Version:  2.0.40 2.2.26 2.4.37 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 4.10

Linux/Documentation/cgroup-v2.txt

  1 
  2 Control Group v2
  3 
  4 October, 2015           Tejun Heo <tj@kernel.org>
  5 
  6 This is the authoritative documentation on the design, interface and
  7 conventions of cgroup v2.  It describes all userland-visible aspects
  8 of cgroup including core and specific controller behaviors.  All
  9 future changes must be reflected in this document.  Documentation for
 10 v1 is available under Documentation/cgroup-v1/.
 11 
 12 CONTENTS
 13 
 14 1. Introduction
 15   1-1. Terminology
 16   1-2. What is cgroup?
 17 2. Basic Operations
 18   2-1. Mounting
 19   2-2. Organizing Processes
 20   2-3. [Un]populated Notification
 21   2-4. Controlling Controllers
 22     2-4-1. Enabling and Disabling
 23     2-4-2. Top-down Constraint
 24     2-4-3. No Internal Process Constraint
 25   2-5. Delegation
 26     2-5-1. Model of Delegation
 27     2-5-2. Delegation Containment
 28   2-6. Guidelines
 29     2-6-1. Organize Once and Control
 30     2-6-2. Avoid Name Collisions
 31 3. Resource Distribution Models
 32   3-1. Weights
 33   3-2. Limits
 34   3-3. Protections
 35   3-4. Allocations
 36 4. Interface Files
 37   4-1. Format
 38   4-2. Conventions
 39   4-3. Core Interface Files
 40 5. Controllers
 41   5-1. CPU
 42     5-1-1. CPU Interface Files
 43   5-2. Memory
 44     5-2-1. Memory Interface Files
 45     5-2-2. Usage Guidelines
 46     5-2-3. Memory Ownership
 47   5-3. IO
 48     5-3-1. IO Interface Files
 49     5-3-2. Writeback
 50 6. Namespace
 51   6-1. Basics
 52   6-2. The Root and Views
 53   6-3. Migration and setns(2)
 54   6-4. Interaction with Other Namespaces
 55 P. Information on Kernel Programming
 56   P-1. Filesystem Support for Writeback
 57 D. Deprecated v1 Core Features
 58 R. Issues with v1 and Rationales for v2
 59   R-1. Multiple Hierarchies
 60   R-2. Thread Granularity
 61   R-3. Competition Between Inner Nodes and Threads
 62   R-4. Other Interface Issues
 63   R-5. Controller Issues and Remedies
 64     R-5-1. Memory
 65 
 66 
 67 1. Introduction
 68 
 69 1-1. Terminology
 70 
 71 "cgroup" stands for "control group" and is never capitalized.  The
 72 singular form is used to designate the whole feature and also as a
 73 qualifier as in "cgroup controllers".  When explicitly referring to
 74 multiple individual control groups, the plural form "cgroups" is used.
 75 
 76 
 77 1-2. What is cgroup?
 78 
 79 cgroup is a mechanism to organize processes hierarchically and
 80 distribute system resources along the hierarchy in a controlled and
 81 configurable manner.
 82 
 83 cgroup is largely composed of two parts - the core and controllers.
 84 cgroup core is primarily responsible for hierarchically organizing
 85 processes.  A cgroup controller is usually responsible for
 86 distributing a specific type of system resource along the hierarchy
 87 although there are utility controllers which serve purposes other than
 88 resource distribution.
 89 
 90 cgroups form a tree structure and every process in the system belongs
 91 to one and only one cgroup.  All threads of a process belong to the
 92 same cgroup.  On creation, all processes are put in the cgroup that
 93 the parent process belongs to at the time.  A process can be migrated
 94 to another cgroup.  Migration of a process doesn't affect already
 95 existing descendant processes.
 96 
 97 Following certain structural constraints, controllers may be enabled or
 98 disabled selectively on a cgroup.  All controller behaviors are
 99 hierarchical - if a controller is enabled on a cgroup, it affects all
100 processes which belong to the cgroups consisting the inclusive
101 sub-hierarchy of the cgroup.  When a controller is enabled on a nested
102 cgroup, it always restricts the resource distribution further.  The
103 restrictions set closer to the root in the hierarchy can not be
104 overridden from further away.
105 
106 
107 2. Basic Operations
108 
109 2-1. Mounting
110 
111 Unlike v1, cgroup v2 has only single hierarchy.  The cgroup v2
112 hierarchy can be mounted with the following mount command.
113 
114   # mount -t cgroup2 none $MOUNT_POINT
115 
116 cgroup2 filesystem has the magic number 0x63677270 ("cgrp").  All
117 controllers which support v2 and are not bound to a v1 hierarchy are
118 automatically bound to the v2 hierarchy and show up at the root.
119 Controllers which are not in active use in the v2 hierarchy can be
120 bound to other hierarchies.  This allows mixing v2 hierarchy with the
121 legacy v1 multiple hierarchies in a fully backward compatible way.
122 
123 A controller can be moved across hierarchies only after the controller
124 is no longer referenced in its current hierarchy.  Because per-cgroup
125 controller states are destroyed asynchronously and controllers may
126 have lingering references, a controller may not show up immediately on
127 the v2 hierarchy after the final umount of the previous hierarchy.
128 Similarly, a controller should be fully disabled to be moved out of
129 the unified hierarchy and it may take some time for the disabled
130 controller to become available for other hierarchies; furthermore, due
131 to inter-controller dependencies, other controllers may need to be
132 disabled too.
133 
134 While useful for development and manual configurations, moving
135 controllers dynamically between the v2 and other hierarchies is
136 strongly discouraged for production use.  It is recommended to decide
137 the hierarchies and controller associations before starting using the
138 controllers after system boot.
139 
140 During transition to v2, system management software might still
141 automount the v1 cgroup filesystem and so hijack all controllers
142 during boot, before manual intervention is possible. To make testing
143 and experimenting easier, the kernel parameter cgroup_no_v1= allows
144 disabling controllers in v1 and make them always available in v2.
145 
146 
147 2-2. Organizing Processes
148 
149 Initially, only the root cgroup exists to which all processes belong.
150 A child cgroup can be created by creating a sub-directory.
151 
152   # mkdir $CGROUP_NAME
153 
154 A given cgroup may have multiple child cgroups forming a tree
155 structure.  Each cgroup has a read-writable interface file
156 "cgroup.procs".  When read, it lists the PIDs of all processes which
157 belong to the cgroup one-per-line.  The PIDs are not ordered and the
158 same PID may show up more than once if the process got moved to
159 another cgroup and then back or the PID got recycled while reading.
160 
161 A process can be migrated into a cgroup by writing its PID to the
162 target cgroup's "cgroup.procs" file.  Only one process can be migrated
163 on a single write(2) call.  If a process is composed of multiple
164 threads, writing the PID of any thread migrates all threads of the
165 process.
166 
167 When a process forks a child process, the new process is born into the
168 cgroup that the forking process belongs to at the time of the
169 operation.  After exit, a process stays associated with the cgroup
170 that it belonged to at the time of exit until it's reaped; however, a
171 zombie process does not appear in "cgroup.procs" and thus can't be
172 moved to another cgroup.
173 
174 A cgroup which doesn't have any children or live processes can be
175 destroyed by removing the directory.  Note that a cgroup which doesn't
176 have any children and is associated only with zombie processes is
177 considered empty and can be removed.
178 
179   # rmdir $CGROUP_NAME
180 
181 "/proc/$PID/cgroup" lists a process's cgroup membership.  If legacy
182 cgroup is in use in the system, this file may contain multiple lines,
183 one for each hierarchy.  The entry for cgroup v2 is always in the
184 format "0::$PATH".
185 
186   # cat /proc/842/cgroup
187   ...
188   0::/test-cgroup/test-cgroup-nested
189 
190 If the process becomes a zombie and the cgroup it was associated with
191 is removed subsequently, " (deleted)" is appended to the path.
192 
193   # cat /proc/842/cgroup
194   ...
195   0::/test-cgroup/test-cgroup-nested (deleted)
196 
197 
198 2-3. [Un]populated Notification
199 
200 Each non-root cgroup has a "cgroup.events" file which contains
201 "populated" field indicating whether the cgroup's sub-hierarchy has
202 live processes in it.  Its value is 0 if there is no live process in
203 the cgroup and its descendants; otherwise, 1.  poll and [id]notify
204 events are triggered when the value changes.  This can be used, for
205 example, to start a clean-up operation after all processes of a given
206 sub-hierarchy have exited.  The populated state updates and
207 notifications are recursive.  Consider the following sub-hierarchy
208 where the numbers in the parentheses represent the numbers of processes
209 in each cgroup.
210 
211   A(4) - B(0) - C(1)
212               \ D(0)
213 
214 A, B and C's "populated" fields would be 1 while D's 0.  After the one
215 process in C exits, B and C's "populated" fields would flip to "0" and
216 file modified events will be generated on the "cgroup.events" files of
217 both cgroups.
218 
219 
220 2-4. Controlling Controllers
221 
222 2-4-1. Enabling and Disabling
223 
224 Each cgroup has a "cgroup.controllers" file which lists all
225 controllers available for the cgroup to enable.
226 
227   # cat cgroup.controllers
228   cpu io memory
229 
230 No controller is enabled by default.  Controllers can be enabled and
231 disabled by writing to the "cgroup.subtree_control" file.
232 
233   # echo "+cpu +memory -io" > cgroup.subtree_control
234 
235 Only controllers which are listed in "cgroup.controllers" can be
236 enabled.  When multiple operations are specified as above, either they
237 all succeed or fail.  If multiple operations on the same controller
238 are specified, the last one is effective.
239 
240 Enabling a controller in a cgroup indicates that the distribution of
241 the target resource across its immediate children will be controlled.
242 Consider the following sub-hierarchy.  The enabled controllers are
243 listed in parentheses.
244 
245   A(cpu,memory) - B(memory) - C()
246                             \ D()
247 
248 As A has "cpu" and "memory" enabled, A will control the distribution
249 of CPU cycles and memory to its children, in this case, B.  As B has
250 "memory" enabled but not "CPU", C and D will compete freely on CPU
251 cycles but their division of memory available to B will be controlled.
252 
253 As a controller regulates the distribution of the target resource to
254 the cgroup's children, enabling it creates the controller's interface
255 files in the child cgroups.  In the above example, enabling "cpu" on B
256 would create the "cpu." prefixed controller interface files in C and
257 D.  Likewise, disabling "memory" from B would remove the "memory."
258 prefixed controller interface files from C and D.  This means that the
259 controller interface files - anything which doesn't start with
260 "cgroup." are owned by the parent rather than the cgroup itself.
261 
262 
263 2-4-2. Top-down Constraint
264 
265 Resources are distributed top-down and a cgroup can further distribute
266 a resource only if the resource has been distributed to it from the
267 parent.  This means that all non-root "cgroup.subtree_control" files
268 can only contain controllers which are enabled in the parent's
269 "cgroup.subtree_control" file.  A controller can be enabled only if
270 the parent has the controller enabled and a controller can't be
271 disabled if one or more children have it enabled.
272 
273 
274 2-4-3. No Internal Process Constraint
275 
276 Non-root cgroups can only distribute resources to their children when
277 they don't have any processes of their own.  In other words, only
278 cgroups which don't contain any processes can have controllers enabled
279 in their "cgroup.subtree_control" files.
280 
281 This guarantees that, when a controller is looking at the part of the
282 hierarchy which has it enabled, processes are always only on the
283 leaves.  This rules out situations where child cgroups compete against
284 internal processes of the parent.
285 
286 The root cgroup is exempt from this restriction.  Root contains
287 processes and anonymous resource consumption which can't be associated
288 with any other cgroups and requires special treatment from most
289 controllers.  How resource consumption in the root cgroup is governed
290 is up to each controller.
291 
292 Note that the restriction doesn't get in the way if there is no
293 enabled controller in the cgroup's "cgroup.subtree_control".  This is
294 important as otherwise it wouldn't be possible to create children of a
295 populated cgroup.  To control resource distribution of a cgroup, the
296 cgroup must create children and transfer all its processes to the
297 children before enabling controllers in its "cgroup.subtree_control"
298 file.
299 
300 
301 2-5. Delegation
302 
303 2-5-1. Model of Delegation
304 
305 A cgroup can be delegated to a less privileged user by granting write
306 access of the directory and its "cgroup.procs" file to the user.  Note
307 that resource control interface files in a given directory control the
308 distribution of the parent's resources and thus must not be delegated
309 along with the directory.
310 
311 Once delegated, the user can build sub-hierarchy under the directory,
312 organize processes as it sees fit and further distribute the resources
313 it received from the parent.  The limits and other settings of all
314 resource controllers are hierarchical and regardless of what happens
315 in the delegated sub-hierarchy, nothing can escape the resource
316 restrictions imposed by the parent.
317 
318 Currently, cgroup doesn't impose any restrictions on the number of
319 cgroups in or nesting depth of a delegated sub-hierarchy; however,
320 this may be limited explicitly in the future.
321 
322 
323 2-5-2. Delegation Containment
324 
325 A delegated sub-hierarchy is contained in the sense that processes
326 can't be moved into or out of the sub-hierarchy by the delegatee.  For
327 a process with a non-root euid to migrate a target process into a
328 cgroup by writing its PID to the "cgroup.procs" file, the following
329 conditions must be met.
330 
331 - The writer's euid must match either uid or suid of the target process.
332 
333 - The writer must have write access to the "cgroup.procs" file.
334 
335 - The writer must have write access to the "cgroup.procs" file of the
336   common ancestor of the source and destination cgroups.
337 
338 The above three constraints ensure that while a delegatee may migrate
339 processes around freely in the delegated sub-hierarchy it can't pull
340 in from or push out to outside the sub-hierarchy.
341 
342 For an example, let's assume cgroups C0 and C1 have been delegated to
343 user U0 who created C00, C01 under C0 and C10 under C1 as follows and
344 all processes under C0 and C1 belong to U0.
345 
346   ~~~~~~~~~~~~~ - C0 - C00
347   ~ cgroup    ~      \ C01
348   ~ hierarchy ~
349   ~~~~~~~~~~~~~ - C1 - C10
350 
351 Let's also say U0 wants to write the PID of a process which is
352 currently in C10 into "C00/cgroup.procs".  U0 has write access to the
353 file and uid match on the process; however, the common ancestor of the
354 source cgroup C10 and the destination cgroup C00 is above the points
355 of delegation and U0 would not have write access to its "cgroup.procs"
356 files and thus the write will be denied with -EACCES.
357 
358 
359 2-6. Guidelines
360 
361 2-6-1. Organize Once and Control
362 
363 Migrating a process across cgroups is a relatively expensive operation
364 and stateful resources such as memory are not moved together with the
365 process.  This is an explicit design decision as there often exist
366 inherent trade-offs between migration and various hot paths in terms
367 of synchronization cost.
368 
369 As such, migrating processes across cgroups frequently as a means to
370 apply different resource restrictions is discouraged.  A workload
371 should be assigned to a cgroup according to the system's logical and
372 resource structure once on start-up.  Dynamic adjustments to resource
373 distribution can be made by changing controller configuration through
374 the interface files.
375 
376 
377 2-6-2. Avoid Name Collisions
378 
379 Interface files for a cgroup and its children cgroups occupy the same
380 directory and it is possible to create children cgroups which collide
381 with interface files.
382 
383 All cgroup core interface files are prefixed with "cgroup." and each
384 controller's interface files are prefixed with the controller name and
385 a dot.  A controller's name is composed of lower case alphabets and
386 '_'s but never begins with an '_' so it can be used as the prefix
387 character for collision avoidance.  Also, interface file names won't
388 start or end with terms which are often used in categorizing workloads
389 such as job, service, slice, unit or workload.
390 
391 cgroup doesn't do anything to prevent name collisions and it's the
392 user's responsibility to avoid them.
393 
394 
395 3. Resource Distribution Models
396 
397 cgroup controllers implement several resource distribution schemes
398 depending on the resource type and expected use cases.  This section
399 describes major schemes in use along with their expected behaviors.
400 
401 
402 3-1. Weights
403 
404 A parent's resource is distributed by adding up the weights of all
405 active children and giving each the fraction matching the ratio of its
406 weight against the sum.  As only children which can make use of the
407 resource at the moment participate in the distribution, this is
408 work-conserving.  Due to the dynamic nature, this model is usually
409 used for stateless resources.
410 
411 All weights are in the range [1, 10000] with the default at 100.  This
412 allows symmetric multiplicative biases in both directions at fine
413 enough granularity while staying in the intuitive range.
414 
415 As long as the weight is in range, all configuration combinations are
416 valid and there is no reason to reject configuration changes or
417 process migrations.
418 
419 "cpu.weight" proportionally distributes CPU cycles to active children
420 and is an example of this type.
421 
422 
423 3-2. Limits
424 
425 A child can only consume upto the configured amount of the resource.
426 Limits can be over-committed - the sum of the limits of children can
427 exceed the amount of resource available to the parent.
428 
429 Limits are in the range [0, max] and defaults to "max", which is noop.
430 
431 As limits can be over-committed, all configuration combinations are
432 valid and there is no reason to reject configuration changes or
433 process migrations.
434 
435 "io.max" limits the maximum BPS and/or IOPS that a cgroup can consume
436 on an IO device and is an example of this type.
437 
438 
439 3-3. Protections
440 
441 A cgroup is protected to be allocated upto the configured amount of
442 the resource if the usages of all its ancestors are under their
443 protected levels.  Protections can be hard guarantees or best effort
444 soft boundaries.  Protections can also be over-committed in which case
445 only upto the amount available to the parent is protected among
446 children.
447 
448 Protections are in the range [0, max] and defaults to 0, which is
449 noop.
450 
451 As protections can be over-committed, all configuration combinations
452 are valid and there is no reason to reject configuration changes or
453 process migrations.
454 
455 "memory.low" implements best-effort memory protection and is an
456 example of this type.
457 
458 
459 3-4. Allocations
460 
461 A cgroup is exclusively allocated a certain amount of a finite
462 resource.  Allocations can't be over-committed - the sum of the
463 allocations of children can not exceed the amount of resource
464 available to the parent.
465 
466 Allocations are in the range [0, max] and defaults to 0, which is no
467 resource.
468 
469 As allocations can't be over-committed, some configuration
470 combinations are invalid and should be rejected.  Also, if the
471 resource is mandatory for execution of processes, process migrations
472 may be rejected.
473 
474 "cpu.rt.max" hard-allocates realtime slices and is an example of this
475 type.
476 
477 
478 4. Interface Files
479 
480 4-1. Format
481 
482 All interface files should be in one of the following formats whenever
483 possible.
484 
485   New-line separated values
486   (when only one value can be written at once)
487 
488         VAL0\n
489         VAL1\n
490         ...
491 
492   Space separated values
493   (when read-only or multiple values can be written at once)
494 
495         VAL0 VAL1 ...\n
496 
497   Flat keyed
498 
499         KEY0 VAL0\n
500         KEY1 VAL1\n
501         ...
502 
503   Nested keyed
504 
505         KEY0 SUB_KEY0=VAL00 SUB_KEY1=VAL01...
506         KEY1 SUB_KEY0=VAL10 SUB_KEY1=VAL11...
507         ...
508 
509 For a writable file, the format for writing should generally match
510 reading; however, controllers may allow omitting later fields or
511 implement restricted shortcuts for most common use cases.
512 
513 For both flat and nested keyed files, only the values for a single key
514 can be written at a time.  For nested keyed files, the sub key pairs
515 may be specified in any order and not all pairs have to be specified.
516 
517 
518 4-2. Conventions
519 
520 - Settings for a single feature should be contained in a single file.
521 
522 - The root cgroup should be exempt from resource control and thus
523   shouldn't have resource control interface files.  Also,
524   informational files on the root cgroup which end up showing global
525   information available elsewhere shouldn't exist.
526 
527 - If a controller implements weight based resource distribution, its
528   interface file should be named "weight" and have the range [1,
529   10000] with 100 as the default.  The values are chosen to allow
530   enough and symmetric bias in both directions while keeping it
531   intuitive (the default is 100%).
532 
533 - If a controller implements an absolute resource guarantee and/or
534   limit, the interface files should be named "min" and "max"
535   respectively.  If a controller implements best effort resource
536   guarantee and/or limit, the interface files should be named "low"
537   and "high" respectively.
538 
539   In the above four control files, the special token "max" should be
540   used to represent upward infinity for both reading and writing.
541 
542 - If a setting has a configurable default value and keyed specific
543   overrides, the default entry should be keyed with "default" and
544   appear as the first entry in the file.
545 
546   The default value can be updated by writing either "default $VAL" or
547   "$VAL".
548 
549   When writing to update a specific override, "default" can be used as
550   the value to indicate removal of the override.  Override entries
551   with "default" as the value must not appear when read.
552 
553   For example, a setting which is keyed by major:minor device numbers
554   with integer values may look like the following.
555 
556     # cat cgroup-example-interface-file
557     default 150
558     8:0 300
559 
560   The default value can be updated by
561 
562     # echo 125 > cgroup-example-interface-file
563 
564   or
565 
566     # echo "default 125" > cgroup-example-interface-file
567 
568   An override can be set by
569 
570     # echo "8:16 170" > cgroup-example-interface-file
571 
572   and cleared by
573 
574     # echo "8:0 default" > cgroup-example-interface-file
575     # cat cgroup-example-interface-file
576     default 125
577     8:16 170
578 
579 - For events which are not very high frequency, an interface file
580   "events" should be created which lists event key value pairs.
581   Whenever a notifiable event happens, file modified event should be
582   generated on the file.
583 
584 
585 4-3. Core Interface Files
586 
587 All cgroup core files are prefixed with "cgroup."
588 
589   cgroup.procs
590 
591         A read-write new-line separated values file which exists on
592         all cgroups.
593 
594         When read, it lists the PIDs of all processes which belong to
595         the cgroup one-per-line.  The PIDs are not ordered and the
596         same PID may show up more than once if the process got moved
597         to another cgroup and then back or the PID got recycled while
598         reading.
599 
600         A PID can be written to migrate the process associated with
601         the PID to the cgroup.  The writer should match all of the
602         following conditions.
603 
604         - Its euid is either root or must match either uid or suid of
605           the target process.
606 
607         - It must have write access to the "cgroup.procs" file.
608 
609         - It must have write access to the "cgroup.procs" file of the
610           common ancestor of the source and destination cgroups.
611 
612         When delegating a sub-hierarchy, write access to this file
613         should be granted along with the containing directory.
614 
615   cgroup.controllers
616 
617         A read-only space separated values file which exists on all
618         cgroups.
619 
620         It shows space separated list of all controllers available to
621         the cgroup.  The controllers are not ordered.
622 
623   cgroup.subtree_control
624 
625         A read-write space separated values file which exists on all
626         cgroups.  Starts out empty.
627 
628         When read, it shows space separated list of the controllers
629         which are enabled to control resource distribution from the
630         cgroup to its children.
631 
632         Space separated list of controllers prefixed with '+' or '-'
633         can be written to enable or disable controllers.  A controller
634         name prefixed with '+' enables the controller and '-'
635         disables.  If a controller appears more than once on the list,
636         the last one is effective.  When multiple enable and disable
637         operations are specified, either all succeed or all fail.
638 
639   cgroup.events
640 
641         A read-only flat-keyed file which exists on non-root cgroups.
642         The following entries are defined.  Unless specified
643         otherwise, a value change in this file generates a file
644         modified event.
645 
646           populated
647 
648                 1 if the cgroup or its descendants contains any live
649                 processes; otherwise, 0.
650 
651 
652 5. Controllers
653 
654 5-1. CPU
655 
656 [NOTE: The interface for the cpu controller hasn't been merged yet]
657 
658 The "cpu" controllers regulates distribution of CPU cycles.  This
659 controller implements weight and absolute bandwidth limit models for
660 normal scheduling policy and absolute bandwidth allocation model for
661 realtime scheduling policy.
662 
663 
664 5-1-1. CPU Interface Files
665 
666 All time durations are in microseconds.
667 
668   cpu.stat
669 
670         A read-only flat-keyed file which exists on non-root cgroups.
671 
672         It reports the following six stats.
673 
674           usage_usec
675           user_usec
676           system_usec
677           nr_periods
678           nr_throttled
679           throttled_usec
680 
681   cpu.weight
682 
683         A read-write single value file which exists on non-root
684         cgroups.  The default is "100".
685 
686         The weight in the range [1, 10000].
687 
688   cpu.max
689 
690         A read-write two value file which exists on non-root cgroups.
691         The default is "max 100000".
692 
693         The maximum bandwidth limit.  It's in the following format.
694 
695           $MAX $PERIOD
696 
697         which indicates that the group may consume upto $MAX in each
698         $PERIOD duration.  "max" for $MAX indicates no limit.  If only
699         one number is written, $MAX is updated.
700 
701   cpu.rt.max
702 
703   [NOTE: The semantics of this file is still under discussion and the
704    interface hasn't been merged yet]
705 
706         A read-write two value file which exists on all cgroups.
707         The default is "0 100000".
708 
709         The maximum realtime runtime allocation.  Over-committing
710         configurations are disallowed and process migrations are
711         rejected if not enough bandwidth is available.  It's in the
712         following format.
713 
714           $MAX $PERIOD
715 
716         which indicates that the group may consume upto $MAX in each
717         $PERIOD duration.  If only one number is written, $MAX is
718         updated.
719 
720 
721 5-2. Memory
722 
723 The "memory" controller regulates distribution of memory.  Memory is
724 stateful and implements both limit and protection models.  Due to the
725 intertwining between memory usage and reclaim pressure and the
726 stateful nature of memory, the distribution model is relatively
727 complex.
728 
729 While not completely water-tight, all major memory usages by a given
730 cgroup are tracked so that the total memory consumption can be
731 accounted and controlled to a reasonable extent.  Currently, the
732 following types of memory usages are tracked.
733 
734 - Userland memory - page cache and anonymous memory.
735 
736 - Kernel data structures such as dentries and inodes.
737 
738 - TCP socket buffers.
739 
740 The above list may expand in the future for better coverage.
741 
742 
743 5-2-1. Memory Interface Files
744 
745 All memory amounts are in bytes.  If a value which is not aligned to
746 PAGE_SIZE is written, the value may be rounded up to the closest
747 PAGE_SIZE multiple when read back.
748 
749   memory.current
750 
751         A read-only single value file which exists on non-root
752         cgroups.
753 
754         The total amount of memory currently being used by the cgroup
755         and its descendants.
756 
757   memory.low
758 
759         A read-write single value file which exists on non-root
760         cgroups.  The default is "0".
761 
762         Best-effort memory protection.  If the memory usages of a
763         cgroup and all its ancestors are below their low boundaries,
764         the cgroup's memory won't be reclaimed unless memory can be
765         reclaimed from unprotected cgroups.
766 
767         Putting more memory than generally available under this
768         protection is discouraged.
769 
770   memory.high
771 
772         A read-write single value file which exists on non-root
773         cgroups.  The default is "max".
774 
775         Memory usage throttle limit.  This is the main mechanism to
776         control memory usage of a cgroup.  If a cgroup's usage goes
777         over the high boundary, the processes of the cgroup are
778         throttled and put under heavy reclaim pressure.
779 
780         Going over the high limit never invokes the OOM killer and
781         under extreme conditions the limit may be breached.
782 
783   memory.max
784 
785         A read-write single value file which exists on non-root
786         cgroups.  The default is "max".
787 
788         Memory usage hard limit.  This is the final protection
789         mechanism.  If a cgroup's memory usage reaches this limit and
790         can't be reduced, the OOM killer is invoked in the cgroup.
791         Under certain circumstances, the usage may go over the limit
792         temporarily.
793 
794         This is the ultimate protection mechanism.  As long as the
795         high limit is used and monitored properly, this limit's
796         utility is limited to providing the final safety net.
797 
798   memory.events
799 
800         A read-only flat-keyed file which exists on non-root cgroups.
801         The following entries are defined.  Unless specified
802         otherwise, a value change in this file generates a file
803         modified event.
804 
805           low
806 
807                 The number of times the cgroup is reclaimed due to
808                 high memory pressure even though its usage is under
809                 the low boundary.  This usually indicates that the low
810                 boundary is over-committed.
811 
812           high
813 
814                 The number of times processes of the cgroup are
815                 throttled and routed to perform direct memory reclaim
816                 because the high memory boundary was exceeded.  For a
817                 cgroup whose memory usage is capped by the high limit
818                 rather than global memory pressure, this event's
819                 occurrences are expected.
820 
821           max
822 
823                 The number of times the cgroup's memory usage was
824                 about to go over the max boundary.  If direct reclaim
825                 fails to bring it down, the OOM killer is invoked.
826 
827           oom
828 
829                 The number of times the OOM killer has been invoked in
830                 the cgroup.  This may not exactly match the number of
831                 processes killed but should generally be close.
832 
833   memory.stat
834 
835         A read-only flat-keyed file which exists on non-root cgroups.
836 
837         This breaks down the cgroup's memory footprint into different
838         types of memory, type-specific details, and other information
839         on the state and past events of the memory management system.
840 
841         All memory amounts are in bytes.
842 
843         The entries are ordered to be human readable, and new entries
844         can show up in the middle. Don't rely on items remaining in a
845         fixed position; use the keys to look up specific values!
846 
847           anon
848 
849                 Amount of memory used in anonymous mappings such as
850                 brk(), sbrk(), and mmap(MAP_ANONYMOUS)
851 
852           file
853 
854                 Amount of memory used to cache filesystem data,
855                 including tmpfs and shared memory.
856 
857           kernel_stack
858 
859                 Amount of memory allocated to kernel stacks.
860 
861           slab
862 
863                 Amount of memory used for storing in-kernel data
864                 structures.
865 
866           sock
867 
868                 Amount of memory used in network transmission buffers
869 
870           file_mapped
871 
872                 Amount of cached filesystem data mapped with mmap()
873 
874           file_dirty
875 
876                 Amount of cached filesystem data that was modified but
877                 not yet written back to disk
878 
879           file_writeback
880 
881                 Amount of cached filesystem data that was modified and
882                 is currently being written back to disk
883 
884           inactive_anon
885           active_anon
886           inactive_file
887           active_file
888           unevictable
889 
890                 Amount of memory, swap-backed and filesystem-backed,
891                 on the internal memory management lists used by the
892                 page reclaim algorithm
893 
894           slab_reclaimable
895 
896                 Part of "slab" that might be reclaimed, such as
897                 dentries and inodes.
898 
899           slab_unreclaimable
900 
901                 Part of "slab" that cannot be reclaimed on memory
902                 pressure.
903 
904           pgfault
905 
906                 Total number of page faults incurred
907 
908           pgmajfault
909 
910                 Number of major page faults incurred
911 
912   memory.swap.current
913 
914         A read-only single value file which exists on non-root
915         cgroups.
916 
917         The total amount of swap currently being used by the cgroup
918         and its descendants.
919 
920   memory.swap.max
921 
922         A read-write single value file which exists on non-root
923         cgroups.  The default is "max".
924 
925         Swap usage hard limit.  If a cgroup's swap usage reaches this
926         limit, anonymous meomry of the cgroup will not be swapped out.
927 
928 
929 5-2-2. Usage Guidelines
930 
931 "memory.high" is the main mechanism to control memory usage.
932 Over-committing on high limit (sum of high limits > available memory)
933 and letting global memory pressure to distribute memory according to
934 usage is a viable strategy.
935 
936 Because breach of the high limit doesn't trigger the OOM killer but
937 throttles the offending cgroup, a management agent has ample
938 opportunities to monitor and take appropriate actions such as granting
939 more memory or terminating the workload.
940 
941 Determining whether a cgroup has enough memory is not trivial as
942 memory usage doesn't indicate whether the workload can benefit from
943 more memory.  For example, a workload which writes data received from
944 network to a file can use all available memory but can also operate as
945 performant with a small amount of memory.  A measure of memory
946 pressure - how much the workload is being impacted due to lack of
947 memory - is necessary to determine whether a workload needs more
948 memory; unfortunately, memory pressure monitoring mechanism isn't
949 implemented yet.
950 
951 
952 5-2-3. Memory Ownership
953 
954 A memory area is charged to the cgroup which instantiated it and stays
955 charged to the cgroup until the area is released.  Migrating a process
956 to a different cgroup doesn't move the memory usages that it
957 instantiated while in the previous cgroup to the new cgroup.
958 
959 A memory area may be used by processes belonging to different cgroups.
960 To which cgroup the area will be charged is in-deterministic; however,
961 over time, the memory area is likely to end up in a cgroup which has
962 enough memory allowance to avoid high reclaim pressure.
963 
964 If a cgroup sweeps a considerable amount of memory which is expected
965 to be accessed repeatedly by other cgroups, it may make sense to use
966 POSIX_FADV_DONTNEED to relinquish the ownership of memory areas
967 belonging to the affected files to ensure correct memory ownership.
968 
969 
970 5-3. IO
971 
972 The "io" controller regulates the distribution of IO resources.  This
973 controller implements both weight based and absolute bandwidth or IOPS
974 limit distribution; however, weight based distribution is available
975 only if cfq-iosched is in use and neither scheme is available for
976 blk-mq devices.
977 
978 
979 5-3-1. IO Interface Files
980 
981   io.stat
982 
983         A read-only nested-keyed file which exists on non-root
984         cgroups.
985 
986         Lines are keyed by $MAJ:$MIN device numbers and not ordered.
987         The following nested keys are defined.
988 
989           rbytes        Bytes read
990           wbytes        Bytes written
991           rios          Number of read IOs
992           wios          Number of write IOs
993 
994         An example read output follows.
995 
996           8:16 rbytes=1459200 wbytes=314773504 rios=192 wios=353
997           8:0 rbytes=90430464 wbytes=299008000 rios=8950 wios=1252
998 
999   io.weight
1000 
1001         A read-write flat-keyed file which exists on non-root cgroups.
1002         The default is "default 100".
1003 
1004         The first line is the default weight applied to devices
1005         without specific override.  The rest are overrides keyed by
1006         $MAJ:$MIN device numbers and not ordered.  The weights are in
1007         the range [1, 10000] and specifies the relative amount IO time
1008         the cgroup can use in relation to its siblings.
1009 
1010         The default weight can be updated by writing either "default
1011         $WEIGHT" or simply "$WEIGHT".  Overrides can be set by writing
1012         "$MAJ:$MIN $WEIGHT" and unset by writing "$MAJ:$MIN default".
1013 
1014         An example read output follows.
1015 
1016           default 100
1017           8:16 200
1018           8:0 50
1019 
1020   io.max
1021 
1022         A read-write nested-keyed file which exists on non-root
1023         cgroups.
1024 
1025         BPS and IOPS based IO limit.  Lines are keyed by $MAJ:$MIN
1026         device numbers and not ordered.  The following nested keys are
1027         defined.
1028 
1029           rbps          Max read bytes per second
1030           wbps          Max write bytes per second
1031           riops         Max read IO operations per second
1032           wiops         Max write IO operations per second
1033 
1034         When writing, any number of nested key-value pairs can be
1035         specified in any order.  "max" can be specified as the value
1036         to remove a specific limit.  If the same key is specified
1037         multiple times, the outcome is undefined.
1038 
1039         BPS and IOPS are measured in each IO direction and IOs are
1040         delayed if limit is reached.  Temporary bursts are allowed.
1041 
1042         Setting read limit at 2M BPS and write at 120 IOPS for 8:16.
1043 
1044           echo "8:16 rbps=2097152 wiops=120" > io.max
1045 
1046         Reading returns the following.
1047 
1048           8:16 rbps=2097152 wbps=max riops=max wiops=120
1049 
1050         Write IOPS limit can be removed by writing the following.
1051 
1052           echo "8:16 wiops=max" > io.max
1053 
1054         Reading now returns the following.
1055 
1056           8:16 rbps=2097152 wbps=max riops=max wiops=max
1057 
1058 
1059 5-3-2. Writeback
1060 
1061 Page cache is dirtied through buffered writes and shared mmaps and
1062 written asynchronously to the backing filesystem by the writeback
1063 mechanism.  Writeback sits between the memory and IO domains and
1064 regulates the proportion of dirty memory by balancing dirtying and
1065 write IOs.
1066 
1067 The io controller, in conjunction with the memory controller,
1068 implements control of page cache writeback IOs.  The memory controller
1069 defines the memory domain that dirty memory ratio is calculated and
1070 maintained for and the io controller defines the io domain which
1071 writes out dirty pages for the memory domain.  Both system-wide and
1072 per-cgroup dirty memory states are examined and the more restrictive
1073 of the two is enforced.
1074 
1075 cgroup writeback requires explicit support from the underlying
1076 filesystem.  Currently, cgroup writeback is implemented on ext2, ext4
1077 and btrfs.  On other filesystems, all writeback IOs are attributed to
1078 the root cgroup.
1079 
1080 There are inherent differences in memory and writeback management
1081 which affects how cgroup ownership is tracked.  Memory is tracked per
1082 page while writeback per inode.  For the purpose of writeback, an
1083 inode is assigned to a cgroup and all IO requests to write dirty pages
1084 from the inode are attributed to that cgroup.
1085 
1086 As cgroup ownership for memory is tracked per page, there can be pages
1087 which are associated with different cgroups than the one the inode is
1088 associated with.  These are called foreign pages.  The writeback
1089 constantly keeps track of foreign pages and, if a particular foreign
1090 cgroup becomes the majority over a certain period of time, switches
1091 the ownership of the inode to that cgroup.
1092 
1093 While this model is enough for most use cases where a given inode is
1094 mostly dirtied by a single cgroup even when the main writing cgroup
1095 changes over time, use cases where multiple cgroups write to a single
1096 inode simultaneously are not supported well.  In such circumstances, a
1097 significant portion of IOs are likely to be attributed incorrectly.
1098 As memory controller assigns page ownership on the first use and
1099 doesn't update it until the page is released, even if writeback
1100 strictly follows page ownership, multiple cgroups dirtying overlapping
1101 areas wouldn't work as expected.  It's recommended to avoid such usage
1102 patterns.
1103 
1104 The sysctl knobs which affect writeback behavior are applied to cgroup
1105 writeback as follows.
1106 
1107   vm.dirty_background_ratio
1108   vm.dirty_ratio
1109 
1110         These ratios apply the same to cgroup writeback with the
1111         amount of available memory capped by limits imposed by the
1112         memory controller and system-wide clean memory.
1113 
1114   vm.dirty_background_bytes
1115   vm.dirty_bytes
1116 
1117         For cgroup writeback, this is calculated into ratio against
1118         total available memory and applied the same way as
1119         vm.dirty[_background]_ratio.
1120 
1121 
1122 6. Namespace
1123 
1124 6-1. Basics
1125 
1126 cgroup namespace provides a mechanism to virtualize the view of the
1127 "/proc/$PID/cgroup" file and cgroup mounts.  The CLONE_NEWCGROUP clone
1128 flag can be used with clone(2) and unshare(2) to create a new cgroup
1129 namespace.  The process running inside the cgroup namespace will have
1130 its "/proc/$PID/cgroup" output restricted to cgroupns root.  The
1131 cgroupns root is the cgroup of the process at the time of creation of
1132 the cgroup namespace.
1133 
1134 Without cgroup namespace, the "/proc/$PID/cgroup" file shows the
1135 complete path of the cgroup of a process.  In a container setup where
1136 a set of cgroups and namespaces are intended to isolate processes the
1137 "/proc/$PID/cgroup" file may leak potential system level information
1138 to the isolated processes.  For Example:
1139 
1140   # cat /proc/self/cgroup
1141   0::/batchjobs/container_id1
1142 
1143 The path '/batchjobs/container_id1' can be considered as system-data
1144 and undesirable to expose to the isolated processes.  cgroup namespace
1145 can be used to restrict visibility of this path.  For example, before
1146 creating a cgroup namespace, one would see:
1147 
1148   # ls -l /proc/self/ns/cgroup
1149   lrwxrwxrwx 1 root root 0 2014-07-15 10:37 /proc/self/ns/cgroup -> cgroup:[4026531835]
1150   # cat /proc/self/cgroup
1151   0::/batchjobs/container_id1
1152 
1153 After unsharing a new namespace, the view changes.
1154 
1155   # ls -l /proc/self/ns/cgroup
1156   lrwxrwxrwx 1 root root 0 2014-07-15 10:35 /proc/self/ns/cgroup -> cgroup:[4026532183]
1157   # cat /proc/self/cgroup
1158   0::/
1159 
1160 When some thread from a multi-threaded process unshares its cgroup
1161 namespace, the new cgroupns gets applied to the entire process (all
1162 the threads).  This is natural for the v2 hierarchy; however, for the
1163 legacy hierarchies, this may be unexpected.
1164 
1165 A cgroup namespace is alive as long as there are processes inside or
1166 mounts pinning it.  When the last usage goes away, the cgroup
1167 namespace is destroyed.  The cgroupns root and the actual cgroups
1168 remain.
1169 
1170 
1171 6-2. The Root and Views
1172 
1173 The 'cgroupns root' for a cgroup namespace is the cgroup in which the
1174 process calling unshare(2) is running.  For example, if a process in
1175 /batchjobs/container_id1 cgroup calls unshare, cgroup
1176 /batchjobs/container_id1 becomes the cgroupns root.  For the
1177 init_cgroup_ns, this is the real root ('/') cgroup.
1178 
1179 The cgroupns root cgroup does not change even if the namespace creator
1180 process later moves to a different cgroup.
1181 
1182   # ~/unshare -c # unshare cgroupns in some cgroup
1183   # cat /proc/self/cgroup
1184   0::/
1185   # mkdir sub_cgrp_1
1186   # echo 0 > sub_cgrp_1/cgroup.procs
1187   # cat /proc/self/cgroup
1188   0::/sub_cgrp_1
1189 
1190 Each process gets its namespace-specific view of "/proc/$PID/cgroup"
1191 
1192 Processes running inside the cgroup namespace will be able to see
1193 cgroup paths (in /proc/self/cgroup) only inside their root cgroup.
1194 From within an unshared cgroupns:
1195 
1196   # sleep 100000 &
1197   [1] 7353
1198   # echo 7353 > sub_cgrp_1/cgroup.procs
1199   # cat /proc/7353/cgroup
1200   0::/sub_cgrp_1
1201 
1202 From the initial cgroup namespace, the real cgroup path will be
1203 visible:
1204 
1205   $ cat /proc/7353/cgroup
1206   0::/batchjobs/container_id1/sub_cgrp_1
1207 
1208 From a sibling cgroup namespace (that is, a namespace rooted at a
1209 different cgroup), the cgroup path relative to its own cgroup
1210 namespace root will be shown.  For instance, if PID 7353's cgroup
1211 namespace root is at '/batchjobs/container_id2', then it will see
1212 
1213   # cat /proc/7353/cgroup
1214   0::/../container_id2/sub_cgrp_1
1215 
1216 Note that the relative path always starts with '/' to indicate that
1217 its relative to the cgroup namespace root of the caller.
1218 
1219 
1220 6-3. Migration and setns(2)
1221 
1222 Processes inside a cgroup namespace can move into and out of the
1223 namespace root if they have proper access to external cgroups.  For
1224 example, from inside a namespace with cgroupns root at
1225 /batchjobs/container_id1, and assuming that the global hierarchy is
1226 still accessible inside cgroupns:
1227 
1228   # cat /proc/7353/cgroup
1229   0::/sub_cgrp_1
1230   # echo 7353 > batchjobs/container_id2/cgroup.procs
1231   # cat /proc/7353/cgroup
1232   0::/../container_id2
1233 
1234 Note that this kind of setup is not encouraged.  A task inside cgroup
1235 namespace should only be exposed to its own cgroupns hierarchy.
1236 
1237 setns(2) to another cgroup namespace is allowed when:
1238 
1239 (a) the process has CAP_SYS_ADMIN against its current user namespace
1240 (b) the process has CAP_SYS_ADMIN against the target cgroup
1241     namespace's userns
1242 
1243 No implicit cgroup changes happen with attaching to another cgroup
1244 namespace.  It is expected that the someone moves the attaching
1245 process under the target cgroup namespace root.
1246 
1247 
1248 6-4. Interaction with Other Namespaces
1249 
1250 Namespace specific cgroup hierarchy can be mounted by a process
1251 running inside a non-init cgroup namespace.
1252 
1253   # mount -t cgroup2 none $MOUNT_POINT
1254 
1255 This will mount the unified cgroup hierarchy with cgroupns root as the
1256 filesystem root.  The process needs CAP_SYS_ADMIN against its user and
1257 mount namespaces.
1258 
1259 The virtualization of /proc/self/cgroup file combined with restricting
1260 the view of cgroup hierarchy by namespace-private cgroupfs mount
1261 provides a properly isolated cgroup view inside the container.
1262 
1263 
1264 P. Information on Kernel Programming
1265 
1266 This section contains kernel programming information in the areas
1267 where interacting with cgroup is necessary.  cgroup core and
1268 controllers are not covered.
1269 
1270 
1271 P-1. Filesystem Support for Writeback
1272 
1273 A filesystem can support cgroup writeback by updating
1274 address_space_operations->writepage[s]() to annotate bio's using the
1275 following two functions.
1276 
1277   wbc_init_bio(@wbc, @bio)
1278 
1279         Should be called for each bio carrying writeback data and
1280         associates the bio with the inode's owner cgroup.  Can be
1281         called anytime between bio allocation and submission.
1282 
1283   wbc_account_io(@wbc, @page, @bytes)
1284 
1285         Should be called for each data segment being written out.
1286         While this function doesn't care exactly when it's called
1287         during the writeback session, it's the easiest and most
1288         natural to call it as data segments are added to a bio.
1289 
1290 With writeback bio's annotated, cgroup support can be enabled per
1291 super_block by setting SB_I_CGROUPWB in ->s_iflags.  This allows for
1292 selective disabling of cgroup writeback support which is helpful when
1293 certain filesystem features, e.g. journaled data mode, are
1294 incompatible.
1295 
1296 wbc_init_bio() binds the specified bio to its cgroup.  Depending on
1297 the configuration, the bio may be executed at a lower priority and if
1298 the writeback session is holding shared resources, e.g. a journal
1299 entry, may lead to priority inversion.  There is no one easy solution
1300 for the problem.  Filesystems can try to work around specific problem
1301 cases by skipping wbc_init_bio() or using bio_associate_blkcg()
1302 directly.
1303 
1304 
1305 D. Deprecated v1 Core Features
1306 
1307 - Multiple hierarchies including named ones are not supported.
1308 
1309 - All mount options and remounting are not supported.
1310 
1311 - The "tasks" file is removed and "cgroup.procs" is not sorted.
1312 
1313 - "cgroup.clone_children" is removed.
1314 
1315 - /proc/cgroups is meaningless for v2.  Use "cgroup.controllers" file
1316   at the root instead.
1317 
1318 
1319 R. Issues with v1 and Rationales for v2
1320 
1321 R-1. Multiple Hierarchies
1322 
1323 cgroup v1 allowed an arbitrary number of hierarchies and each
1324 hierarchy could host any number of controllers.  While this seemed to
1325 provide a high level of flexibility, it wasn't useful in practice.
1326 
1327 For example, as there is only one instance of each controller, utility
1328 type controllers such as freezer which can be useful in all
1329 hierarchies could only be used in one.  The issue is exacerbated by
1330 the fact that controllers couldn't be moved to another hierarchy once
1331 hierarchies were populated.  Another issue was that all controllers
1332 bound to a hierarchy were forced to have exactly the same view of the
1333 hierarchy.  It wasn't possible to vary the granularity depending on
1334 the specific controller.
1335 
1336 In practice, these issues heavily limited which controllers could be
1337 put on the same hierarchy and most configurations resorted to putting
1338 each controller on its own hierarchy.  Only closely related ones, such
1339 as the cpu and cpuacct controllers, made sense to be put on the same
1340 hierarchy.  This often meant that userland ended up managing multiple
1341 similar hierarchies repeating the same steps on each hierarchy
1342 whenever a hierarchy management operation was necessary.
1343 
1344 Furthermore, support for multiple hierarchies came at a steep cost.
1345 It greatly complicated cgroup core implementation but more importantly
1346 the support for multiple hierarchies restricted how cgroup could be
1347 used in general and what controllers was able to do.
1348 
1349 There was no limit on how many hierarchies there might be, which meant
1350 that a thread's cgroup membership couldn't be described in finite
1351 length.  The key might contain any number of entries and was unlimited
1352 in length, which made it highly awkward to manipulate and led to
1353 addition of controllers which existed only to identify membership,
1354 which in turn exacerbated the original problem of proliferating number
1355 of hierarchies.
1356 
1357 Also, as a controller couldn't have any expectation regarding the
1358 topologies of hierarchies other controllers might be on, each
1359 controller had to assume that all other controllers were attached to
1360 completely orthogonal hierarchies.  This made it impossible, or at
1361 least very cumbersome, for controllers to cooperate with each other.
1362 
1363 In most use cases, putting controllers on hierarchies which are
1364 completely orthogonal to each other isn't necessary.  What usually is
1365 called for is the ability to have differing levels of granularity
1366 depending on the specific controller.  In other words, hierarchy may
1367 be collapsed from leaf towards root when viewed from specific
1368 controllers.  For example, a given configuration might not care about
1369 how memory is distributed beyond a certain level while still wanting
1370 to control how CPU cycles are distributed.
1371 
1372 
1373 R-2. Thread Granularity
1374 
1375 cgroup v1 allowed threads of a process to belong to different cgroups.
1376 This didn't make sense for some controllers and those controllers
1377 ended up implementing different ways to ignore such situations but
1378 much more importantly it blurred the line between API exposed to
1379 individual applications and system management interface.
1380 
1381 Generally, in-process knowledge is available only to the process
1382 itself; thus, unlike service-level organization of processes,
1383 categorizing threads of a process requires active participation from
1384 the application which owns the target process.
1385 
1386 cgroup v1 had an ambiguously defined delegation model which got abused
1387 in combination with thread granularity.  cgroups were delegated to
1388 individual applications so that they can create and manage their own
1389 sub-hierarchies and control resource distributions along them.  This
1390 effectively raised cgroup to the status of a syscall-like API exposed
1391 to lay programs.
1392 
1393 First of all, cgroup has a fundamentally inadequate interface to be
1394 exposed this way.  For a process to access its own knobs, it has to
1395 extract the path on the target hierarchy from /proc/self/cgroup,
1396 construct the path by appending the name of the knob to the path, open
1397 and then read and/or write to it.  This is not only extremely clunky
1398 and unusual but also inherently racy.  There is no conventional way to
1399 define transaction across the required steps and nothing can guarantee
1400 that the process would actually be operating on its own sub-hierarchy.
1401 
1402 cgroup controllers implemented a number of knobs which would never be
1403 accepted as public APIs because they were just adding control knobs to
1404 system-management pseudo filesystem.  cgroup ended up with interface
1405 knobs which were not properly abstracted or refined and directly
1406 revealed kernel internal details.  These knobs got exposed to
1407 individual applications through the ill-defined delegation mechanism
1408 effectively abusing cgroup as a shortcut to implementing public APIs
1409 without going through the required scrutiny.
1410 
1411 This was painful for both userland and kernel.  Userland ended up with
1412 misbehaving and poorly abstracted interfaces and kernel exposing and
1413 locked into constructs inadvertently.
1414 
1415 
1416 R-3. Competition Between Inner Nodes and Threads
1417 
1418 cgroup v1 allowed threads to be in any cgroups which created an
1419 interesting problem where threads belonging to a parent cgroup and its
1420 children cgroups competed for resources.  This was nasty as two
1421 different types of entities competed and there was no obvious way to
1422 settle it.  Different controllers did different things.
1423 
1424 The cpu controller considered threads and cgroups as equivalents and
1425 mapped nice levels to cgroup weights.  This worked for some cases but
1426 fell flat when children wanted to be allocated specific ratios of CPU
1427 cycles and the number of internal threads fluctuated - the ratios
1428 constantly changed as the number of competing entities fluctuated.
1429 There also were other issues.  The mapping from nice level to weight
1430 wasn't obvious or universal, and there were various other knobs which
1431 simply weren't available for threads.
1432 
1433 The io controller implicitly created a hidden leaf node for each
1434 cgroup to host the threads.  The hidden leaf had its own copies of all
1435 the knobs with "leaf_" prefixed.  While this allowed equivalent
1436 control over internal threads, it was with serious drawbacks.  It
1437 always added an extra layer of nesting which wouldn't be necessary
1438 otherwise, made the interface messy and significantly complicated the
1439 implementation.
1440 
1441 The memory controller didn't have a way to control what happened
1442 between internal tasks and child cgroups and the behavior was not
1443 clearly defined.  There were attempts to add ad-hoc behaviors and
1444 knobs to tailor the behavior to specific workloads which would have
1445 led to problems extremely difficult to resolve in the long term.
1446 
1447 Multiple controllers struggled with internal tasks and came up with
1448 different ways to deal with it; unfortunately, all the approaches were
1449 severely flawed and, furthermore, the widely different behaviors
1450 made cgroup as a whole highly inconsistent.
1451 
1452 This clearly is a problem which needs to be addressed from cgroup core
1453 in a uniform way.
1454 
1455 
1456 R-4. Other Interface Issues
1457 
1458 cgroup v1 grew without oversight and developed a large number of
1459 idiosyncrasies and inconsistencies.  One issue on the cgroup core side
1460 was how an empty cgroup was notified - a userland helper binary was
1461 forked and executed for each event.  The event delivery wasn't
1462 recursive or delegatable.  The limitations of the mechanism also led
1463 to in-kernel event delivery filtering mechanism further complicating
1464 the interface.
1465 
1466 Controller interfaces were problematic too.  An extreme example is
1467 controllers completely ignoring hierarchical organization and treating
1468 all cgroups as if they were all located directly under the root
1469 cgroup.  Some controllers exposed a large amount of inconsistent
1470 implementation details to userland.
1471 
1472 There also was no consistency across controllers.  When a new cgroup
1473 was created, some controllers defaulted to not imposing extra
1474 restrictions while others disallowed any resource usage until
1475 explicitly configured.  Configuration knobs for the same type of
1476 control used widely differing naming schemes and formats.  Statistics
1477 and information knobs were named arbitrarily and used different
1478 formats and units even in the same controller.
1479 
1480 cgroup v2 establishes common conventions where appropriate and updates
1481 controllers so that they expose minimal and consistent interfaces.
1482 
1483 
1484 R-5. Controller Issues and Remedies
1485 
1486 R-5-1. Memory
1487 
1488 The original lower boundary, the soft limit, is defined as a limit
1489 that is per default unset.  As a result, the set of cgroups that
1490 global reclaim prefers is opt-in, rather than opt-out.  The costs for
1491 optimizing these mostly negative lookups are so high that the
1492 implementation, despite its enormous size, does not even provide the
1493 basic desirable behavior.  First off, the soft limit has no
1494 hierarchical meaning.  All configured groups are organized in a global
1495 rbtree and treated like equal peers, regardless where they are located
1496 in the hierarchy.  This makes subtree delegation impossible.  Second,
1497 the soft limit reclaim pass is so aggressive that it not just
1498 introduces high allocation latencies into the system, but also impacts
1499 system performance due to overreclaim, to the point where the feature
1500 becomes self-defeating.
1501 
1502 The memory.low boundary on the other hand is a top-down allocated
1503 reserve.  A cgroup enjoys reclaim protection when it and all its
1504 ancestors are below their low boundaries, which makes delegation of
1505 subtrees possible.  Secondly, new cgroups have no reserve per default
1506 and in the common case most cgroups are eligible for the preferred
1507 reclaim pass.  This allows the new low boundary to be efficiently
1508 implemented with just a minor addition to the generic reclaim code,
1509 without the need for out-of-band data structures and reclaim passes.
1510 Because the generic reclaim code considers all cgroups except for the
1511 ones running low in the preferred first reclaim pass, overreclaim of
1512 individual groups is eliminated as well, resulting in much better
1513 overall workload performance.
1514 
1515 The original high boundary, the hard limit, is defined as a strict
1516 limit that can not budge, even if the OOM killer has to be called.
1517 But this generally goes against the goal of making the most out of the
1518 available memory.  The memory consumption of workloads varies during
1519 runtime, and that requires users to overcommit.  But doing that with a
1520 strict upper limit requires either a fairly accurate prediction of the
1521 working set size or adding slack to the limit.  Since working set size
1522 estimation is hard and error prone, and getting it wrong results in
1523 OOM kills, most users tend to err on the side of a looser limit and
1524 end up wasting precious resources.
1525 
1526 The memory.high boundary on the other hand can be set much more
1527 conservatively.  When hit, it throttles allocations by forcing them
1528 into direct reclaim to work off the excess, but it never invokes the
1529 OOM killer.  As a result, a high boundary that is chosen too
1530 aggressively will not terminate the processes, but instead it will
1531 lead to gradual performance degradation.  The user can monitor this
1532 and make corrections until the minimal memory footprint that still
1533 gives acceptable performance is found.
1534 
1535 In extreme cases, with many concurrent allocations and a complete
1536 breakdown of reclaim progress within the group, the high boundary can
1537 be exceeded.  But even then it's mostly better to satisfy the
1538 allocation from the slack available in other groups or the rest of the
1539 system than killing the group.  Otherwise, memory.max is there to
1540 limit this type of spillover and ultimately contain buggy or even
1541 malicious applications.
1542 
1543 Setting the original memory.limit_in_bytes below the current usage was
1544 subject to a race condition, where concurrent charges could cause the
1545 limit setting to fail. memory.max on the other hand will first set the
1546 limit to prevent new charges, and then reclaim and OOM kill until the
1547 new limit is met - or the task writing to memory.max is killed.
1548 
1549 The combined memory+swap accounting and limiting is replaced by real
1550 control over swap space.
1551 
1552 The main argument for a combined memory+swap facility in the original
1553 cgroup design was that global or parental pressure would always be
1554 able to swap all anonymous memory of a child group, regardless of the
1555 child's own (possibly untrusted) configuration.  However, untrusted
1556 groups can sabotage swapping by other means - such as referencing its
1557 anonymous memory in a tight loop - and an admin can not assume full
1558 swappability when overcommitting untrusted jobs.
1559 
1560 For trusted jobs, on the other hand, a combined counter is not an
1561 intuitive userspace interface, and it flies in the face of the idea
1562 that cgroup controllers should account and limit specific physical
1563 resources.  Swap space is a resource like all others in the system,
1564 and that's why unified hierarchy allows distributing it separately.

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us