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