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 979Controllers 980=========== 981 982.. _cgroup-v2-cpu: 983 984CPU 985--- 986 987The "cpu" controllers regulates distribution of CPU cycles. This 988controller implements weight and absolute bandwidth limit models for 989normal scheduling policy and absolute bandwidth allocation model for 990realtime scheduling policy. 991 992In all the above models, cycles distribution is defined only on a temporal 993base and it does not account for the frequency at which tasks are executed. 994The (optional) utilization clamping support allows to hint the schedutil 995cpufreq governor about the minimum desired frequency which should always be 996provided by a CPU, as well as the maximum desired frequency, which should not 997be exceeded by a CPU. 998 999WARNING: cgroup2 doesn't yet support control of realtime processes and 1000the cpu controller can only be enabled when all RT processes are in 1001the root cgroup. Be aware that system management software may already 1002have placed RT processes into nonroot cgroups during the system boot 1003process, and these processes may need to be moved to the root cgroup 1004before the cpu controller can be enabled. 1005 1006 1007CPU Interface Files 1008~~~~~~~~~~~~~~~~~~~ 1009 1010All time durations are in microseconds. 1011 1012 cpu.stat 1013 A read-only flat-keyed file. 1014 This file exists whether the controller is enabled or not. 1015 1016 It always reports the following three stats: 1017 1018 - usage_usec 1019 - user_usec 1020 - system_usec 1021 1022 and the following three when the controller is enabled: 1023 1024 - nr_periods 1025 - nr_throttled 1026 - throttled_usec 1027 - nr_bursts 1028 - burst_usec 1029 1030 cpu.weight 1031 A read-write single value file which exists on non-root 1032 cgroups. The default is "100". 1033 1034 The weight in the range [1, 10000]. 1035 1036 cpu.weight.nice 1037 A read-write single value file which exists on non-root 1038 cgroups. The default is "0". 1039 1040 The nice value is in the range [-20, 19]. 1041 1042 This interface file is an alternative interface for 1043 "cpu.weight" and allows reading and setting weight using the 1044 same values used by nice(2). Because the range is smaller and 1045 granularity is coarser for the nice values, the read value is 1046 the closest approximation of the current weight. 1047 1048 cpu.max 1049 A read-write two value file which exists on non-root cgroups. 1050 The default is "max 100000". 1051 1052 The maximum bandwidth limit. It's in the following format:: 1053 1054 $MAX $PERIOD 1055 1056 which indicates that the group may consume upto $MAX in each 1057 $PERIOD duration. "max" for $MAX indicates no limit. If only 1058 one number is written, $MAX is updated. 1059 1060 cpu.max.burst 1061 A read-write single value file which exists on non-root 1062 cgroups. The default is "0". 1063 1064 The burst in the range [0, $MAX]. 1065 1066 cpu.pressure 1067 A read-write nested-keyed file. 1068 1069 Shows pressure stall information for CPU. See 1070 :ref:`Documentation/accounting/psi.rst <psi>` for details. 1071 1072 cpu.uclamp.min 1073 A read-write single value file which exists on non-root cgroups. 1074 The default is "0", i.e. no utilization boosting. 1075 1076 The requested minimum utilization (protection) as a percentage 1077 rational number, e.g. 12.34 for 12.34%. 1078 1079 This interface allows reading and setting minimum utilization clamp 1080 values similar to the sched_setattr(2). This minimum utilization 1081 value is used to clamp the task specific minimum utilization clamp. 1082 1083 The requested minimum utilization (protection) is always capped by 1084 the current value for the maximum utilization (limit), i.e. 1085 `cpu.uclamp.max`. 1086 1087 cpu.uclamp.max 1088 A read-write single value file which exists on non-root cgroups. 1089 The default is "max". i.e. no utilization capping 1090 1091 The requested maximum utilization (limit) as a percentage rational 1092 number, e.g. 98.76 for 98.76%. 1093 1094 This interface allows reading and setting maximum utilization clamp 1095 values similar to the sched_setattr(2). This maximum utilization 1096 value is used to clamp the task specific maximum utilization clamp. 1097 1098 1099 1100Memory 1101------ 1102 1103The "memory" controller regulates distribution of memory. Memory is 1104stateful and implements both limit and protection models. Due to the 1105intertwining between memory usage and reclaim pressure and the 1106stateful nature of memory, the distribution model is relatively 1107complex. 1108 1109While not completely water-tight, all major memory usages by a given 1110cgroup are tracked so that the total memory consumption can be 1111accounted and controlled to a reasonable extent. Currently, the 1112following types of memory usages are tracked. 1113 1114- Userland memory - page cache and anonymous memory. 1115 1116- Kernel data structures such as dentries and inodes. 1117 1118- TCP socket buffers. 1119 1120The above list may expand in the future for better coverage. 1121 1122 1123Memory Interface Files 1124~~~~~~~~~~~~~~~~~~~~~~ 1125 1126All memory amounts are in bytes. If a value which is not aligned to 1127PAGE_SIZE is written, the value may be rounded up to the closest 1128PAGE_SIZE multiple when read back. 1129 1130 memory.current 1131 A read-only single value file which exists on non-root 1132 cgroups. 1133 1134 The total amount of memory currently being used by the cgroup 1135 and its descendants. 1136 1137 memory.min 1138 A read-write single value file which exists on non-root 1139 cgroups. The default is "0". 1140 1141 Hard memory protection. If the memory usage of a cgroup 1142 is within its effective min boundary, the cgroup's memory 1143 won't be reclaimed under any conditions. If there is no 1144 unprotected reclaimable memory available, OOM killer 1145 is invoked. Above the effective min boundary (or 1146 effective low boundary if it is higher), pages are reclaimed 1147 proportionally to the overage, reducing reclaim pressure for 1148 smaller overages. 1149 1150 Effective min boundary is limited by memory.min values of 1151 all ancestor cgroups. If there is memory.min overcommitment 1152 (child cgroup or cgroups are requiring more protected memory 1153 than parent will allow), then each child cgroup will get 1154 the part of parent's protection proportional to its 1155 actual memory usage below memory.min. 1156 1157 Putting more memory than generally available under this 1158 protection is discouraged and may lead to constant OOMs. 1159 1160 If a memory cgroup is not populated with processes, 1161 its memory.min is ignored. 1162 1163 memory.low 1164 A read-write single value file which exists on non-root 1165 cgroups. The default is "0". 1166 1167 Best-effort memory protection. If the memory usage of a 1168 cgroup is within its effective low boundary, the cgroup's 1169 memory won't be reclaimed unless there is no reclaimable 1170 memory available in unprotected cgroups. 1171 Above the effective low boundary (or 1172 effective min boundary if it is higher), pages are reclaimed 1173 proportionally to the overage, reducing reclaim pressure for 1174 smaller overages. 1175 1176 Effective low boundary is limited by memory.low values of 1177 all ancestor cgroups. If there is memory.low overcommitment 1178 (child cgroup or cgroups are requiring more protected memory 1179 than parent will allow), then each child cgroup will get 1180 the part of parent's protection proportional to its 1181 actual memory usage below memory.low. 1182 1183 Putting more memory than generally available under this 1184 protection is discouraged. 1185 1186 memory.high 1187 A read-write single value file which exists on non-root 1188 cgroups. The default is "max". 1189 1190 Memory usage throttle limit. This is the main mechanism to 1191 control memory usage of a cgroup. If a cgroup's usage goes 1192 over the high boundary, the processes of the cgroup are 1193 throttled and put under heavy reclaim pressure. 1194 1195 Going over the high limit never invokes the OOM killer and 1196 under extreme conditions the limit may be breached. 1197 1198 memory.max 1199 A read-write single value file which exists on non-root 1200 cgroups. The default is "max". 1201 1202 Memory usage hard limit. This is the final protection 1203 mechanism. If a cgroup's memory usage reaches this limit and 1204 can't be reduced, the OOM killer is invoked in the cgroup. 1205 Under certain circumstances, the usage may go over the limit 1206 temporarily. 1207 1208 In default configuration regular 0-order allocations always 1209 succeed unless OOM killer chooses current task as a victim. 1210 1211 Some kinds of allocations don't invoke the OOM killer. 1212 Caller could retry them differently, return into userspace 1213 as -ENOMEM or silently ignore in cases like disk readahead. 1214 1215 This is the ultimate protection mechanism. As long as the 1216 high limit is used and monitored properly, this limit's 1217 utility is limited to providing the final safety net. 1218 1219 memory.reclaim 1220 A write-only nested-keyed file which exists for all cgroups. 1221 1222 This is a simple interface to trigger memory reclaim in the 1223 target cgroup. 1224 1225 This file accepts a single key, the number of bytes to reclaim. 1226 No nested keys are currently supported. 1227 1228 Example:: 1229 1230 echo "1G" > memory.reclaim 1231 1232 The interface can be later extended with nested keys to 1233 configure the reclaim behavior. For example, specify the 1234 type of memory to reclaim from (anon, file, ..). 1235 1236 Please note that the kernel can over or under reclaim from 1237 the target cgroup. If less bytes are reclaimed than the 1238 specified amount, -EAGAIN is returned. 1239 1240 memory.peak 1241 A read-only single value file which exists on non-root 1242 cgroups. 1243 1244 The max memory usage recorded for the cgroup and its 1245 descendants since the creation of the cgroup. 1246 1247 memory.oom.group 1248 A read-write single value file which exists on non-root 1249 cgroups. The default value is "0". 1250 1251 Determines whether the cgroup should be treated as 1252 an indivisible workload by the OOM killer. If set, 1253 all tasks belonging to the cgroup or to its descendants 1254 (if the memory cgroup is not a leaf cgroup) are killed 1255 together or not at all. This can be used to avoid 1256 partial kills to guarantee workload integrity. 1257 1258 Tasks with the OOM protection (oom_score_adj set to -1000) 1259 are treated as an exception and are never killed. 1260 1261 If the OOM killer is invoked in a cgroup, it's not going 1262 to kill any tasks outside of this cgroup, regardless 1263 memory.oom.group values of ancestor cgroups. 1264 1265 memory.events 1266 A read-only flat-keyed file which exists on non-root cgroups. 1267 The following entries are defined. Unless specified 1268 otherwise, a value change in this file generates a file 1269 modified event. 1270 1271 Note that all fields in this file are hierarchical and the 1272 file modified event can be generated due to an event down the 1273 hierarchy. For the local events at the cgroup level see 1274 memory.events.local. 1275 1276 low 1277 The number of times the cgroup is reclaimed due to 1278 high memory pressure even though its usage is under 1279 the low boundary. This usually indicates that the low 1280 boundary is over-committed. 1281 1282 high 1283 The number of times processes of the cgroup are 1284 throttled and routed to perform direct memory reclaim 1285 because the high memory boundary was exceeded. For a 1286 cgroup whose memory usage is capped by the high limit 1287 rather than global memory pressure, this event's 1288 occurrences are expected. 1289 1290 max 1291 The number of times the cgroup's memory usage was 1292 about to go over the max boundary. If direct reclaim 1293 fails to bring it down, the cgroup goes to OOM state. 1294 1295 oom 1296 The number of time the cgroup's memory usage was 1297 reached the limit and allocation was about to fail. 1298 1299 This event is not raised if the OOM killer is not 1300 considered as an option, e.g. for failed high-order 1301 allocations or if caller asked to not retry attempts. 1302 1303 oom_kill 1304 The number of processes belonging to this cgroup 1305 killed by any kind of OOM killer. 1306 1307 oom_group_kill 1308 The number of times a group OOM has occurred. 1309 1310 memory.events.local 1311 Similar to memory.events but the fields in the file are local 1312 to the cgroup i.e. not hierarchical. The file modified event 1313 generated on this file reflects only the local events. 1314 1315 memory.stat 1316 A read-only flat-keyed file which exists on non-root cgroups. 1317 1318 This breaks down the cgroup's memory footprint into different 1319 types of memory, type-specific details, and other information 1320 on the state and past events of the memory management system. 1321 1322 All memory amounts are in bytes. 1323 1324 The entries are ordered to be human readable, and new entries 1325 can show up in the middle. Don't rely on items remaining in a 1326 fixed position; use the keys to look up specific values! 1327 1328 If the entry has no per-node counter (or not show in the 1329 memory.numa_stat). We use 'npn' (non-per-node) as the tag 1330 to indicate that it will not show in the memory.numa_stat. 1331 1332 anon 1333 Amount of memory used in anonymous mappings such as 1334 brk(), sbrk(), and mmap(MAP_ANONYMOUS) 1335 1336 file 1337 Amount of memory used to cache filesystem data, 1338 including tmpfs and shared memory. 1339 1340 kernel (npn) 1341 Amount of total kernel memory, including 1342 (kernel_stack, pagetables, percpu, vmalloc, slab) in 1343 addition to other kernel memory use cases. 1344 1345 kernel_stack 1346 Amount of memory allocated to kernel stacks. 1347 1348 pagetables 1349 Amount of memory allocated for page tables. 1350 1351 percpu (npn) 1352 Amount of memory used for storing per-cpu kernel 1353 data structures. 1354 1355 sock (npn) 1356 Amount of memory used in network transmission buffers 1357 1358 vmalloc (npn) 1359 Amount of memory used for vmap backed memory. 1360 1361 shmem 1362 Amount of cached filesystem data that is swap-backed, 1363 such as tmpfs, shm segments, shared anonymous mmap()s 1364 1365 zswap 1366 Amount of memory consumed by the zswap compression backend. 1367 1368 zswapped 1369 Amount of application memory swapped out to zswap. 1370 1371 file_mapped 1372 Amount of cached filesystem data mapped with mmap() 1373 1374 file_dirty 1375 Amount of cached filesystem data that was modified but 1376 not yet written back to disk 1377 1378 file_writeback 1379 Amount of cached filesystem data that was modified and 1380 is currently being written back to disk 1381 1382 swapcached 1383 Amount of swap cached in memory. The swapcache is accounted 1384 against both memory and swap usage. 1385 1386 anon_thp 1387 Amount of memory used in anonymous mappings backed by 1388 transparent hugepages 1389 1390 file_thp 1391 Amount of cached filesystem data backed by transparent 1392 hugepages 1393 1394 shmem_thp 1395 Amount of shm, tmpfs, shared anonymous mmap()s backed by 1396 transparent hugepages 1397 1398 inactive_anon, active_anon, inactive_file, active_file, unevictable 1399 Amount of memory, swap-backed and filesystem-backed, 1400 on the internal memory management lists used by the 1401 page reclaim algorithm. 1402 1403 As these represent internal list state (eg. shmem pages are on anon 1404 memory management lists), inactive_foo + active_foo may not be equal to 1405 the value for the foo counter, since the foo counter is type-based, not 1406 list-based. 1407 1408 slab_reclaimable 1409 Part of "slab" that might be reclaimed, such as 1410 dentries and inodes. 1411 1412 slab_unreclaimable 1413 Part of "slab" that cannot be reclaimed on memory 1414 pressure. 1415 1416 slab (npn) 1417 Amount of memory used for storing in-kernel data 1418 structures. 1419 1420 workingset_refault_anon 1421 Number of refaults of previously evicted anonymous pages. 1422 1423 workingset_refault_file 1424 Number of refaults of previously evicted file pages. 1425 1426 workingset_activate_anon 1427 Number of refaulted anonymous pages that were immediately 1428 activated. 1429 1430 workingset_activate_file 1431 Number of refaulted file pages that were immediately activated. 1432 1433 workingset_restore_anon 1434 Number of restored anonymous pages which have been detected as 1435 an active workingset before they got reclaimed. 1436 1437 workingset_restore_file 1438 Number of restored file pages which have been detected as an 1439 active workingset before they got reclaimed. 1440 1441 workingset_nodereclaim 1442 Number of times a shadow node has been reclaimed 1443 1444 pgfault (npn) 1445 Total number of page faults incurred 1446 1447 pgmajfault (npn) 1448 Number of major page faults incurred 1449 1450 pgrefill (npn) 1451 Amount of scanned pages (in an active LRU list) 1452 1453 pgscan (npn) 1454 Amount of scanned pages (in an inactive LRU list) 1455 1456 pgsteal (npn) 1457 Amount of reclaimed pages 1458 1459 pgactivate (npn) 1460 Amount of pages moved to the active LRU list 1461 1462 pgdeactivate (npn) 1463 Amount of pages moved to the inactive LRU list 1464 1465 pglazyfree (npn) 1466 Amount of pages postponed to be freed under memory pressure 1467 1468 pglazyfreed (npn) 1469 Amount of reclaimed lazyfree pages 1470 1471 thp_fault_alloc (npn) 1472 Number of transparent hugepages which were allocated to satisfy 1473 a page fault. This counter is not present when CONFIG_TRANSPARENT_HUGEPAGE 1474 is not set. 1475 1476 thp_collapse_alloc (npn) 1477 Number of transparent hugepages which were allocated to allow 1478 collapsing an existing range of pages. This counter is not 1479 present when CONFIG_TRANSPARENT_HUGEPAGE is not set. 1480 1481 memory.numa_stat 1482 A read-only nested-keyed file which exists on non-root cgroups. 1483 1484 This breaks down the cgroup's memory footprint into different 1485 types of memory, type-specific details, and other information 1486 per node on the state of the memory management system. 1487 1488 This is useful for providing visibility into the NUMA locality 1489 information within an memcg since the pages are allowed to be 1490 allocated from any physical node. One of the use case is evaluating 1491 application performance by combining this information with the 1492 application's CPU allocation. 1493 1494 All memory amounts are in bytes. 1495 1496 The output format of memory.numa_stat is:: 1497 1498 type N0=<bytes in node 0> N1=<bytes in node 1> ... 1499 1500 The entries are ordered to be human readable, and new entries 1501 can show up in the middle. Don't rely on items remaining in a 1502 fixed position; use the keys to look up specific values! 1503 1504 The entries can refer to the memory.stat. 1505 1506 memory.swap.current 1507 A read-only single value file which exists on non-root 1508 cgroups. 1509 1510 The total amount of swap currently being used by the cgroup 1511 and its descendants. 1512 1513 memory.swap.high 1514 A read-write single value file which exists on non-root 1515 cgroups. The default is "max". 1516 1517 Swap usage throttle limit. If a cgroup's swap usage exceeds 1518 this limit, all its further allocations will be throttled to 1519 allow userspace to implement custom out-of-memory procedures. 1520 1521 This limit marks a point of no return for the cgroup. It is NOT 1522 designed to manage the amount of swapping a workload does 1523 during regular operation. Compare to memory.swap.max, which 1524 prohibits swapping past a set amount, but lets the cgroup 1525 continue unimpeded as long as other memory can be reclaimed. 1526 1527 Healthy workloads are not expected to reach this limit. 1528 1529 memory.swap.max 1530 A read-write single value file which exists on non-root 1531 cgroups. The default is "max". 1532 1533 Swap usage hard limit. If a cgroup's swap usage reaches this 1534 limit, anonymous memory of the cgroup will not be swapped out. 1535 1536 memory.swap.events 1537 A read-only flat-keyed file which exists on non-root cgroups. 1538 The following entries are defined. Unless specified 1539 otherwise, a value change in this file generates a file 1540 modified event. 1541 1542 high 1543 The number of times the cgroup's swap usage was over 1544 the high threshold. 1545 1546 max 1547 The number of times the cgroup's swap usage was about 1548 to go over the max boundary and swap allocation 1549 failed. 1550 1551 fail 1552 The number of times swap allocation failed either 1553 because of running out of swap system-wide or max 1554 limit. 1555 1556 When reduced under the current usage, the existing swap 1557 entries are reclaimed gradually and the swap usage may stay 1558 higher than the limit for an extended period of time. This 1559 reduces the impact on the workload and memory management. 1560 1561 memory.zswap.current 1562 A read-only single value file which exists on non-root 1563 cgroups. 1564 1565 The total amount of memory consumed by the zswap compression 1566 backend. 1567 1568 memory.zswap.max 1569 A read-write single value file which exists on non-root 1570 cgroups. The default is "max". 1571 1572 Zswap usage hard limit. If a cgroup's zswap pool reaches this 1573 limit, it will refuse to take any more stores before existing 1574 entries fault back in or are written out to disk. 1575 1576 memory.pressure 1577 A read-only nested-keyed file. 1578 1579 Shows pressure stall information for memory. See 1580 :ref:`Documentation/accounting/psi.rst <psi>` for details. 1581 1582 1583Usage Guidelines 1584~~~~~~~~~~~~~~~~ 1585 1586"memory.high" is the main mechanism to control memory usage. 1587Over-committing on high limit (sum of high limits > available memory) 1588and letting global memory pressure to distribute memory according to 1589usage is a viable strategy. 1590 1591Because breach of the high limit doesn't trigger the OOM killer but 1592throttles the offending cgroup, a management agent has ample 1593opportunities to monitor and take appropriate actions such as granting 1594more memory or terminating the workload. 1595 1596Determining whether a cgroup has enough memory is not trivial as 1597memory usage doesn't indicate whether the workload can benefit from 1598more memory. For example, a workload which writes data received from 1599network to a file can use all available memory but can also operate as 1600performant with a small amount of memory. A measure of memory 1601pressure - how much the workload is being impacted due to lack of 1602memory - is necessary to determine whether a workload needs more 1603memory; unfortunately, memory pressure monitoring mechanism isn't 1604implemented yet. 1605 1606 1607Memory Ownership 1608~~~~~~~~~~~~~~~~ 1609 1610A memory area is charged to the cgroup which instantiated it and stays 1611charged to the cgroup until the area is released. Migrating a process 1612to a different cgroup doesn't move the memory usages that it 1613instantiated while in the previous cgroup to the new cgroup. 1614 1615A memory area may be used by processes belonging to different cgroups. 1616To which cgroup the area will be charged is in-deterministic; however, 1617over time, the memory area is likely to end up in a cgroup which has 1618enough memory allowance to avoid high reclaim pressure. 1619 1620If a cgroup sweeps a considerable amount of memory which is expected 1621to be accessed repeatedly by other cgroups, it may make sense to use 1622POSIX_FADV_DONTNEED to relinquish the ownership of memory areas 1623belonging to the affected files to ensure correct memory ownership. 1624 1625 1626IO 1627-- 1628 1629The "io" controller regulates the distribution of IO resources. This 1630controller implements both weight based and absolute bandwidth or IOPS 1631limit distribution; however, weight based distribution is available 1632only if cfq-iosched is in use and neither scheme is available for 1633blk-mq devices. 1634 1635 1636IO Interface Files 1637~~~~~~~~~~~~~~~~~~ 1638 1639 io.stat 1640 A read-only nested-keyed file. 1641 1642 Lines are keyed by $MAJ:$MIN device numbers and not ordered. 1643 The following nested keys are defined. 1644 1645 ====== ===================== 1646 rbytes Bytes read 1647 wbytes Bytes written 1648 rios Number of read IOs 1649 wios Number of write IOs 1650 dbytes Bytes discarded 1651 dios Number of discard IOs 1652 ====== ===================== 1653 1654 An example read output follows:: 1655 1656 8:16 rbytes=1459200 wbytes=314773504 rios=192 wios=353 dbytes=0 dios=0 1657 8:0 rbytes=90430464 wbytes=299008000 rios=8950 wios=1252 dbytes=50331648 dios=3021 1658 1659 io.cost.qos 1660 A read-write nested-keyed file which exists only on the root 1661 cgroup. 1662 1663 This file configures the Quality of Service of the IO cost 1664 model based controller (CONFIG_BLK_CGROUP_IOCOST) which 1665 currently implements "io.weight" proportional control. Lines 1666 are keyed by $MAJ:$MIN device numbers and not ordered. The 1667 line for a given device is populated on the first write for 1668 the device on "io.cost.qos" or "io.cost.model". The following 1669 nested keys are defined. 1670 1671 ====== ===================================== 1672 enable Weight-based control enable 1673 ctrl "auto" or "user" 1674 rpct Read latency percentile [0, 100] 1675 rlat Read latency threshold 1676 wpct Write latency percentile [0, 100] 1677 wlat Write latency threshold 1678 min Minimum scaling percentage [1, 10000] 1679 max Maximum scaling percentage [1, 10000] 1680 ====== ===================================== 1681 1682 The controller is disabled by default and can be enabled by 1683 setting "enable" to 1. "rpct" and "wpct" parameters default 1684 to zero and the controller uses internal device saturation 1685 state to adjust the overall IO rate between "min" and "max". 1686 1687 When a better control quality is needed, latency QoS 1688 parameters can be configured. For example:: 1689 1690 8:16 enable=1 ctrl=auto rpct=95.00 rlat=75000 wpct=95.00 wlat=150000 min=50.00 max=150.0 1691 1692 shows that on sdb, the controller is enabled, will consider 1693 the device saturated if the 95th percentile of read completion 1694 latencies is above 75ms or write 150ms, and adjust the overall 1695 IO issue rate between 50% and 150% accordingly. 1696 1697 The lower the saturation point, the better the latency QoS at 1698 the cost of aggregate bandwidth. The narrower the allowed 1699 adjustment range between "min" and "max", the more conformant 1700 to the cost model the IO behavior. Note that the IO issue 1701 base rate may be far off from 100% and setting "min" and "max" 1702 blindly can lead to a significant loss of device capacity or 1703 control quality. "min" and "max" are useful for regulating 1704 devices which show wide temporary behavior changes - e.g. a 1705 ssd which accepts writes at the line speed for a while and 1706 then completely stalls for multiple seconds. 1707 1708 When "ctrl" is "auto", the parameters are controlled by the 1709 kernel and may change automatically. Setting "ctrl" to "user" 1710 or setting any of the percentile and latency parameters puts 1711 it into "user" mode and disables the automatic changes. The 1712 automatic mode can be restored by setting "ctrl" to "auto". 1713 1714 io.cost.model 1715 A read-write nested-keyed file which exists only on the root 1716 cgroup. 1717 1718 This file configures the cost model of the IO cost model based 1719 controller (CONFIG_BLK_CGROUP_IOCOST) which currently 1720 implements "io.weight" proportional control. Lines are keyed 1721 by $MAJ:$MIN device numbers and not ordered. The line for a 1722 given device is populated on the first write for the device on 1723 "io.cost.qos" or "io.cost.model". The following nested keys 1724 are defined. 1725 1726 ===== ================================ 1727 ctrl "auto" or "user" 1728 model The cost model in use - "linear" 1729 ===== ================================ 1730 1731 When "ctrl" is "auto", the kernel may change all parameters 1732 dynamically. When "ctrl" is set to "user" or any other 1733 parameters are written to, "ctrl" become "user" and the 1734 automatic changes are disabled. 1735 1736 When "model" is "linear", the following model parameters are 1737 defined. 1738 1739 ============= ======================================== 1740 [r|w]bps The maximum sequential IO throughput 1741 [r|w]seqiops The maximum 4k sequential IOs per second 1742 [r|w]randiops The maximum 4k random IOs per second 1743 ============= ======================================== 1744 1745 From the above, the builtin linear model determines the base 1746 costs of a sequential and random IO and the cost coefficient 1747 for the IO size. While simple, this model can cover most 1748 common device classes acceptably. 1749 1750 The IO cost model isn't expected to be accurate in absolute 1751 sense and is scaled to the device behavior dynamically. 1752 1753 If needed, tools/cgroup/iocost_coef_gen.py can be used to 1754 generate device-specific coefficients. 1755 1756 io.weight 1757 A read-write flat-keyed file which exists on non-root cgroups. 1758 The default is "default 100". 1759 1760 The first line is the default weight applied to devices 1761 without specific override. The rest are overrides keyed by 1762 $MAJ:$MIN device numbers and not ordered. The weights are in 1763 the range [1, 10000] and specifies the relative amount IO time 1764 the cgroup can use in relation to its siblings. 1765 1766 The default weight can be updated by writing either "default 1767 $WEIGHT" or simply "$WEIGHT". Overrides can be set by writing 1768 "$MAJ:$MIN $WEIGHT" and unset by writing "$MAJ:$MIN default". 1769 1770 An example read output follows:: 1771 1772 default 100 1773 8:16 200 1774 8:0 50 1775 1776 io.max 1777 A read-write nested-keyed file which exists on non-root 1778 cgroups. 1779 1780 BPS and IOPS based IO limit. Lines are keyed by $MAJ:$MIN 1781 device numbers and not ordered. The following nested keys are 1782 defined. 1783 1784 ===== ================================== 1785 rbps Max read bytes per second 1786 wbps Max write bytes per second 1787 riops Max read IO operations per second 1788 wiops Max write IO operations per second 1789 ===== ================================== 1790 1791 When writing, any number of nested key-value pairs can be 1792 specified in any order. "max" can be specified as the value 1793 to remove a specific limit. If the same key is specified 1794 multiple times, the outcome is undefined. 1795 1796 BPS and IOPS are measured in each IO direction and IOs are 1797 delayed if limit is reached. Temporary bursts are allowed. 1798 1799 Setting read limit at 2M BPS and write at 120 IOPS for 8:16:: 1800 1801 echo "8:16 rbps=2097152 wiops=120" > io.max 1802 1803 Reading returns the following:: 1804 1805 8:16 rbps=2097152 wbps=max riops=max wiops=120 1806 1807 Write IOPS limit can be removed by writing the following:: 1808 1809 echo "8:16 wiops=max" > io.max 1810 1811 Reading now returns the following:: 1812 1813 8:16 rbps=2097152 wbps=max riops=max wiops=max 1814 1815 io.pressure 1816 A read-only nested-keyed file. 1817 1818 Shows pressure stall information for IO. See 1819 :ref:`Documentation/accounting/psi.rst <psi>` for details. 1820 1821 1822Writeback 1823~~~~~~~~~ 1824 1825Page cache is dirtied through buffered writes and shared mmaps and 1826written asynchronously to the backing filesystem by the writeback 1827mechanism. Writeback sits between the memory and IO domains and 1828regulates the proportion of dirty memory by balancing dirtying and 1829write IOs. 1830 1831The io controller, in conjunction with the memory controller, 1832implements control of page cache writeback IOs. The memory controller 1833defines the memory domain that dirty memory ratio is calculated and 1834maintained for and the io controller defines the io domain which 1835writes out dirty pages for the memory domain. Both system-wide and 1836per-cgroup dirty memory states are examined and the more restrictive 1837of the two is enforced. 1838 1839cgroup writeback requires explicit support from the underlying 1840filesystem. Currently, cgroup writeback is implemented on ext2, ext4, 1841btrfs, f2fs, and xfs. On other filesystems, all writeback IOs are 1842attributed to the root cgroup. 1843 1844There are inherent differences in memory and writeback management 1845which affects how cgroup ownership is tracked. Memory is tracked per 1846page while writeback per inode. For the purpose of writeback, an 1847inode is assigned to a cgroup and all IO requests to write dirty pages 1848from the inode are attributed to that cgroup. 1849 1850As cgroup ownership for memory is tracked per page, there can be pages 1851which are associated with different cgroups than the one the inode is 1852associated with. These are called foreign pages. The writeback 1853constantly keeps track of foreign pages and, if a particular foreign 1854cgroup becomes the majority over a certain period of time, switches 1855the ownership of the inode to that cgroup. 1856 1857While this model is enough for most use cases where a given inode is 1858mostly dirtied by a single cgroup even when the main writing cgroup 1859changes over time, use cases where multiple cgroups write to a single 1860inode simultaneously are not supported well. In such circumstances, a 1861significant portion of IOs are likely to be attributed incorrectly. 1862As memory controller assigns page ownership on the first use and 1863doesn't update it until the page is released, even if writeback 1864strictly follows page ownership, multiple cgroups dirtying overlapping 1865areas wouldn't work as expected. It's recommended to avoid such usage 1866patterns. 1867 1868The sysctl knobs which affect writeback behavior are applied to cgroup 1869writeback as follows. 1870 1871 vm.dirty_background_ratio, vm.dirty_ratio 1872 These ratios apply the same to cgroup writeback with the 1873 amount of available memory capped by limits imposed by the 1874 memory controller and system-wide clean memory. 1875 1876 vm.dirty_background_bytes, vm.dirty_bytes 1877 For cgroup writeback, this is calculated into ratio against 1878 total available memory and applied the same way as 1879 vm.dirty[_background]_ratio. 1880 1881 1882IO Latency 1883~~~~~~~~~~ 1884 1885This is a cgroup v2 controller for IO workload protection. You provide a group 1886with a latency target, and if the average latency exceeds that target the 1887controller will throttle any peers that have a lower latency target than the 1888protected workload. 1889 1890The limits are only applied at the peer level in the hierarchy. This means that 1891in the diagram below, only groups A, B, and C will influence each other, and 1892groups D and F will influence each other. Group G will influence nobody:: 1893 1894 [root] 1895 / | \ 1896 A B C 1897 / \ | 1898 D F G 1899 1900 1901So the ideal way to configure this is to set io.latency in groups A, B, and C. 1902Generally you do not want to set a value lower than the latency your device 1903supports. Experiment to find the value that works best for your workload. 1904Start at higher than the expected latency for your device and watch the 1905avg_lat value in io.stat for your workload group to get an idea of the 1906latency you see during normal operation. Use the avg_lat value as a basis for 1907your real setting, setting at 10-15% higher than the value in io.stat. 1908 1909How IO Latency Throttling Works 1910~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 1911 1912io.latency is work conserving; so as long as everybody is meeting their latency 1913target the controller doesn't do anything. Once a group starts missing its 1914target it begins throttling any peer group that has a higher target than itself. 1915This throttling takes 2 forms: 1916 1917- Queue depth throttling. This is the number of outstanding IO's a group is 1918 allowed to have. We will clamp down relatively quickly, starting at no limit 1919 and going all the way down to 1 IO at a time. 1920 1921- Artificial delay induction. There are certain types of IO that cannot be 1922 throttled without possibly adversely affecting higher priority groups. This 1923 includes swapping and metadata IO. These types of IO are allowed to occur 1924 normally, however they are "charged" to the originating group. If the 1925 originating group is being throttled you will see the use_delay and delay 1926 fields in io.stat increase. The delay value is how many microseconds that are 1927 being added to any process that runs in this group. Because this number can 1928 grow quite large if there is a lot of swapping or metadata IO occurring we 1929 limit the individual delay events to 1 second at a time. 1930 1931Once the victimized group starts meeting its latency target again it will start 1932unthrottling any peer groups that were throttled previously. If the victimized 1933group simply stops doing IO the global counter will unthrottle appropriately. 1934 1935IO Latency Interface Files 1936~~~~~~~~~~~~~~~~~~~~~~~~~~ 1937 1938 io.latency 1939 This takes a similar format as the other controllers. 1940 1941 "MAJOR:MINOR target=<target time in microseconds>" 1942 1943 io.stat 1944 If the controller is enabled you will see extra stats in io.stat in 1945 addition to the normal ones. 1946 1947 depth 1948 This is the current queue depth for the group. 1949 1950 avg_lat 1951 This is an exponential moving average with a decay rate of 1/exp 1952 bound by the sampling interval. The decay rate interval can be 1953 calculated by multiplying the win value in io.stat by the 1954 corresponding number of samples based on the win value. 1955 1956 win 1957 The sampling window size in milliseconds. This is the minimum 1958 duration of time between evaluation events. Windows only elapse 1959 with IO activity. Idle periods extend the most recent window. 1960 1961IO Priority 1962~~~~~~~~~~~ 1963 1964A single attribute controls the behavior of the I/O priority cgroup policy, 1965namely the blkio.prio.class attribute. The following values are accepted for 1966that attribute: 1967 1968 no-change 1969 Do not modify the I/O priority class. 1970 1971 none-to-rt 1972 For requests that do not have an I/O priority class (NONE), 1973 change the I/O priority class into RT. Do not modify 1974 the I/O priority class of other requests. 1975 1976 restrict-to-be 1977 For requests that do not have an I/O priority class or that have I/O 1978 priority class RT, change it into BE. Do not modify the I/O priority 1979 class of requests that have priority class IDLE. 1980 1981 idle 1982 Change the I/O priority class of all requests into IDLE, the lowest 1983 I/O priority class. 1984 1985The following numerical values are associated with the I/O priority policies: 1986 1987+-------------+---+ 1988| no-change | 0 | 1989+-------------+---+ 1990| none-to-rt | 1 | 1991+-------------+---+ 1992| rt-to-be | 2 | 1993+-------------+---+ 1994| all-to-idle | 3 | 1995+-------------+---+ 1996 1997The numerical value that corresponds to each I/O priority class is as follows: 1998 1999+-------------------------------+---+ 2000| IOPRIO_CLASS_NONE | 0 | 2001+-------------------------------+---+ 2002| IOPRIO_CLASS_RT (real-time) | 1 | 2003+-------------------------------+---+ 2004| IOPRIO_CLASS_BE (best effort) | 2 | 2005+-------------------------------+---+ 2006| IOPRIO_CLASS_IDLE | 3 | 2007+-------------------------------+---+ 2008 2009The algorithm to set the I/O priority class for a request is as follows: 2010 2011- Translate the I/O priority class policy into a number. 2012- Change the request I/O priority class into the maximum of the I/O priority 2013 class policy number and the numerical I/O priority class. 2014 2015PID 2016--- 2017 2018The process number controller is used to allow a cgroup to stop any 2019new tasks from being fork()'d or clone()'d after a specified limit is 2020reached. 2021 2022The number of tasks in a cgroup can be exhausted in ways which other 2023controllers cannot prevent, thus warranting its own controller. For 2024example, a fork bomb is likely to exhaust the number of tasks before 2025hitting memory restrictions. 2026 2027Note that PIDs used in this controller refer to TIDs, process IDs as 2028used by the kernel. 2029 2030 2031PID Interface Files 2032~~~~~~~~~~~~~~~~~~~ 2033 2034 pids.max 2035 A read-write single value file which exists on non-root 2036 cgroups. The default is "max". 2037 2038 Hard limit of number of processes. 2039 2040 pids.current 2041 A read-only single value file which exists on all cgroups. 2042 2043 The number of processes currently in the cgroup and its 2044 descendants. 2045 2046Organisational operations are not blocked by cgroup policies, so it is 2047possible to have pids.current > pids.max. This can be done by either 2048setting the limit to be smaller than pids.current, or attaching enough 2049processes to the cgroup such that pids.current is larger than 2050pids.max. However, it is not possible to violate a cgroup PID policy 2051through fork() or clone(). These will return -EAGAIN if the creation 2052of a new process would cause a cgroup policy to be violated. 2053 2054 2055Cpuset 2056------ 2057 2058The "cpuset" controller provides a mechanism for constraining 2059the CPU and memory node placement of tasks to only the resources 2060specified in the cpuset interface files in a task's current cgroup. 2061This is especially valuable on large NUMA systems where placing jobs 2062on properly sized subsets of the systems with careful processor and 2063memory placement to reduce cross-node memory access and contention 2064can improve overall system performance. 2065 2066The "cpuset" controller is hierarchical. That means the controller 2067cannot use CPUs or memory nodes not allowed in its parent. 2068 2069 2070Cpuset Interface Files 2071~~~~~~~~~~~~~~~~~~~~~~ 2072 2073 cpuset.cpus 2074 A read-write multiple values file which exists on non-root 2075 cpuset-enabled cgroups. 2076 2077 It lists the requested CPUs to be used by tasks within this 2078 cgroup. The actual list of CPUs to be granted, however, is 2079 subjected to constraints imposed by its parent and can differ 2080 from the requested CPUs. 2081 2082 The CPU numbers are comma-separated numbers or ranges. 2083 For example:: 2084 2085 # cat cpuset.cpus 2086 0-4,6,8-10 2087 2088 An empty value indicates that the cgroup is using the same 2089 setting as the nearest cgroup ancestor with a non-empty 2090 "cpuset.cpus" or all the available CPUs if none is found. 2091 2092 The value of "cpuset.cpus" stays constant until the next update 2093 and won't be affected by any CPU hotplug events. 2094 2095 cpuset.cpus.effective 2096 A read-only multiple values file which exists on all 2097 cpuset-enabled cgroups. 2098 2099 It lists the onlined CPUs that are actually granted to this 2100 cgroup by its parent. These CPUs are allowed to be used by 2101 tasks within the current cgroup. 2102 2103 If "cpuset.cpus" is empty, the "cpuset.cpus.effective" file shows 2104 all the CPUs from the parent cgroup that can be available to 2105 be used by this cgroup. Otherwise, it should be a subset of 2106 "cpuset.cpus" unless none of the CPUs listed in "cpuset.cpus" 2107 can be granted. In this case, it will be treated just like an 2108 empty "cpuset.cpus". 2109 2110 Its value will be affected by CPU hotplug events. 2111 2112 cpuset.mems 2113 A read-write multiple values file which exists on non-root 2114 cpuset-enabled cgroups. 2115 2116 It lists the requested memory nodes to be used by tasks within 2117 this cgroup. The actual list of memory nodes granted, however, 2118 is subjected to constraints imposed by its parent and can differ 2119 from the requested memory nodes. 2120 2121 The memory node numbers are comma-separated numbers or ranges. 2122 For example:: 2123 2124 # cat cpuset.mems 2125 0-1,3 2126 2127 An empty value indicates that the cgroup is using the same 2128 setting as the nearest cgroup ancestor with a non-empty 2129 "cpuset.mems" or all the available memory nodes if none 2130 is found. 2131 2132 The value of "cpuset.mems" stays constant until the next update 2133 and won't be affected by any memory nodes hotplug events. 2134 2135 Setting a non-empty value to "cpuset.mems" causes memory of 2136 tasks within the cgroup to be migrated to the designated nodes if 2137 they are currently using memory outside of the designated nodes. 2138 2139 There is a cost for this memory migration. The migration 2140 may not be complete and some memory pages may be left behind. 2141 So it is recommended that "cpuset.mems" should be set properly 2142 before spawning new tasks into the cpuset. Even if there is 2143 a need to change "cpuset.mems" with active tasks, it shouldn't 2144 be done frequently. 2145 2146 cpuset.mems.effective 2147 A read-only multiple values file which exists on all 2148 cpuset-enabled cgroups. 2149 2150 It lists the onlined memory nodes that are actually granted to 2151 this cgroup by its parent. These memory nodes are allowed to 2152 be used by tasks within the current cgroup. 2153 2154 If "cpuset.mems" is empty, it shows all the memory nodes from the 2155 parent cgroup that will be available to be used by this cgroup. 2156 Otherwise, it should be a subset of "cpuset.mems" unless none of 2157 the memory nodes listed in "cpuset.mems" can be granted. In this 2158 case, it will be treated just like an empty "cpuset.mems". 2159 2160 Its value will be affected by memory nodes hotplug events. 2161 2162 cpuset.cpus.partition 2163 A read-write single value file which exists on non-root 2164 cpuset-enabled cgroups. This flag is owned by the parent cgroup 2165 and is not delegatable. 2166 2167 It accepts only the following input values when written to. 2168 2169 ======== ================================ 2170 "root" a partition root 2171 "member" a non-root member of a partition 2172 ======== ================================ 2173 2174 When set to be a partition root, the current cgroup is the 2175 root of a new partition or scheduling domain that comprises 2176 itself and all its descendants except those that are separate 2177 partition roots themselves and their descendants. The root 2178 cgroup is always a partition root. 2179 2180 There are constraints on where a partition root can be set. 2181 It can only be set in a cgroup if all the following conditions 2182 are true. 2183 2184 1) The "cpuset.cpus" is not empty and the list of CPUs are 2185 exclusive, i.e. they are not shared by any of its siblings. 2186 2) The parent cgroup is a partition root. 2187 3) The "cpuset.cpus" is also a proper subset of the parent's 2188 "cpuset.cpus.effective". 2189 4) There is no child cgroups with cpuset enabled. This is for 2190 eliminating corner cases that have to be handled if such a 2191 condition is allowed. 2192 2193 Setting it to partition root will take the CPUs away from the 2194 effective CPUs of the parent cgroup. Once it is set, this 2195 file cannot be reverted back to "member" if there are any child 2196 cgroups with cpuset enabled. 2197 2198 A parent partition cannot distribute all its CPUs to its 2199 child partitions. There must be at least one cpu left in the 2200 parent partition. 2201 2202 Once becoming a partition root, changes to "cpuset.cpus" is 2203 generally allowed as long as the first condition above is true, 2204 the change will not take away all the CPUs from the parent 2205 partition and the new "cpuset.cpus" value is a superset of its 2206 children's "cpuset.cpus" values. 2207 2208 Sometimes, external factors like changes to ancestors' 2209 "cpuset.cpus" or cpu hotplug can cause the state of the partition 2210 root to change. On read, the "cpuset.sched.partition" file 2211 can show the following values. 2212 2213 ============== ============================== 2214 "member" Non-root member of a partition 2215 "root" Partition root 2216 "root invalid" Invalid partition root 2217 ============== ============================== 2218 2219 It is a partition root if the first 2 partition root conditions 2220 above are true and at least one CPU from "cpuset.cpus" is 2221 granted by the parent cgroup. 2222 2223 A partition root can become invalid if none of CPUs requested 2224 in "cpuset.cpus" can be granted by the parent cgroup or the 2225 parent cgroup is no longer a partition root itself. In this 2226 case, it is not a real partition even though the restriction 2227 of the first partition root condition above will still apply. 2228 The cpu affinity of all the tasks in the cgroup will then be 2229 associated with CPUs in the nearest ancestor partition. 2230 2231 An invalid partition root can be transitioned back to a 2232 real partition root if at least one of the requested CPUs 2233 can now be granted by its parent. In this case, the cpu 2234 affinity of all the tasks in the formerly invalid partition 2235 will be associated to the CPUs of the newly formed partition. 2236 Changing the partition state of an invalid partition root to 2237 "member" is always allowed even if child cpusets are present. 2238 2239 2240Device controller 2241----------------- 2242 2243Device controller manages access to device files. It includes both 2244creation of new device files (using mknod), and access to the 2245existing device files. 2246 2247Cgroup v2 device controller has no interface files and is implemented 2248on top of cgroup BPF. To control access to device files, a user may 2249create bpf programs of type BPF_PROG_TYPE_CGROUP_DEVICE and attach 2250them to cgroups with BPF_CGROUP_DEVICE flag. On an attempt to access a 2251device file, corresponding BPF programs will be executed, and depending 2252on the return value the attempt will succeed or fail with -EPERM. 2253 2254A BPF_PROG_TYPE_CGROUP_DEVICE program takes a pointer to the 2255bpf_cgroup_dev_ctx structure, which describes the device access attempt: 2256access type (mknod/read/write) and device (type, major and minor numbers). 2257If the program returns 0, the attempt fails with -EPERM, otherwise it 2258succeeds. 2259 2260An example of BPF_PROG_TYPE_CGROUP_DEVICE program may be found in 2261tools/testing/selftests/bpf/progs/dev_cgroup.c in the kernel source tree. 2262 2263 2264RDMA 2265---- 2266 2267The "rdma" controller regulates the distribution and accounting of 2268RDMA resources. 2269 2270RDMA Interface Files 2271~~~~~~~~~~~~~~~~~~~~ 2272 2273 rdma.max 2274 A readwrite nested-keyed file that exists for all the cgroups 2275 except root that describes current configured resource limit 2276 for a RDMA/IB device. 2277 2278 Lines are keyed by device name and are not ordered. 2279 Each line contains space separated resource name and its configured 2280 limit that can be distributed. 2281 2282 The following nested keys are defined. 2283 2284 ========== ============================= 2285 hca_handle Maximum number of HCA Handles 2286 hca_object Maximum number of HCA Objects 2287 ========== ============================= 2288 2289 An example for mlx4 and ocrdma device follows:: 2290 2291 mlx4_0 hca_handle=2 hca_object=2000 2292 ocrdma1 hca_handle=3 hca_object=max 2293 2294 rdma.current 2295 A read-only file that describes current resource usage. 2296 It exists for all the cgroup except root. 2297 2298 An example for mlx4 and ocrdma device follows:: 2299 2300 mlx4_0 hca_handle=1 hca_object=20 2301 ocrdma1 hca_handle=1 hca_object=23 2302 2303HugeTLB 2304------- 2305 2306The HugeTLB controller allows to limit the HugeTLB usage per control group and 2307enforces the controller limit during page fault. 2308 2309HugeTLB Interface Files 2310~~~~~~~~~~~~~~~~~~~~~~~ 2311 2312 hugetlb.<hugepagesize>.current 2313 Show current usage for "hugepagesize" hugetlb. It exists for all 2314 the cgroup except root. 2315 2316 hugetlb.<hugepagesize>.max 2317 Set/show the hard limit of "hugepagesize" hugetlb usage. 2318 The default value is "max". It exists for all the cgroup except root. 2319 2320 hugetlb.<hugepagesize>.events 2321 A read-only flat-keyed file which exists on non-root cgroups. 2322 2323 max 2324 The number of allocation failure due to HugeTLB limit 2325 2326 hugetlb.<hugepagesize>.events.local 2327 Similar to hugetlb.<hugepagesize>.events but the fields in the file 2328 are local to the cgroup i.e. not hierarchical. The file modified event 2329 generated on this file reflects only the local events. 2330 2331 hugetlb.<hugepagesize>.numa_stat 2332 Similar to memory.numa_stat, it shows the numa information of the 2333 hugetlb pages of <hugepagesize> in this cgroup. Only active in 2334 use hugetlb pages are included. The per-node values are in bytes. 2335 2336Misc 2337---- 2338 2339The Miscellaneous cgroup provides the resource limiting and tracking 2340mechanism for the scalar resources which cannot be abstracted like the other 2341cgroup resources. Controller is enabled by the CONFIG_CGROUP_MISC config 2342option. 2343 2344A resource can be added to the controller via enum misc_res_type{} in the 2345include/linux/misc_cgroup.h file and the corresponding name via misc_res_name[] 2346in the kernel/cgroup/misc.c file. Provider of the resource must set its 2347capacity prior to using the resource by calling misc_cg_set_capacity(). 2348 2349Once a capacity is set then the resource usage can be updated using charge and 2350uncharge APIs. All of the APIs to interact with misc controller are in 2351include/linux/misc_cgroup.h. 2352 2353Misc Interface Files 2354~~~~~~~~~~~~~~~~~~~~ 2355 2356Miscellaneous controller provides 3 interface files. If two misc resources (res_a and res_b) are registered then: 2357 2358 misc.capacity 2359 A read-only flat-keyed file shown only in the root cgroup. It shows 2360 miscellaneous scalar resources available on the platform along with 2361 their quantities:: 2362 2363 $ cat misc.capacity 2364 res_a 50 2365 res_b 10 2366 2367 misc.current 2368 A read-only flat-keyed file shown in the non-root cgroups. It shows 2369 the current usage of the resources in the cgroup and its children.:: 2370 2371 $ cat misc.current 2372 res_a 3 2373 res_b 0 2374 2375 misc.max 2376 A read-write flat-keyed file shown in the non root cgroups. Allowed 2377 maximum usage of the resources in the cgroup and its children.:: 2378 2379 $ cat misc.max 2380 res_a max 2381 res_b 4 2382 2383 Limit can be set by:: 2384 2385 # echo res_a 1 > misc.max 2386 2387 Limit can be set to max by:: 2388 2389 # echo res_a max > misc.max 2390 2391 Limits can be set higher than the capacity value in the misc.capacity 2392 file. 2393 2394 misc.events 2395 A read-only flat-keyed file which exists on non-root cgroups. The 2396 following entries are defined. Unless specified otherwise, a value 2397 change in this file generates a file modified event. All fields in 2398 this file are hierarchical. 2399 2400 max 2401 The number of times the cgroup's resource usage was 2402 about to go over the max boundary. 2403 2404Migration and Ownership 2405~~~~~~~~~~~~~~~~~~~~~~~ 2406 2407A miscellaneous scalar resource is charged to the cgroup in which it is used 2408first, and stays charged to that cgroup until that resource is freed. Migrating 2409a process to a different cgroup does not move the charge to the destination 2410cgroup where the process has moved. 2411 2412Others 2413------ 2414 2415perf_event 2416~~~~~~~~~~ 2417 2418perf_event controller, if not mounted on a legacy hierarchy, is 2419automatically enabled on the v2 hierarchy so that perf events can 2420always be filtered by cgroup v2 path. The controller can still be 2421moved to a legacy hierarchy after v2 hierarchy is populated. 2422 2423 2424Non-normative information 2425------------------------- 2426 2427This section contains information that isn't considered to be a part of 2428the stable kernel API and so is subject to change. 2429 2430 2431CPU controller root cgroup process behaviour 2432~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 2433 2434When distributing CPU cycles in the root cgroup each thread in this 2435cgroup is treated as if it was hosted in a separate child cgroup of the 2436root cgroup. This child cgroup weight is dependent on its thread nice 2437level. 2438 2439For details of this mapping see sched_prio_to_weight array in 2440kernel/sched/core.c file (values from this array should be scaled 2441appropriately so the neutral - nice 0 - value is 100 instead of 1024). 2442 2443 2444IO controller root cgroup process behaviour 2445~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 2446 2447Root cgroup processes are hosted in an implicit leaf child node. 2448When distributing IO resources this implicit child node is taken into 2449account as if it was a normal child cgroup of the root cgroup with a 2450weight value of 200. 2451 2452 2453Namespace 2454========= 2455 2456Basics 2457------ 2458 2459cgroup namespace provides a mechanism to virtualize the view of the 2460"/proc/$PID/cgroup" file and cgroup mounts. The CLONE_NEWCGROUP clone 2461flag can be used with clone(2) and unshare(2) to create a new cgroup 2462namespace. The process running inside the cgroup namespace will have 2463its "/proc/$PID/cgroup" output restricted to cgroupns root. The 2464cgroupns root is the cgroup of the process at the time of creation of 2465the cgroup namespace. 2466 2467Without cgroup namespace, the "/proc/$PID/cgroup" file shows the 2468complete path of the cgroup of a process. In a container setup where 2469a set of cgroups and namespaces are intended to isolate processes the 2470"/proc/$PID/cgroup" file may leak potential system level information 2471to the isolated processes. For example:: 2472 2473 # cat /proc/self/cgroup 2474 0::/batchjobs/container_id1 2475 2476The path '/batchjobs/container_id1' can be considered as system-data 2477and undesirable to expose to the isolated processes. cgroup namespace 2478can be used to restrict visibility of this path. For example, before 2479creating a cgroup namespace, one would see:: 2480 2481 # ls -l /proc/self/ns/cgroup 2482 lrwxrwxrwx 1 root root 0 2014-07-15 10:37 /proc/self/ns/cgroup -> cgroup:[4026531835] 2483 # cat /proc/self/cgroup 2484 0::/batchjobs/container_id1 2485 2486After unsharing a new namespace, the view changes:: 2487 2488 # ls -l /proc/self/ns/cgroup 2489 lrwxrwxrwx 1 root root 0 2014-07-15 10:35 /proc/self/ns/cgroup -> cgroup:[4026532183] 2490 # cat /proc/self/cgroup 2491 0::/ 2492 2493When some thread from a multi-threaded process unshares its cgroup 2494namespace, the new cgroupns gets applied to the entire process (all 2495the threads). This is natural for the v2 hierarchy; however, for the 2496legacy hierarchies, this may be unexpected. 2497 2498A cgroup namespace is alive as long as there are processes inside or 2499mounts pinning it. When the last usage goes away, the cgroup 2500namespace is destroyed. The cgroupns root and the actual cgroups 2501remain. 2502 2503 2504The Root and Views 2505------------------ 2506 2507The 'cgroupns root' for a cgroup namespace is the cgroup in which the 2508process calling unshare(2) is running. For example, if a process in 2509/batchjobs/container_id1 cgroup calls unshare, cgroup 2510/batchjobs/container_id1 becomes the cgroupns root. For the 2511init_cgroup_ns, this is the real root ('/') cgroup. 2512 2513The cgroupns root cgroup does not change even if the namespace creator 2514process later moves to a different cgroup:: 2515 2516 # ~/unshare -c # unshare cgroupns in some cgroup 2517 # cat /proc/self/cgroup 2518 0::/ 2519 # mkdir sub_cgrp_1 2520 # echo 0 > sub_cgrp_1/cgroup.procs 2521 # cat /proc/self/cgroup 2522 0::/sub_cgrp_1 2523 2524Each process gets its namespace-specific view of "/proc/$PID/cgroup" 2525 2526Processes running inside the cgroup namespace will be able to see 2527cgroup paths (in /proc/self/cgroup) only inside their root cgroup. 2528From within an unshared cgroupns:: 2529 2530 # sleep 100000 & 2531 [1] 7353 2532 # echo 7353 > sub_cgrp_1/cgroup.procs 2533 # cat /proc/7353/cgroup 2534 0::/sub_cgrp_1 2535 2536From the initial cgroup namespace, the real cgroup path will be 2537visible:: 2538 2539 $ cat /proc/7353/cgroup 2540 0::/batchjobs/container_id1/sub_cgrp_1 2541 2542From a sibling cgroup namespace (that is, a namespace rooted at a 2543different cgroup), the cgroup path relative to its own cgroup 2544namespace root will be shown. For instance, if PID 7353's cgroup 2545namespace root is at '/batchjobs/container_id2', then it will see:: 2546 2547 # cat /proc/7353/cgroup 2548 0::/../container_id2/sub_cgrp_1 2549 2550Note that the relative path always starts with '/' to indicate that 2551its relative to the cgroup namespace root of the caller. 2552 2553 2554Migration and setns(2) 2555---------------------- 2556 2557Processes inside a cgroup namespace can move into and out of the 2558namespace root if they have proper access to external cgroups. For 2559example, from inside a namespace with cgroupns root at 2560/batchjobs/container_id1, and assuming that the global hierarchy is 2561still accessible inside cgroupns:: 2562 2563 # cat /proc/7353/cgroup 2564 0::/sub_cgrp_1 2565 # echo 7353 > batchjobs/container_id2/cgroup.procs 2566 # cat /proc/7353/cgroup 2567 0::/../container_id2 2568 2569Note that this kind of setup is not encouraged. A task inside cgroup 2570namespace should only be exposed to its own cgroupns hierarchy. 2571 2572setns(2) to another cgroup namespace is allowed when: 2573 2574(a) the process has CAP_SYS_ADMIN against its current user namespace 2575(b) the process has CAP_SYS_ADMIN against the target cgroup 2576 namespace's userns 2577 2578No implicit cgroup changes happen with attaching to another cgroup 2579namespace. It is expected that the someone moves the attaching 2580process under the target cgroup namespace root. 2581 2582 2583Interaction with Other Namespaces 2584--------------------------------- 2585 2586Namespace specific cgroup hierarchy can be mounted by a process 2587running inside a non-init cgroup namespace:: 2588 2589 # mount -t cgroup2 none $MOUNT_POINT 2590 2591This will mount the unified cgroup hierarchy with cgroupns root as the 2592filesystem root. The process needs CAP_SYS_ADMIN against its user and 2593mount namespaces. 2594 2595The virtualization of /proc/self/cgroup file combined with restricting 2596the view of cgroup hierarchy by namespace-private cgroupfs mount 2597provides a properly isolated cgroup view inside the container. 2598 2599 2600Information on Kernel Programming 2601================================= 2602 2603This section contains kernel programming information in the areas 2604where interacting with cgroup is necessary. cgroup core and 2605controllers are not covered. 2606 2607 2608Filesystem Support for Writeback 2609-------------------------------- 2610 2611A filesystem can support cgroup writeback by updating 2612address_space_operations->writepage[s]() to annotate bio's using the 2613following two functions. 2614 2615 wbc_init_bio(@wbc, @bio) 2616 Should be called for each bio carrying writeback data and 2617 associates the bio with the inode's owner cgroup and the 2618 corresponding request queue. This must be called after 2619 a queue (device) has been associated with the bio and 2620 before submission. 2621 2622 wbc_account_cgroup_owner(@wbc, @page, @bytes) 2623 Should be called for each data segment being written out. 2624 While this function doesn't care exactly when it's called 2625 during the writeback session, it's the easiest and most 2626 natural to call it as data segments are added to a bio. 2627 2628With writeback bio's annotated, cgroup support can be enabled per 2629super_block by setting SB_I_CGROUPWB in ->s_iflags. This allows for 2630selective disabling of cgroup writeback support which is helpful when 2631certain filesystem features, e.g. journaled data mode, are 2632incompatible. 2633 2634wbc_init_bio() binds the specified bio to its cgroup. Depending on 2635the configuration, the bio may be executed at a lower priority and if 2636the writeback session is holding shared resources, e.g. a journal 2637entry, may lead to priority inversion. There is no one easy solution 2638for the problem. Filesystems can try to work around specific problem 2639cases by skipping wbc_init_bio() and using bio_associate_blkg() 2640directly. 2641 2642 2643Deprecated v1 Core Features 2644=========================== 2645 2646- Multiple hierarchies including named ones are not supported. 2647 2648- All v1 mount options are not supported. 2649 2650- The "tasks" file is removed and "cgroup.procs" is not sorted. 2651 2652- "cgroup.clone_children" is removed. 2653 2654- /proc/cgroups is meaningless for v2. Use "cgroup.controllers" file 2655 at the root instead. 2656 2657 2658Issues with v1 and Rationales for v2 2659==================================== 2660 2661Multiple Hierarchies 2662-------------------- 2663 2664cgroup v1 allowed an arbitrary number of hierarchies and each 2665hierarchy could host any number of controllers. While this seemed to 2666provide a high level of flexibility, it wasn't useful in practice. 2667 2668For example, as there is only one instance of each controller, utility 2669type controllers such as freezer which can be useful in all 2670hierarchies could only be used in one. The issue is exacerbated by 2671the fact that controllers couldn't be moved to another hierarchy once 2672hierarchies were populated. Another issue was that all controllers 2673bound to a hierarchy were forced to have exactly the same view of the 2674hierarchy. It wasn't possible to vary the granularity depending on 2675the specific controller. 2676 2677In practice, these issues heavily limited which controllers could be 2678put on the same hierarchy and most configurations resorted to putting 2679each controller on its own hierarchy. Only closely related ones, such 2680as the cpu and cpuacct controllers, made sense to be put on the same 2681hierarchy. This often meant that userland ended up managing multiple 2682similar hierarchies repeating the same steps on each hierarchy 2683whenever a hierarchy management operation was necessary. 2684 2685Furthermore, support for multiple hierarchies came at a steep cost. 2686It greatly complicated cgroup core implementation but more importantly 2687the support for multiple hierarchies restricted how cgroup could be 2688used in general and what controllers was able to do. 2689 2690There was no limit on how many hierarchies there might be, which meant 2691that a thread's cgroup membership couldn't be described in finite 2692length. The key might contain any number of entries and was unlimited 2693in length, which made it highly awkward to manipulate and led to 2694addition of controllers which existed only to identify membership, 2695which in turn exacerbated the original problem of proliferating number 2696of hierarchies. 2697 2698Also, as a controller couldn't have any expectation regarding the 2699topologies of hierarchies other controllers might be on, each 2700controller had to assume that all other controllers were attached to 2701completely orthogonal hierarchies. This made it impossible, or at 2702least very cumbersome, for controllers to cooperate with each other. 2703 2704In most use cases, putting controllers on hierarchies which are 2705completely orthogonal to each other isn't necessary. What usually is 2706called for is the ability to have differing levels of granularity 2707depending on the specific controller. In other words, hierarchy may 2708be collapsed from leaf towards root when viewed from specific 2709controllers. For example, a given configuration might not care about 2710how memory is distributed beyond a certain level while still wanting 2711to control how CPU cycles are distributed. 2712 2713 2714Thread Granularity 2715------------------ 2716 2717cgroup v1 allowed threads of a process to belong to different cgroups. 2718This didn't make sense for some controllers and those controllers 2719ended up implementing different ways to ignore such situations but 2720much more importantly it blurred the line between API exposed to 2721individual applications and system management interface. 2722 2723Generally, in-process knowledge is available only to the process 2724itself; thus, unlike service-level organization of processes, 2725categorizing threads of a process requires active participation from 2726the application which owns the target process. 2727 2728cgroup v1 had an ambiguously defined delegation model which got abused 2729in combination with thread granularity. cgroups were delegated to 2730individual applications so that they can create and manage their own 2731sub-hierarchies and control resource distributions along them. This 2732effectively raised cgroup to the status of a syscall-like API exposed 2733to lay programs. 2734 2735First of all, cgroup has a fundamentally inadequate interface to be 2736exposed this way. For a process to access its own knobs, it has to 2737extract the path on the target hierarchy from /proc/self/cgroup, 2738construct the path by appending the name of the knob to the path, open 2739and then read and/or write to it. This is not only extremely clunky 2740and unusual but also inherently racy. There is no conventional way to 2741define transaction across the required steps and nothing can guarantee 2742that the process would actually be operating on its own sub-hierarchy. 2743 2744cgroup controllers implemented a number of knobs which would never be 2745accepted as public APIs because they were just adding control knobs to 2746system-management pseudo filesystem. cgroup ended up with interface 2747knobs which were not properly abstracted or refined and directly 2748revealed kernel internal details. These knobs got exposed to 2749individual applications through the ill-defined delegation mechanism 2750effectively abusing cgroup as a shortcut to implementing public APIs 2751without going through the required scrutiny. 2752 2753This was painful for both userland and kernel. Userland ended up with 2754misbehaving and poorly abstracted interfaces and kernel exposing and 2755locked into constructs inadvertently. 2756 2757 2758Competition Between Inner Nodes and Threads 2759------------------------------------------- 2760 2761cgroup v1 allowed threads to be in any cgroups which created an 2762interesting problem where threads belonging to a parent cgroup and its 2763children cgroups competed for resources. This was nasty as two 2764different types of entities competed and there was no obvious way to 2765settle it. Different controllers did different things. 2766 2767The cpu controller considered threads and cgroups as equivalents and 2768mapped nice levels to cgroup weights. This worked for some cases but 2769fell flat when children wanted to be allocated specific ratios of CPU 2770cycles and the number of internal threads fluctuated - the ratios 2771constantly changed as the number of competing entities fluctuated. 2772There also were other issues. The mapping from nice level to weight 2773wasn't obvious or universal, and there were various other knobs which 2774simply weren't available for threads. 2775 2776The io controller implicitly created a hidden leaf node for each 2777cgroup to host the threads. The hidden leaf had its own copies of all 2778the knobs with ``leaf_`` prefixed. While this allowed equivalent 2779control over internal threads, it was with serious drawbacks. It 2780always added an extra layer of nesting which wouldn't be necessary 2781otherwise, made the interface messy and significantly complicated the 2782implementation. 2783 2784The memory controller didn't have a way to control what happened 2785between internal tasks and child cgroups and the behavior was not 2786clearly defined. There were attempts to add ad-hoc behaviors and 2787knobs to tailor the behavior to specific workloads which would have 2788led to problems extremely difficult to resolve in the long term. 2789 2790Multiple controllers struggled with internal tasks and came up with 2791different ways to deal with it; unfortunately, all the approaches were 2792severely flawed and, furthermore, the widely different behaviors 2793made cgroup as a whole highly inconsistent. 2794 2795This clearly is a problem which needs to be addressed from cgroup core 2796in a uniform way. 2797 2798 2799Other Interface Issues 2800---------------------- 2801 2802cgroup v1 grew without oversight and developed a large number of 2803idiosyncrasies and inconsistencies. One issue on the cgroup core side 2804was how an empty cgroup was notified - a userland helper binary was 2805forked and executed for each event. The event delivery wasn't 2806recursive or delegatable. The limitations of the mechanism also led 2807to in-kernel event delivery filtering mechanism further complicating 2808the interface. 2809 2810Controller interfaces were problematic too. An extreme example is 2811controllers completely ignoring hierarchical organization and treating 2812all cgroups as if they were all located directly under the root 2813cgroup. Some controllers exposed a large amount of inconsistent 2814implementation details to userland. 2815 2816There also was no consistency across controllers. When a new cgroup 2817was created, some controllers defaulted to not imposing extra 2818restrictions while others disallowed any resource usage until 2819explicitly configured. Configuration knobs for the same type of 2820control used widely differing naming schemes and formats. Statistics 2821and information knobs were named arbitrarily and used different 2822formats and units even in the same controller. 2823 2824cgroup v2 establishes common conventions where appropriate and updates 2825controllers so that they expose minimal and consistent interfaces. 2826 2827 2828Controller Issues and Remedies 2829------------------------------ 2830 2831Memory 2832~~~~~~ 2833 2834The original lower boundary, the soft limit, is defined as a limit 2835that is per default unset. As a result, the set of cgroups that 2836global reclaim prefers is opt-in, rather than opt-out. The costs for 2837optimizing these mostly negative lookups are so high that the 2838implementation, despite its enormous size, does not even provide the 2839basic desirable behavior. First off, the soft limit has no 2840hierarchical meaning. All configured groups are organized in a global 2841rbtree and treated like equal peers, regardless where they are located 2842in the hierarchy. This makes subtree delegation impossible. Second, 2843the soft limit reclaim pass is so aggressive that it not just 2844introduces high allocation latencies into the system, but also impacts 2845system performance due to overreclaim, to the point where the feature 2846becomes self-defeating. 2847 2848The memory.low boundary on the other hand is a top-down allocated 2849reserve. A cgroup enjoys reclaim protection when it's within its 2850effective low, which makes delegation of subtrees possible. It also 2851enjoys having reclaim pressure proportional to its overage when 2852above its effective low. 2853 2854The original high boundary, the hard limit, is defined as a strict 2855limit that can not budge, even if the OOM killer has to be called. 2856But this generally goes against the goal of making the most out of the 2857available memory. The memory consumption of workloads varies during 2858runtime, and that requires users to overcommit. But doing that with a 2859strict upper limit requires either a fairly accurate prediction of the 2860working set size or adding slack to the limit. Since working set size 2861estimation is hard and error prone, and getting it wrong results in 2862OOM kills, most users tend to err on the side of a looser limit and 2863end up wasting precious resources. 2864 2865The memory.high boundary on the other hand can be set much more 2866conservatively. When hit, it throttles allocations by forcing them 2867into direct reclaim to work off the excess, but it never invokes the 2868OOM killer. As a result, a high boundary that is chosen too 2869aggressively will not terminate the processes, but instead it will 2870lead to gradual performance degradation. The user can monitor this 2871and make corrections until the minimal memory footprint that still 2872gives acceptable performance is found. 2873 2874In extreme cases, with many concurrent allocations and a complete 2875breakdown of reclaim progress within the group, the high boundary can 2876be exceeded. But even then it's mostly better to satisfy the 2877allocation from the slack available in other groups or the rest of the 2878system than killing the group. Otherwise, memory.max is there to 2879limit this type of spillover and ultimately contain buggy or even 2880malicious applications. 2881 2882Setting the original memory.limit_in_bytes below the current usage was 2883subject to a race condition, where concurrent charges could cause the 2884limit setting to fail. memory.max on the other hand will first set the 2885limit to prevent new charges, and then reclaim and OOM kill until the 2886new limit is met - or the task writing to memory.max is killed. 2887 2888The combined memory+swap accounting and limiting is replaced by real 2889control over swap space. 2890 2891The main argument for a combined memory+swap facility in the original 2892cgroup design was that global or parental pressure would always be 2893able to swap all anonymous memory of a child group, regardless of the 2894child's own (possibly untrusted) configuration. However, untrusted 2895groups can sabotage swapping by other means - such as referencing its 2896anonymous memory in a tight loop - and an admin can not assume full 2897swappability when overcommitting untrusted jobs. 2898 2899For trusted jobs, on the other hand, a combined counter is not an 2900intuitive userspace interface, and it flies in the face of the idea 2901that cgroup controllers should account and limit specific physical 2902resources. Swap space is a resource like all others in the system, 2903and that's why unified hierarchy allows distributing it separately. 2904