing cgroup 1915changes over time, use cases where multiple cgroups write to a single 1916inode simultaneously are not supported well. In such circumstances, a 1917significant portion of IOs are likely to be attributed incorrectly. 1918As memory controller assigns page ownership on the first use and 1919doesn't update it until the page is released, even if writeback 1920strictly follows page ownership, multiple cgroups dirtying overlapping 1921areas wouldn't work as expected. It's recommended to avoid such usage 1922patterns. 1923 1924The sysctl knobs which affect writeback behavior are applied to cgroup 1925writeback as follows. 1926 1927 vm.dirty_background_ratio, vm.dirty_ratio 1928 These ratios apply the same to cgroup writeback with the 1929 amount of available memory capped by limits imposed by the 1930 memory controller and system-wide clean memory. 1931 1932 vm.dirty_background_bytes, vm.dirty_bytes 1933 For cgroup writeback, this is calculated into ratio against 1934 total available memory and applied the same way as 1935 vm.dirty[_background]_ratio. 1936 1937 1938IO Latency 1939~~~~~~~~~~ 1940 1941This is a cgroup v2 controller for IO workload protection. You provide a group 1942with a latency target, and if the average latency exceeds that target the 1943controller will throttle any peers that have a lower latency target than the 1944protected workload. 1945 1946The limits are only applied at the peer level in the hierarchy. This means that 1947in the diagram below, only groups A, B, and C will influence each other, and 1948groups D and F will influence each other. Group G will influence nobody:: 1949 1950 [root] 1951 / | \ 1952 A B C 1953 / \ | 1954 D F G 1955 1956 1957So the ideal way to configure this is to set io.latency in groups A, B, and C. 1958Generally you do not want to set a value lower than the latency your device 1959supports. Experiment to find the value that works best for your workload. 1960Start at higher than the expected latency for your device and watch the 1961avg_lat value in io.stat for your workload group to get an idea of the 1962latency you see during normal operation. Use the avg_lat value as a basis for 1963your real setting, setting at 10-15% higher than the value in io.stat. 1964 1965How IO Latency Throttling Works 1966~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 1967 1968io.latency is work conserving; so as long as everybody is meeting their latency 1969target the controller doesn't do anything. Once a group starts missing its 1970target it begins throttling any peer group that has a higher target than itself. 1971This throttling takes 2 forms: 1972 1973- Queue depth throttling. This is the number of outstanding IO's a group is 1974 allowed to have. We will clamp down relatively quickly, starting at no limit 1975 and going all the way down to 1 IO at a time. 1976 1977- Artificial delay induction. There are certain types of IO that cannot be 1978 throttled without possibly adversely affecting higher priority groups. This 1979 includes swapping and metadata IO. These types of IO are allowed to occur 1980 normally, however they are "charged" to the originating group. If the 1981 originating group is being throttled you will see the use_delay and delay 1982 fields in io.stat increase. The delay value is how many microseconds that are 1983 being added to any process that runs in this group. Because this number can 1984 grow quite large if there is a lot of swapping or metadata IO occurring we 1985 limit the individual delay events to 1 second at a time. 1986 1987Once the victimized group starts meeting its latency target again it will start 1988unthrottling any peer groups that were throttled previously. If the victimized 1989group simply stops doing IO the global counter will unthrottle appropriately. 1990 1991IO Latency Interface Files 1992~~~~~~~~~~~~~~~~~~~~~~~~~~ 1993 1994 io.latency 1995 This takes a similar format as the other controllers. 1996 1997 "MAJOR:MINOR target=<target time in microseconds>" 1998 1999 io.stat 2000 If the controller is enabled you will see extra stats in io.stat in 2001 addition to the normal ones. 2002 2003 depth 2004 This is the current queue depth for the group. 2005 2006 avg_lat 2007 This is an exponential moving average with a decay rate of 1/exp 2008 bound by the sampling interval. The decay rate interval can be 2009 calculated by multiplying the win value in io.stat by the 2010 corresponding number of samples based on the win value. 2011 2012 win 2013 The sampling window size in milliseconds. This is the minimum 2014 duration of time between evaluation events. Windows only elapse 2015 with IO activity. Idle periods extend the most recent window. 2016 2017IO Priority 2018~~~~~~~~~~~ 2019 2020A single attribute controls the behavior of the I/O priority cgroup policy, 2021namely the blkio.prio.class attribute. The following values are accepted for 2022that attribute: 2023 2024 no-change 2025 Do not modify the I/O priority class. 2026 2027 none-to-rt 2028 For requests that do not have an I/O priority class (NONE), 2029 change the I/O priority class into RT. Do not modify 2030 the I/O priority class of other requests. 2031 2032 restrict-to-be 2033 For requests that do not have an I/O priority class or that have I/O 2034 priority class RT, change it into BE. Do not modify the I/O priority 2035 class of requests that have priority class IDLE. 2036 2037 idle 2038 Change the I/O priority class of all requests into IDLE, the lowest 2039 I/O priority class. 2040 2041The following numerical values are associated with the I/O priority policies: 2042 2043+-------------+---+ 2044| no-change | 0 | 2045+-------------+---+ 2046| none-to-rt | 1 | 2047+-------------+---+ 2048| rt-to-be | 2 | 2049+-------------+---+ 2050| all-to-idle | 3 | 2051+-------------+---+ 2052 2053The numerical value that corresponds to each I/O priority class is as follows: 2054 2055+-------------------------------+---+ 2056| IOPRIO_CLASS_NONE | 0 | 2057+-------------------------------+---+ 2058| IOPRIO_CLASS_RT (real-time) | 1 | 2059+-------------------------------+---+ 2060| IOPRIO_CLASS_BE (best effort) | 2 | 2061+-------------------------------+---+ 2062| IOPRIO_CLASS_IDLE | 3 | 2063+-------------------------------+---+ 2064 2065The algorithm to set the I/O priority class for a request is as follows: 2066 2067- Translate the I/O priority class policy into a number. 2068- Change the request I/O priority class into the maximum of the I/O priority 2069 class policy number and the numerical I/O priority class. 2070 2071PID 2072--- 2073 2074The process number controller is used to allow a cgroup to stop any 2075new tasks from being fork()'d or clone()'d after a specified limit is 2076reached. 2077 2078The number of tasks in a cgroup can be exhausted in ways which other 2079controllers cannot prevent, thus warranting its own controller. For 2080example, a fork bomb is likely to exhaust the number of tasks before 2081hitting memory restrictions. 2082 2083Note that PIDs used in this controller refer to TIDs, process IDs as 2084used by the kernel. 2085 2086 2087PID Interface Files 2088~~~~~~~~~~~~~~~~~~~ 2089 2090 pids.max 2091 A read-write single value file which exists on non-root 2092 cgroups. The default is "max". 2093 2094 Hard limit of number of processes. 2095 2096 pids.current 2097 A read-only single value file which exists on all cgroups. 2098 2099 The number of processes currently in the cgroup and its 2100 descendants. 2101 2102Organisational operations are not blocked by cgroup policies, so it is 2103possible to have pids.current > pids.max. This can be done by either 2104setting the limit to be smaller than pids.current, or attaching enough 2105processes to the cgroup such that pids.current is larger than 2106pids.max. However, it is not possible to violate a cgroup PID policy 2107through fork() or clone(). These will return -EAGAIN if the creation 2108of a new process would cause a cgroup policy to be violated. 2109 2110 2111Cpuset 2112------ 2113 2114The "cpuset" controller provides a mechanism for constraining 2115the CPU and memory node placement of tasks to only the resources 2116specified in the cpuset interface files in a task's current cgroup. 2117This is especially valuable on large NUMA systems where placing jobs 2118on properly sized subsets of the systems with careful processor and 2119memory placement to reduce cross-node memory access and contention 2120can improve overall system performance. 2121 2122The "cpuset" controller is hierarchical. That means the controller 2123cannot use CPUs or memory nodes not allowed in its parent. 2124 2125 2126Cpuset Interface Files 2127~~~~~~~~~~~~~~~~~~~~~~ 2128 2129 cpuset.cpus 2130 A read-write multiple values file which exists on non-root 2131 cpuset-enabled cgroups. 2132 2133 It lists the requested CPUs to be used by tasks within this 2134 cgroup. The actual list of CPUs to be granted, however, is 2135 subjected to constraints imposed by its parent and can differ 2136 from the requested CPUs. 2137 2138 The CPU numbers are comma-separated numbers or ranges. 2139 For example:: 2140 2141 # cat cpuset.cpus 2142 0-4,6,8-10 2143 2144 An empty value indicates that the cgroup is using the same 2145 setting as the nearest cgroup ancestor with a non-empty 2146 "cpuset.cpus" or all the available CPUs if none is found. 2147 2148 The value of "cpuset.cpus" stays constant until the next update 2149 and won't be affected by any CPU hotplug events. 2150 2151 cpuset.cpus.effective 2152 A read-only multiple values file which exists on all 2153 cpuset-enabled cgroups. 2154 2155 It lists the onlined CPUs that are actually granted to this 2156 cgroup by its parent. These CPUs are allowed to be used by 2157 tasks within the current cgroup. 2158 2159 If "cpuset.cpus" is empty, the "cpuset.cpus.effective" file shows 2160 all the CPUs from the parent cgroup that can be available to 2161 be used by this cgroup. Otherwise, it should be a subset of 2162 "cpuset.cpus" unless none of the CPUs listed in "cpuset.cpus" 2163 can be granted. In this case, it will be treated just like an 2164 empty "cpuset.cpus". 2165 2166 Its value will be affected by CPU hotplug events. 2167 2168 cpuset.mems 2169 A read-write multiple values file which exists on non-root 2170 cpuset-enabled cgroups. 2171 2172 It lists the requested memory nodes to be used by tasks within 2173 this cgroup. The actual list of memory nodes granted, however, 2174 is subjected to constraints imposed by its parent and can differ 2175 from the requested memory nodes. 2176 2177 The memory node numbers are comma-separated numbers or ranges. 2178 For example:: 2179 2180 # cat cpuset.mems 2181 0-1,3 2182 2183 An empty value indicates that the cgroup is using the same 2184 setting as the nearest cgroup ancestor with a non-empty 2185 "cpuset.mems" or all the available memory nodes if none 2186 is found. 2187 2188 The value of "cpuset.mems" stays constant until the next update 2189 and won't be affected by any memory nodes hotplug events. 2190 2191 Setting a non-empty value to "cpuset.mems" causes memory of 2192 tasks within the cgroup to be migrated to the designated nodes if 2193 they are currently using memory outside of the designated nodes. 2194 2195 There is a cost for this memory migration. The migration 2196 may not be complete and some memory pages may be left behind. 2197 So it is recommended that "cpuset.mems" should be set properly 2198 before spawning new tasks into the cpuset. Even if there is 2199 a need to change "cpuset.mems" with active tasks, it shouldn't 2200 be done frequently. 2201 2202 cpuset.mems.effective 2203 A read-only multiple values file which exists on all 2204 cpuset-enabled cgroups. 2205 2206 It lists the onlined memory nodes that are actually granted to 2207 this cgroup by its parent. These memory nodes are allowed to 2208 be used by tasks within the current cgroup. 2209 2210 If "cpuset.mems" is empty, it shows all the memory nodes from the 2211 parent cgroup that will be available to be used by this cgroup. 2212 Otherwise, it should be a subset of "cpuset.mems" unless none of 2213 the memory nodes listed in "cpuset.mems" can be granted. In this 2214 case, it will be treated just like an empty "cpuset.mems". 2215 2216 Its value will be affected by memory nodes hotplug events. 2217 2218 cpuset.cpus.partition 2219 A read-write single value file which exists on non-root 2220 cpuset-enabled cgroups. This flag is owned by the parent cgroup 2221 and is not delegatable. 2222 2223 It accepts only the following input values when written to. 2224 2225 ========== ===================================== 2226 "member" Non-root member of a partition 2227 "root" Partition root 2228 "isolated" Partition root without load balancing 2229 ========== ===================================== 2230 2231 The root cgroup is always a partition root and its state 2232 cannot be changed. All other non-root cgroups start out as 2233 "member". 2234 2235 When set to "root", the current cgroup is the root of a new 2236 partition or scheduling domain that comprises itself and all 2237 its descendants except those that are separate partition roots 2238 themselves and their descendants. 2239 2240 When set to "isolated", the CPUs in that partition root will 2241 be in an isolated state without any load balancing from the 2242 scheduler. Tasks placed in such a partition with multiple 2243 CPUs should be carefully distributed and bound to each of the 2244 individual CPUs for optimal performance. 2245 2246 The value shown in "cpuset.cpus.effective" of a partition root 2247 is the CPUs that the partition root can dedicate to a potential 2248 new child partition root. The new child subtracts available 2249 CPUs from its parent "cpuset.cpus.effective". 2250 2251 A partition root ("root" or "isolated") can be in one of the 2252 two possible states - valid or invalid. An invalid partition 2253 root is in a degraded state where some state information may 2254 be retained, but behaves more like a "member". 2255 2256 All possible state transitions among "member", "root" and 2257 "isolated" are allowed. 2258 2259 On read, the "cpuset.cpus.partition" file can show the following 2260 values. 2261 2262 ============================= ===================================== 2263 "member" Non-root member of a partition 2264 "root" Partition root 2265 "isolated" Partition root without load balancing 2266 "root invalid (<reason>)" Invalid partition root 2267 "isolated invalid (<reason>)" Invalid isolated partition root 2268 ============================= ===================================== 2269 2270 In the case of an invalid partition root, a descriptive string on 2271 why the partition is invalid is included within parentheses. 2272 2273 For a partition root to become valid, the following conditions 2274 must be met. 2275 2276 1) The "cpuset.cpus" is exclusive with its siblings , i.e. they 2277 are not shared by any of its siblings (exclusivity rule). 2278 2) The parent cgroup is a valid partition root. 2279 3) The "cpuset.cpus" is not empty and must contain at least 2280 one of the CPUs from parent's "cpuset.cpus", i.e. they overlap. 2281 4) The "cpuset.cpus.effective" cannot be empty unless there is 2282 no task associated with this partition. 2283 2284 External events like hotplug or changes to "cpuset.cpus" can 2285 cause a valid partition root to become invalid and vice versa. 2286 Note that a task cannot be moved to a cgroup with empty 2287 "cpuset.cpus.effective". 2288 2289 For a valid partition root with the sibling cpu exclusivity 2290 rule enabled, changes made to "cpuset.cpus" that violate the 2291 exclusivity rule will invalidate the partition as well as its 2292 sibiling partitions with conflicting cpuset.cpus values. So 2293 care must be taking in changing "cpuset.cpus". 2294 2295 A valid non-root parent partition may distribute out all its CPUs 2296 to its child partitions when there is no task associated with it. 2297 2298 Care must be taken to change a valid partition root to 2299 "member" as all its child partitions, if present, will become 2300 invalid causing disruption to tasks running in those child 2301 partitions. These inactivated partitions could be recovered if 2302 their parent is switched back to a partition root with a proper 2303 set of "cpuset.cpus". 2304 2305 Poll and inotify events are triggered whenever the state of 2306 "cpuset.cpus.partition" changes. That includes changes caused 2307 by write to "cpuset.cpus.partition", cpu hotplug or other 2308 changes that modify the validity status of the partition. 2309 This will allow user space agents to monitor unexpected changes 2310 to "cpuset.cpus.partition" without the need to do continuous 2311 polling. 2312 2313 2314Device controller 2315----------------- 2316 2317Device controller manages access to device files. It includes both 2318creation of new device files (using mknod), and access to the 2319existing device files. 2320 2321Cgroup v2 device controller has no interface files and is implemented 2322on top of cgroup BPF. To control access to device files, a user may 2323create bpf programs of type BPF_PROG_TYPE_CGROUP_DEVICE and attach 2324them to cgroups with BPF_CGROUP_DEVICE flag. On an attempt to access a 2325device file, corresponding BPF programs will be executed, and depending 2326on the return value the attempt will succeed or fail with -EPERM. 2327 2328A BPF_PROG_TYPE_CGROUP_DEVICE program takes a pointer to the 2329bpf_cgroup_dev_ctx structure, which describes the device access attempt: 2330access type (mknod/read/write) and device (type, major and minor numbers). 2331If the program returns 0, the attempt fails with -EPERM, otherwise it 2332succeeds. 2333 2334An example of BPF_PROG_TYPE_CGROUP_DEVICE program may be found in 2335tools/testing/selftests/bpf/progs/dev_cgroup.c in the kernel source tree. 2336 2337 2338RDMA 2339---- 2340 2341The "rdma" controller regulates the distribution and accounting of 2342RDMA resources. 2343 2344RDMA Interface Files 2345~~~~~~~~~~~~~~~~~~~~ 2346 2347 rdma.max 2348 A readwrite nested-keyed file that exists for all the cgroups 2349 except root that describes current configured resource limit 2350 for a RDMA/IB device. 2351 2352 Lines are keyed by device name and are not ordered. 2353 Each line contains space separated resource name and its configured 2354 limit that can be distributed. 2355 2356 The following nested keys are defined. 2357 2358 ========== ============================= 2359 hca_handle Maximum number of HCA Handles 2360 hca_object Maximum number of HCA Objects 2361 ========== ============================= 2362 2363 An example for mlx4 and ocrdma device follows:: 2364 2365 mlx4_0 hca_handle=2 hca_object=2000 2366 ocrdma1 hca_handle=3 hca_object=max 2367 2368 rdma.current 2369 A read-only file that describes current resource usage. 2370 It exists for all the cgroup except root. 2371 2372 An example for mlx4 and ocrdma device follows:: 2373 2374 mlx4_0 hca_handle=1 hca_object=20 2375 ocrdma1 hca_handle=1 hca_object=23 2376 2377HugeTLB 2378------- 2379 2380The HugeTLB controller allows to limit the HugeTLB usage per control group and 2381enforces the controller limit during page fault. 2382 2383HugeTLB Interface Files 2384~~~~~~~~~~~~~~~~~~~~~~~ 2385 2386 hugetlb.<hugepagesize>.current 2387 Show current usage for "hugepagesize" hugetlb. It exists for all 2388 the cgroup except root. 2389 2390 hugetlb.<hugepagesize>.max 2391 Set/show the hard limit of "hugepagesize" hugetlb usage. 2392 The default value is "max". It exists for all the cgroup except root. 2393 2394 hugetlb.<hugepagesize>.events 2395 A read-only flat-keyed file which exists on non-root cgroups. 2396 2397 max 2398 The number of allocation failure due to HugeTLB limit 2399 2400 hugetlb.<hugepagesize>.events.local 2401 Similar to hugetlb.<hugepagesize>.events but the fields in the file 2402 are local to the cgroup i.e. not hierarchical. The file modified event 2403 generated on this file reflects only the local events. 2404 2405 hugetlb.<hugepagesize>.numa_stat 2406 Similar to memory.numa_stat, it shows the numa information of the 2407 hugetlb pages of <hugepagesize> in this cgroup. Only active in 2408 use hugetlb pages are included. The per-node values are in bytes. 2409 2410Misc 2411---- 2412 2413The Miscellaneous cgroup provides the resource limiting and tracking 2414mechanism for the scalar resources which cannot be abstracted like the other 2415cgroup resources. Controller is enabled by the CONFIG_CGROUP_MISC config 2416option. 2417 2418A resource can be added to the controller via enum misc_res_type{} in the 2419include/linux/misc_cgroup.h file and the corresponding name via misc_res_name[] 2420in the kernel/cgroup/misc.c file. Provider of the resource must set its 2421capacity prior to using the resource by calling misc_cg_set_capacity(). 2422 2423Once a capacity is set then the resource usage can be updated using charge and 2424uncharge APIs. All of the APIs to interact with misc controller are in 2425include/linux/misc_cgroup.h. 2426 2427Misc Interface Files 2428~~~~~~~~~~~~~~~~~~~~ 2429 2430Miscellaneous controller provides 3 interface files. If two misc resources (res_a and res_b) are registered then: 2431 2432 misc.capacity 2433 A read-only flat-keyed file shown only in the root cgroup. It shows 2434 miscellaneous scalar resources available on the platform along with 2435 their quantities:: 2436 2437 $ cat misc.capacity 2438 res_a 50 2439 res_b 10 2440 2441 misc.current 2442 A read-only flat-keyed file shown in the non-root cgroups. It shows 2443 the current usage of the resources in the cgroup and its children.:: 2444 2445 $ cat misc.current 2446 res_a 3 2447 res_b 0 2448 2449 misc.max 2450 A read-write flat-keyed file shown in the non root cgroups. Allowed 2451 maximum usage of the resources in the cgroup and its children.:: 2452 2453 $ cat misc.max 2454 res_a max 2455 res_b 4 2456 2457 Limit can be set by:: 2458 2459 # echo res_a 1 > misc.max 2460 2461 Limit can be set to max by:: 2462 2463 # echo res_a max > misc.max 2464 2465 Limits can be set higher than the capacity value in the misc.capacity 2466 file. 2467 2468 misc.events 2469 A read-only flat-keyed file which exists on non-root cgroups. The 2470 following entries are defined. Unless specified otherwise, a value 2471 change in this file generates a file modified event. All fields in 2472 this file are hierarchical. 2473 2474 max 2475 The number of times the cgroup's resource usage was 2476 about to go over the max boundary. 2477 2478Migration and Ownership 2479~~~~~~~~~~~~~~~~~~~~~~~ 2480 2481A miscellaneous scalar resource is charged to the cgroup in which it is used 2482first, and stays charged to that cgroup until that resource is freed. Migrating 2483a process to a different cgroup does not move the charge to the destination 2484cgroup where the process has moved. 2485 2486Others 2487------ 2488 2489perf_event 2490~~~~~~~~~~ 2491 2492perf_event controller, if not mounted on a legacy hierarchy, is 2493automatically enabled on the v2 hierarchy so that perf events can 2494always be filtered by cgroup v2 path. The controller can still be 2495moved to a legacy hierarchy after v2 hierarchy is populated. 2496 2497 2498Non-normative information 2499------------------------- 2500 2501This section contains information that isn't considered to be a part of 2502the stable kernel API and so is subject to change. 2503 2504 2505CPU controller root cgroup process behaviour 2506~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 2507 2508When distributing CPU cycles in the root cgroup each thread in this 2509cgroup is treated as if it was hosted in a separate child cgroup of the 2510root cgroup. This child cgroup weight is dependent on its thread nice 2511level. 2512 2513For details of this mapping see sched_prio_to_weight array in 2514kernel/sched/core.c file (values from this array should be scaled 2515appropriately so the neutral - nice 0 - value is 100 instead of 1024). 2516 2517 2518IO controller root cgroup process behaviour 2519~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 2520 2521Root cgroup processes are hosted in an implicit leaf child node. 2522When distributing IO resources this implicit child node is taken into 2523account as if it was a normal child cgroup of the root cgroup with a 2524weight value of 200. 2525 2526 2527Namespace 2528========= 2529 2530Basics 2531------ 2532 2533cgroup namespace provides a mechanism to virtualize the view of the 2534"/proc/$PID/cgroup" file and cgroup mounts. The CLONE_NEWCGROUP clone 2535flag can be used with clone(2) and unshare(2) to create a new cgroup 2536namespace. The process running inside the cgroup namespace will have 2537its "/proc/$PID/cgroup" output restricted to cgroupns root. The 2538cgroupns root is the cgroup of the process at the time of creation of 2539the cgroup namespace. 2540 2541Without cgroup namespace, the "/proc/$PID/cgroup" file shows the 2542complete path of the cgroup of a process. In a container setup where 2543a set of cgroups and namespaces are intended to isolate processes the 2544"/proc/$PID/cgroup" file may leak potential system level information 2545to the isolated processes. For example:: 2546 2547 # cat /proc/self/cgroup 2548 0::/batchjobs/container_id1 2549 2550The path '/batchjobs/container_id1' can be considered as system-data 2551and undesirable to expose to the isolated processes. cgroup namespace 2552can be used to restrict visibility of this path. For example, before 2553creating a cgroup namespace, one would see:: 2554 2555 # ls -l /proc/self/ns/cgroup 2556 lrwxrwxrwx 1 root root 0 2014-07-15 10:37 /proc/self/ns/cgroup -> cgroup:[4026531835] 2557 # cat /proc/self/cgroup 2558 0::/batchjobs/container_id1 2559 2560After unsharing a new namespace, the view changes:: 2561 2562 # ls -l /proc/self/ns/cgroup 2563 lrwxrwxrwx 1 root root 0 2014-07-15 10:35 /proc/self/ns/cgroup -> cgroup:[4026532183] 2564 # cat /proc/self/cgroup 2565 0::/ 2566 2567When some thread from a multi-threaded process unshares its cgroup 2568namespace, the new cgroupns gets applied to the entire process (all 2569the threads). This is natural for the v2 hierarchy; however, for the 2570legacy hierarchies, this may be unexpected. 2571 2572A cgroup namespace is alive as long as there are processes inside or 2573mounts pinning it. When the last usage goes away, the cgroup 2574namespace is destroyed. The cgroupns root and the actual cgroups 2575remain. 2576 2577 2578The Root and Views 2579------------------ 2580 2581The 'cgroupns root' for a cgroup namespace is the cgroup in which the 2582process calling unshare(2) is running. For example, if a process in 2583/batchjobs/container_id1 cgroup calls unshare, cgroup 2584/batchjobs/container_id1 becomes the cgroupns root. For the 2585init_cgroup_ns, this is the real root ('/') cgroup. 2586 2587The cgroupns root cgroup does not change even if the namespace creator 2588process later moves to a different cgroup:: 2589 2590 # ~/unshare -c # unshare cgroupns in some cgroup 2591 # cat /proc/self/cgroup 2592 0::/ 2593 # mkdir sub_cgrp_1 2594 # echo 0 > sub_cgrp_1/cgroup.procs 2595 # cat /proc/self/cgroup 2596 0::/sub_cgrp_1 2597 2598Each process gets its namespace-specific view of "/proc/$PID/cgroup" 2599 2600Processes running inside the cgroup namespace will be able to see 2601cgroup paths (in /proc/self/cgroup) only inside their root cgroup. 2602From within an unshared cgroupns:: 2603 2604 # sleep 100000 & 2605 [1] 7353 2606 # echo 7353 > sub_cgrp_1/cgroup.procs 2607 # cat /proc/7353/cgroup 2608 0::/sub_cgrp_1 2609 2610From the initial cgroup namespace, the real cgroup path will be 2611visible:: 2612 2613 $ cat /proc/7353/cgroup 2614 0::/batchjobs/container_id1/sub_cgrp_1 2615 2616From a sibling cgroup namespace (that is, a namespace rooted at a 2617different cgroup), the cgroup path relative to its own cgroup 2618namespace root will be shown. For instance, if PID 7353's cgroup 2619namespace root is at '/batchjobs/container_id2', then it will see:: 2620 2621 # cat /proc/7353/cgroup 2622 0::/../container_id2/sub_cgrp_1 2623 2624Note that the relative path always starts with '/' to indicate that 2625its relative to the cgroup namespace root of the caller. 2626 2627 2628Migration and setns(2) 2629---------------------- 2630 2631Processes inside a cgroup namespace can move into and out of the 2632namespace root if they have proper access to external cgroups. For 2633example, from inside a namespace with cgroupns root at 2634/batchjobs/container_id1, and assuming that the global hierarchy is 2635still accessible inside cgroupns:: 2636 2637 # cat /proc/7353/cgroup 2638 0::/sub_cgrp_1 2639 # echo 7353 > batchjobs/container_id2/cgroup.procs 2640 # cat /proc/7353/cgroup 2641 0::/../container_id2 2642 2643Note that this kind of setup is not encouraged. A task inside cgroup 2644namespace should only be exposed to its own cgroupns hierarchy. 2645 2646setns(2) to another cgroup namespace is allowed when: 2647 2648(a) the process has CAP_SYS_ADMIN against its current user namespace 2649(b) the process has CAP_SYS_ADMIN against the target cgroup 2650 namespace's userns 2651 2652No implicit cgroup changes happen with attaching to another cgroup 2653namespace. It is expected that the someone moves the attaching 2654process under the target cgroup namespace root. 2655 2656 2657Interaction with Other Namespaces 2658--------------------------------- 2659 2660Namespace specific cgroup hierarchy can be mounted by a process 2661running inside a non-init cgroup namespace:: 2662 2663 # mount -t cgroup2 none $MOUNT_POINT 2664 2665This will mount the unified cgroup hierarchy with cgroupns root as the 2666filesystem root. The process needs CAP_SYS_ADMIN against its user and 2667mount namespaces. 2668 2669The virtualization of /proc/self/cgroup file combined with restricting 2670the view of cgroup hierarchy by namespace-private cgroupfs mount 2671provides a properly isolated cgroup view inside the container. 2672 2673 2674Information on Kernel Programming 2675================================= 2676 2677This section contains kernel programming information in the areas 2678where interacting with cgroup is necessary. cgroup core and 2679controllers are not covered. 2680 2681 2682Filesystem Support for Writeback 2683-------------------------------- 2684 2685A filesystem can support cgroup writeback by updating 2686address_space_operations->writepage[s]() to annotate bio's using the 2687following two functions. 2688 2689 wbc_init_bio(@wbc, @bio) 2690 Should be called for each bio carrying writeback data and 2691 associates the bio with the inode's owner cgroup and the 2692 corresponding request queue. This must be called after 2693 a queue (device) has been associated with the bio and 2694 before submission. 2695 2696 wbc_account_cgroup_owner(@wbc, @page, @bytes) 2697 Should be called for each data segment being written out. 2698 While this function doesn't care exactly when it's called 2699 during the writeback session, it's the easiest and most 2700 natural to call it as data segments are added to a bio. 2701 2702With writeback bio's annotated, cgroup support can be enabled per 2703super_block by setting SB_I_CGROUPWB in ->s_iflags. This allows for 2704selective disabling of cgroup writeback support which is helpful when 2705certain filesystem features, e.g. journaled data mode, are 2706incompatible. 2707 2708wbc_init_bio() binds the specified bio to its cgroup. Depending on 2709the configuration, the bio may be executed at a lower priority and if 2710the writeback session is holding shared resources, e.g. a journal 2711entry, may lead to priority inversion. There is no one easy solution 2712for the problem. Filesystems can try to work around specific problem 2713cases by skipping wbc_init_bio() and using bio_associate_blkg() 2714directly. 2715 2716 2717Deprecated v1 Core Features 2718=========================== 2719 2720- Multiple hierarchies including named ones are not supported. 2721 2722- All v1 mount options are not supported. 2723 2724- The "tasks" file is removed and "cgroup.procs" is not sorted. 2725 2726- "cgroup.clone_children" is removed. 2727 2728- /proc/cgroups is meaningless for v2. Use "cgroup.controllers" file 2729 at the root instead. 2730 2731 2732Issues with v1 and Rationales for v2 2733==================================== 2734 2735Multiple Hierarchies 2736-------------------- 2737 2738cgroup v1 allowed an arbitrary number of hierarchies and each 2739hierarchy could host any number of controllers. While this seemed to 2740provide a high level of flexibility, it wasn't useful in practice. 2741 2742For example, as there is only one instance of each controller, utility 2743type controllers such as freezer which can be useful in all 2744hierarchies could only be used in one. The issue is exacerbated by 2745the fact that controllers couldn't be moved to another hierarchy once 2746hierarchies were populated. Another issue was that all controllers 2747bound to a hierarchy were forced to have exactly the same view of the 2748hierarchy. It wasn't possible to vary the granularity depending on 2749the specific controller. 2750 2751In practice, these issues heavily limited which controllers could be 2752put on the same hierarchy and most configurations resorted to putting 2753each controller on its own hierarchy. Only closely related ones, such 2754as the cpu and cpuacct controllers, made sense to be put on the same 2755hierarchy. This often meant that userland ended up managing multiple 2756similar hierarchies repeating the same steps on each hierarchy 2757whenever a hierarchy management operation was necessary. 2758 2759Furthermore, support for multiple hierarchies came at a steep cost. 2760It greatly complicated cgroup core implementation but more importantly 2761the support for multiple hierarchies restricted how cgroup could be 2762used in general and what controllers was able to do. 2763 2764There was no limit on how many hierarchies there might be, which meant 2765that a thread's cgroup membership couldn't be described in finite 2766length. The key might contain any number of entries and was unlimited 2767in length, which made it highly awkward to manipulate and led to 2768addition of controllers which existed only to identify membership, 2769which in turn exacerbated the original problem of proliferating number 2770of hierarchies. 2771 2772Also, as a controller couldn't have any expectation regarding the 2773topologies of hierarchies other controllers might be on, each 2774controller had to assume that all other controllers were attached to 2775completely orthogonal hierarchies. This made it impossible, or at 2776least very cumbersome, for controllers to cooperate with each other. 2777 2778In most use cases, putting controllers on hierarchies which are 2779completely orthogonal to each other isn't necessary. What usually is 2780called for is the ability to have differing levels of granularity 2781depending on the specific controller. In other words, hierarchy may 2782be collapsed from leaf towards root when viewed from specific 2783controllers. For example, a given configuration might not care about 2784how memory is distributed beyond a certain level while still wanting 2785to control how CPU cycles are distributed. 2786 2787 2788Thread Granularity 2789------------------ 2790 2791cgroup v1 allowed threads of a process to belong to different cgroups. 2792This didn't make sense for some controllers and those controllers 2793ended up implementing different ways to ignore such situations but 2794much more importantly it blurred the line between API exposed to 2795individual applications and system management interface. 2796 2797Generally, in-process knowledge is available only to the process 2798itself; thus, unlike service-level organization of processes, 2799categorizing threads of a process requires active participation from 2800the application which owns the target process. 2801 2802cgroup v1 had an ambiguously defined delegation model which got abused 2803in combination with thread granularity. cgroups were delegated to 2804individual applications so that they can create and manage their own 2805sub-hierarchies and control resource distributions along them. This 2806effectively raised cgroup to the status of a syscall-like API exposed 2807to lay programs. 2808 2809First of all, cgroup has a fundamentally inadequate interface to be 2810exposed this way. For a process to access its own knobs, it has to 2811extract the path on the target hierarchy from /proc/self/cgroup, 2812construct the path by appending the name of the knob to the path, open 2813and then read and/or write to it. This is not only extremely clunky 2814and unusual but also inherently racy. There is no conventional way to 2815define transaction across the required steps and nothing can guarantee 2816that the process would actually be operating on its own sub-hierarchy. 2817 2818cgroup controllers implemented a number of knobs which would never be 2819accepted as public APIs because they were just adding control knobs to 2820system-management pseudo filesystem. cgroup ended up with interface 2821knobs which were not properly abstracted or refined and directly 2822revealed kernel internal details. These knobs got exposed to 2823individual applications through the ill-defined delegation mechanism 2824effectively abusing cgroup as a shortcut to implementing public APIs 2825without going through the required scrutiny. 2826 2827This was painful for both userland and kernel. Userland ended up with 2828misbehaving and poorly abstracted interfaces and kernel exposing and 2829locked into constructs inadvertently. 2830 2831 2832Competition Between Inner Nodes and Threads 2833------------------------------------------- 2834 2835cgroup v1 allowed threads to be in any cgroups which created an 2836interesting problem where threads belonging to a parent cgroup and its 2837children cgroups competed for resources. This was nasty as two 2838different types of entities competed and there was no obvious way to 2839settle it. Different controllers did different things. 2840 2841The cpu controller considered threads and cgroups as equivalents and 2842mapped nice levels to cgroup weights. This worked for some cases but 2843fell flat when children wanted to be allocated specific ratios of CPU 2844cycles and the number of internal threads fluctuated - the ratios 2845constantly changed as the number of competing entities fluctuated. 2846There also were other issues. The mapping from nice level to weight 2847wasn't obvious or universal, and there were various other knobs which 2848simply weren't available for threads. 2849 2850The io controller implicitly created a hidden leaf node for each 2851cgroup to host the threads. The hidden leaf had its own copies of all 2852the knobs with ``leaf_`` prefixed. While this allowed equivalent 2853control over internal threads, it was with serious drawbacks. It 2854always added an extra layer of nesting which wouldn't be necessary 2855otherwise, made the interface messy and significantly complicated the 2856implementation. 2857 2858The memory controller didn't have a way to control what happened 2859between internal tasks and child cgroups and the behavior was not 2860clearly defined. There were attempts to add ad-hoc behaviors and 2861knobs to tailor the behavior to specific workloads which would have 2862led to problems extremely difficult to resolve in the long term. 2863 2864Multiple controllers struggled with internal tasks and came up with 2865different ways to deal with it; unfortunately, all the approaches were 2866severely flawed and, furthermore, the widely different behaviors 2867made cgroup as a whole highly inconsistent. 2868 2869This clearly is a problem which needs to be addressed from cgroup core 2870in a uniform way. 2871 2872 2873Other Interface Issues 2874---------------------- 2875 2876cgroup v1 grew without oversight and developed a large number of 2877idiosyncrasies and inconsistencies. One issue on the cgroup core side 2878was how an empty cgroup was notified - a userland helper binary was 2879forked and executed for each event. The event delivery wasn't 2880recursive or delegatable. The limitations of the mechanism also led 2881to in-kernel event delivery filtering mechanism further complicating 2882the interface. 2883 2884Controller interfaces were problematic too. An extreme example is 2885controllers completely ignoring hierarchical organization and treating 2886all cgroups as if they were all located directly under the root 2887cgroup. Some controllers exposed a large amount of inconsistent 2888implementation details to userland. 2889 2890There also was no consistency across controllers. When a new cgroup 2891was created, some controllers defaulted to not imposing extra 2892restrictions while others disallowed any resource usage until 2893explicitly configured. Configuration knobs for the same type of 2894control used widely differing naming schemes and formats. Statistics 2895and information knobs were named arbitrarily and used different 2896formats and units even in the same controller. 2897 2898cgroup v2 establishes common conventions where appropriate and updates 2899controllers so that they expose minimal and consistent interfaces. 2900 2901 2902Controller Issues and Remedies 2903------------------------------ 2904 2905Memory 2906~~~~~~ 2907 2908The original lower boundary, the soft limit, is defined as a limit 2909that is per default unset. As a result, the set of cgroups that 2910global reclaim prefers is opt-in, rather than opt-out. The costs for 2911optimizing these mostly negative lookups are so high that the 2912implementation, despite its enormous size, does not even provide the 2913basic desirable behavior. First off, the soft limit has no 2914hierarchical meaning. All configured groups are organized in a global 2915rbtree and treated like equal peers, regardless where they are located 2916in the hierarchy. This makes subtree delegation impossible. Second, 2917the soft limit reclaim pass is so aggressive that it not just 2918introduces high allocation latencies into the system, but also impacts 2919system performance due to overreclaim, to the point where the feature 2920becomes self-defeating. 2921 2922The memory.low boundary on the other hand is a top-down allocated 2923reserve. A cgroup enjoys reclaim protection when it's within its 2924effective low, which makes delegation of subtrees possible. It also 2925enjoys having reclaim pressure proportional to its overage when 2926above its effective low. 2927 2928The original high boundary, the hard limit, is defined as a strict 2929limit that can not budge, even if the OOM killer has to be called. 2930But this generally goes against the goal of making the most out of the 2931available memory. The memory consumption of workloads varies during 2932runtime, and that requires users to overcommit. But doing that with a 2933strict upper limit requires either a fairly accurate prediction of the 2934working set size or adding slack to the limit. Since working set size 2935estimation is hard and error prone, and getting it wrong results in 2936OOM kills, most users tend to err on the side of a looser limit and 2937end up wasting precious resources. 2938 2939The memory.high boundary on the other hand can be set much more 2940conservatively. When hit, it throttles allocations by forcing them 2941into direct reclaim to work off the excess, but it never invokes the 2942OOM killer. As a result, a high boundary that is chosen too 2943aggressively will not terminate the processes, but instead it will 2944lead to gradual performance degradation. The user can monitor this 2945and make corrections until the minimal memory footprint that still 2946gives acceptable performance is found. 2947 2948In extreme cases, with many concurrent allocations and a complete 2949breakdown of reclaim progress within the group, the high boundary can 2950be exceeded. But even then it's mostly better to satisfy the 2951allocation from the slack available in other groups or the rest of the 2952system than killing the group. Otherwise, memory.max is there to 2953limit this type of spillover and ultimately contain buggy or even 2954malicious applications. 2955 2956Setting the original memory.limit_in_bytes below the current usage was 2957subject to a race condition, where concurrent charges could cause the 2958limit setting to fail. memory.max on the other hand will first set the 2959limit to prevent new charges, and then reclaim and OOM kill until the 2960new limit is met - or the task writing to memory.max is killed. 2961 2962The combined memory+swap accounting and limiting is replaced by real 2963control over swap space. 2964 2965The main argument for a combined memory+swap facility in the original 2966cgroup design was that global or parental pressure would always be 2967able to swap all anonymous memory of a child group, regardless of the 2968child's own (possibly untrusted) configuration. However, untrusted 2969groups can sabotage swapping by other means - such as referencing its 2970anonymous memory in a tight loop - and an admin can not assume full 2971swappability when overcommitting untrusted jobs. 2972 2973For trusted jobs, on the other hand, a combined counter is not an 2974intuitive userspace interface, and it flies in the face of the idea 2975that cgroup controllers should account and limit specific physical 2976resources. Swap space is a resource like all others in the system, 2977and that's why unified hierarchy allows distributing it separately. 2978