]>
Commit | Line | Data |
---|---|---|
633b11be | 1 | ================ |
6c292092 | 2 | Control Group v2 |
633b11be | 3 | ================ |
6c292092 | 4 | |
633b11be MCC |
5 | :Date: October, 2015 |
6 | :Author: Tejun Heo <tj@kernel.org> | |
6c292092 TH |
7 | |
8 | This is the authoritative documentation on the design, interface and | |
9 | conventions of cgroup v2. It describes all userland-visible aspects | |
10 | of cgroup including core and specific controller behaviors. All | |
11 | future changes must be reflected in this document. Documentation for | |
9a2ddda5 | 12 | v1 is available under Documentation/cgroup-v1/. |
6c292092 | 13 | |
633b11be MCC |
14 | .. CONTENTS |
15 | ||
16 | 1. Introduction | |
17 | 1-1. Terminology | |
18 | 1-2. What is cgroup? | |
19 | 2. Basic Operations | |
20 | 2-1. Mounting | |
8cfd8147 TH |
21 | 2-2. Organizing Processes and Threads |
22 | 2-2-1. Processes | |
23 | 2-2-2. Threads | |
633b11be MCC |
24 | 2-3. [Un]populated Notification |
25 | 2-4. Controlling Controllers | |
26 | 2-4-1. Enabling and Disabling | |
27 | 2-4-2. Top-down Constraint | |
28 | 2-4-3. No Internal Process Constraint | |
29 | 2-5. Delegation | |
30 | 2-5-1. Model of Delegation | |
31 | 2-5-2. Delegation Containment | |
32 | 2-6. Guidelines | |
33 | 2-6-1. Organize Once and Control | |
34 | 2-6-2. Avoid Name Collisions | |
35 | 3. Resource Distribution Models | |
36 | 3-1. Weights | |
37 | 3-2. Limits | |
38 | 3-3. Protections | |
39 | 3-4. Allocations | |
40 | 4. Interface Files | |
41 | 4-1. Format | |
42 | 4-2. Conventions | |
43 | 4-3. Core Interface Files | |
44 | 5. Controllers | |
45 | 5-1. CPU | |
46 | 5-1-1. CPU Interface Files | |
47 | 5-2. Memory | |
48 | 5-2-1. Memory Interface Files | |
49 | 5-2-2. Usage Guidelines | |
50 | 5-2-3. Memory Ownership | |
51 | 5-3. IO | |
52 | 5-3-1. IO Interface Files | |
53 | 5-3-2. Writeback | |
54 | 5-4. PID | |
55 | 5-4-1. PID Interface Files | |
56 | 5-5. RDMA | |
57 | 5-5-1. RDMA Interface Files | |
58 | 5-6. Misc | |
59 | 5-6-1. perf_event | |
60 | 6. Namespace | |
61 | 6-1. Basics | |
62 | 6-2. The Root and Views | |
63 | 6-3. Migration and setns(2) | |
64 | 6-4. Interaction with Other Namespaces | |
65 | P. Information on Kernel Programming | |
66 | P-1. Filesystem Support for Writeback | |
67 | D. Deprecated v1 Core Features | |
68 | R. Issues with v1 and Rationales for v2 | |
69 | R-1. Multiple Hierarchies | |
70 | R-2. Thread Granularity | |
71 | R-3. Competition Between Inner Nodes and Threads | |
72 | R-4. Other Interface Issues | |
73 | R-5. Controller Issues and Remedies | |
74 | R-5-1. Memory | |
75 | ||
76 | ||
77 | Introduction | |
78 | ============ | |
79 | ||
80 | Terminology | |
81 | ----------- | |
6c292092 TH |
82 | |
83 | "cgroup" stands for "control group" and is never capitalized. The | |
84 | singular form is used to designate the whole feature and also as a | |
85 | qualifier as in "cgroup controllers". When explicitly referring to | |
86 | multiple individual control groups, the plural form "cgroups" is used. | |
87 | ||
88 | ||
633b11be MCC |
89 | What is cgroup? |
90 | --------------- | |
6c292092 TH |
91 | |
92 | cgroup is a mechanism to organize processes hierarchically and | |
93 | distribute system resources along the hierarchy in a controlled and | |
94 | configurable manner. | |
95 | ||
96 | cgroup is largely composed of two parts - the core and controllers. | |
97 | cgroup core is primarily responsible for hierarchically organizing | |
98 | processes. A cgroup controller is usually responsible for | |
99 | distributing a specific type of system resource along the hierarchy | |
100 | although there are utility controllers which serve purposes other than | |
101 | resource distribution. | |
102 | ||
103 | cgroups form a tree structure and every process in the system belongs | |
104 | to one and only one cgroup. All threads of a process belong to the | |
105 | same cgroup. On creation, all processes are put in the cgroup that | |
106 | the parent process belongs to at the time. A process can be migrated | |
107 | to another cgroup. Migration of a process doesn't affect already | |
108 | existing descendant processes. | |
109 | ||
110 | Following certain structural constraints, controllers may be enabled or | |
111 | disabled selectively on a cgroup. All controller behaviors are | |
112 | hierarchical - if a controller is enabled on a cgroup, it affects all | |
113 | processes which belong to the cgroups consisting the inclusive | |
114 | sub-hierarchy of the cgroup. When a controller is enabled on a nested | |
115 | cgroup, it always restricts the resource distribution further. The | |
116 | restrictions set closer to the root in the hierarchy can not be | |
117 | overridden from further away. | |
118 | ||
119 | ||
633b11be MCC |
120 | Basic Operations |
121 | ================ | |
6c292092 | 122 | |
633b11be MCC |
123 | Mounting |
124 | -------- | |
6c292092 TH |
125 | |
126 | Unlike v1, cgroup v2 has only single hierarchy. The cgroup v2 | |
633b11be | 127 | hierarchy can be mounted with the following mount command:: |
6c292092 TH |
128 | |
129 | # mount -t cgroup2 none $MOUNT_POINT | |
130 | ||
131 | cgroup2 filesystem has the magic number 0x63677270 ("cgrp"). All | |
132 | controllers which support v2 and are not bound to a v1 hierarchy are | |
133 | automatically bound to the v2 hierarchy and show up at the root. | |
134 | Controllers which are not in active use in the v2 hierarchy can be | |
135 | bound to other hierarchies. This allows mixing v2 hierarchy with the | |
136 | legacy v1 multiple hierarchies in a fully backward compatible way. | |
137 | ||
138 | A controller can be moved across hierarchies only after the controller | |
139 | is no longer referenced in its current hierarchy. Because per-cgroup | |
140 | controller states are destroyed asynchronously and controllers may | |
141 | have lingering references, a controller may not show up immediately on | |
142 | the v2 hierarchy after the final umount of the previous hierarchy. | |
143 | Similarly, a controller should be fully disabled to be moved out of | |
144 | the unified hierarchy and it may take some time for the disabled | |
145 | controller to become available for other hierarchies; furthermore, due | |
146 | to inter-controller dependencies, other controllers may need to be | |
147 | disabled too. | |
148 | ||
149 | While useful for development and manual configurations, moving | |
150 | controllers dynamically between the v2 and other hierarchies is | |
151 | strongly discouraged for production use. It is recommended to decide | |
152 | the hierarchies and controller associations before starting using the | |
153 | controllers after system boot. | |
154 | ||
1619b6d4 JW |
155 | During transition to v2, system management software might still |
156 | automount the v1 cgroup filesystem and so hijack all controllers | |
157 | during boot, before manual intervention is possible. To make testing | |
158 | and experimenting easier, the kernel parameter cgroup_no_v1= allows | |
159 | disabling controllers in v1 and make them always available in v2. | |
160 | ||
5136f636 TH |
161 | cgroup v2 currently supports the following mount options. |
162 | ||
163 | nsdelegate | |
164 | ||
165 | Consider cgroup namespaces as delegation boundaries. This | |
166 | option is system wide and can only be set on mount or modified | |
167 | through remount from the init namespace. The mount option is | |
168 | ignored on non-init namespace mounts. Please refer to the | |
169 | Delegation section for details. | |
170 | ||
6c292092 | 171 | |
8cfd8147 TH |
172 | Organizing Processes and Threads |
173 | -------------------------------- | |
174 | ||
175 | Processes | |
176 | ~~~~~~~~~ | |
6c292092 TH |
177 | |
178 | Initially, only the root cgroup exists to which all processes belong. | |
633b11be | 179 | A child cgroup can be created by creating a sub-directory:: |
6c292092 TH |
180 | |
181 | # mkdir $CGROUP_NAME | |
182 | ||
183 | A given cgroup may have multiple child cgroups forming a tree | |
184 | structure. Each cgroup has a read-writable interface file | |
185 | "cgroup.procs". When read, it lists the PIDs of all processes which | |
186 | belong to the cgroup one-per-line. The PIDs are not ordered and the | |
187 | same PID may show up more than once if the process got moved to | |
188 | another cgroup and then back or the PID got recycled while reading. | |
189 | ||
190 | A process can be migrated into a cgroup by writing its PID to the | |
191 | target cgroup's "cgroup.procs" file. Only one process can be migrated | |
192 | on a single write(2) call. If a process is composed of multiple | |
193 | threads, writing the PID of any thread migrates all threads of the | |
194 | process. | |
195 | ||
196 | When a process forks a child process, the new process is born into the | |
197 | cgroup that the forking process belongs to at the time of the | |
198 | operation. After exit, a process stays associated with the cgroup | |
199 | that it belonged to at the time of exit until it's reaped; however, a | |
200 | zombie process does not appear in "cgroup.procs" and thus can't be | |
201 | moved to another cgroup. | |
202 | ||
203 | A cgroup which doesn't have any children or live processes can be | |
204 | destroyed by removing the directory. Note that a cgroup which doesn't | |
205 | have any children and is associated only with zombie processes is | |
633b11be | 206 | considered empty and can be removed:: |
6c292092 TH |
207 | |
208 | # rmdir $CGROUP_NAME | |
209 | ||
210 | "/proc/$PID/cgroup" lists a process's cgroup membership. If legacy | |
211 | cgroup is in use in the system, this file may contain multiple lines, | |
212 | one for each hierarchy. The entry for cgroup v2 is always in the | |
633b11be | 213 | format "0::$PATH":: |
6c292092 TH |
214 | |
215 | # cat /proc/842/cgroup | |
216 | ... | |
217 | 0::/test-cgroup/test-cgroup-nested | |
218 | ||
219 | If the process becomes a zombie and the cgroup it was associated with | |
633b11be | 220 | is removed subsequently, " (deleted)" is appended to the path:: |
6c292092 TH |
221 | |
222 | # cat /proc/842/cgroup | |
223 | ... | |
224 | 0::/test-cgroup/test-cgroup-nested (deleted) | |
225 | ||
226 | ||
8cfd8147 TH |
227 | Threads |
228 | ~~~~~~~ | |
229 | ||
230 | cgroup v2 supports thread granularity for a subset of controllers to | |
231 | support use cases requiring hierarchical resource distribution across | |
232 | the threads of a group of processes. By default, all threads of a | |
233 | process belong to the same cgroup, which also serves as the resource | |
234 | domain to host resource consumptions which are not specific to a | |
235 | process or thread. The thread mode allows threads to be spread across | |
236 | a subtree while still maintaining the common resource domain for them. | |
237 | ||
238 | Controllers which support thread mode are called threaded controllers. | |
239 | The ones which don't are called domain controllers. | |
240 | ||
241 | Marking a cgroup threaded makes it join the resource domain of its | |
242 | parent as a threaded cgroup. The parent may be another threaded | |
243 | cgroup whose resource domain is further up in the hierarchy. The root | |
244 | of a threaded subtree, that is, the nearest ancestor which is not | |
245 | threaded, is called threaded domain or thread root interchangeably and | |
246 | serves as the resource domain for the entire subtree. | |
247 | ||
248 | Inside a threaded subtree, threads of a process can be put in | |
249 | different cgroups and are not subject to the no internal process | |
250 | constraint - threaded controllers can be enabled on non-leaf cgroups | |
251 | whether they have threads in them or not. | |
252 | ||
253 | As the threaded domain cgroup hosts all the domain resource | |
254 | consumptions of the subtree, it is considered to have internal | |
255 | resource consumptions whether there are processes in it or not and | |
256 | can't have populated child cgroups which aren't threaded. Because the | |
257 | root cgroup is not subject to no internal process constraint, it can | |
258 | serve both as a threaded domain and a parent to domain cgroups. | |
259 | ||
260 | The current operation mode or type of the cgroup is shown in the | |
261 | "cgroup.type" file which indicates whether the cgroup is a normal | |
262 | domain, a domain which is serving as the domain of a threaded subtree, | |
263 | or a threaded cgroup. | |
264 | ||
265 | On creation, a cgroup is always a domain cgroup and can be made | |
266 | threaded by writing "threaded" to the "cgroup.type" file. The | |
267 | operation is single direction:: | |
268 | ||
269 | # echo threaded > cgroup.type | |
270 | ||
271 | Once threaded, the cgroup can't be made a domain again. To enable the | |
272 | thread mode, the following conditions must be met. | |
273 | ||
274 | - As the cgroup will join the parent's resource domain. The parent | |
275 | must either be a valid (threaded) domain or a threaded cgroup. | |
276 | ||
918a8c2c TH |
277 | - When the parent is an unthreaded domain, it must not have any domain |
278 | controllers enabled or populated domain children. The root is | |
279 | exempt from this requirement. | |
8cfd8147 TH |
280 | |
281 | Topology-wise, a cgroup can be in an invalid state. Please consider | |
282 | the following toplogy:: | |
283 | ||
284 | A (threaded domain) - B (threaded) - C (domain, just created) | |
285 | ||
286 | C is created as a domain but isn't connected to a parent which can | |
287 | host child domains. C can't be used until it is turned into a | |
288 | threaded cgroup. "cgroup.type" file will report "domain (invalid)" in | |
289 | these cases. Operations which fail due to invalid topology use | |
290 | EOPNOTSUPP as the errno. | |
291 | ||
292 | A domain cgroup is turned into a threaded domain when one of its child | |
293 | cgroup becomes threaded or threaded controllers are enabled in the | |
294 | "cgroup.subtree_control" file while there are processes in the cgroup. | |
295 | A threaded domain reverts to a normal domain when the conditions | |
296 | clear. | |
297 | ||
298 | When read, "cgroup.threads" contains the list of the thread IDs of all | |
299 | threads in the cgroup. Except that the operations are per-thread | |
300 | instead of per-process, "cgroup.threads" has the same format and | |
301 | behaves the same way as "cgroup.procs". While "cgroup.threads" can be | |
302 | written to in any cgroup, as it can only move threads inside the same | |
303 | threaded domain, its operations are confined inside each threaded | |
304 | subtree. | |
305 | ||
306 | The threaded domain cgroup serves as the resource domain for the whole | |
307 | subtree, and, while the threads can be scattered across the subtree, | |
308 | all the processes are considered to be in the threaded domain cgroup. | |
309 | "cgroup.procs" in a threaded domain cgroup contains the PIDs of all | |
310 | processes in the subtree and is not readable in the subtree proper. | |
311 | However, "cgroup.procs" can be written to from anywhere in the subtree | |
312 | to migrate all threads of the matching process to the cgroup. | |
313 | ||
314 | Only threaded controllers can be enabled in a threaded subtree. When | |
315 | a threaded controller is enabled inside a threaded subtree, it only | |
316 | accounts for and controls resource consumptions associated with the | |
317 | threads in the cgroup and its descendants. All consumptions which | |
318 | aren't tied to a specific thread belong to the threaded domain cgroup. | |
319 | ||
320 | Because a threaded subtree is exempt from no internal process | |
321 | constraint, a threaded controller must be able to handle competition | |
322 | between threads in a non-leaf cgroup and its child cgroups. Each | |
323 | threaded controller defines how such competitions are handled. | |
324 | ||
325 | ||
633b11be MCC |
326 | [Un]populated Notification |
327 | -------------------------- | |
6c292092 TH |
328 | |
329 | Each non-root cgroup has a "cgroup.events" file which contains | |
330 | "populated" field indicating whether the cgroup's sub-hierarchy has | |
331 | live processes in it. Its value is 0 if there is no live process in | |
332 | the cgroup and its descendants; otherwise, 1. poll and [id]notify | |
333 | events are triggered when the value changes. This can be used, for | |
334 | example, to start a clean-up operation after all processes of a given | |
335 | sub-hierarchy have exited. The populated state updates and | |
336 | notifications are recursive. Consider the following sub-hierarchy | |
337 | where the numbers in the parentheses represent the numbers of processes | |
633b11be | 338 | in each cgroup:: |
6c292092 TH |
339 | |
340 | A(4) - B(0) - C(1) | |
341 | \ D(0) | |
342 | ||
343 | A, B and C's "populated" fields would be 1 while D's 0. After the one | |
344 | process in C exits, B and C's "populated" fields would flip to "0" and | |
345 | file modified events will be generated on the "cgroup.events" files of | |
346 | both cgroups. | |
347 | ||
348 | ||
633b11be MCC |
349 | Controlling Controllers |
350 | ----------------------- | |
6c292092 | 351 | |
633b11be MCC |
352 | Enabling and Disabling |
353 | ~~~~~~~~~~~~~~~~~~~~~~ | |
6c292092 TH |
354 | |
355 | Each cgroup has a "cgroup.controllers" file which lists all | |
633b11be | 356 | controllers available for the cgroup to enable:: |
6c292092 TH |
357 | |
358 | # cat cgroup.controllers | |
359 | cpu io memory | |
360 | ||
361 | No controller is enabled by default. Controllers can be enabled and | |
633b11be | 362 | disabled by writing to the "cgroup.subtree_control" file:: |
6c292092 TH |
363 | |
364 | # echo "+cpu +memory -io" > cgroup.subtree_control | |
365 | ||
366 | Only controllers which are listed in "cgroup.controllers" can be | |
367 | enabled. When multiple operations are specified as above, either they | |
368 | all succeed or fail. If multiple operations on the same controller | |
369 | are specified, the last one is effective. | |
370 | ||
371 | Enabling a controller in a cgroup indicates that the distribution of | |
372 | the target resource across its immediate children will be controlled. | |
373 | Consider the following sub-hierarchy. The enabled controllers are | |
633b11be | 374 | listed in parentheses:: |
6c292092 TH |
375 | |
376 | A(cpu,memory) - B(memory) - C() | |
377 | \ D() | |
378 | ||
379 | As A has "cpu" and "memory" enabled, A will control the distribution | |
380 | of CPU cycles and memory to its children, in this case, B. As B has | |
381 | "memory" enabled but not "CPU", C and D will compete freely on CPU | |
382 | cycles but their division of memory available to B will be controlled. | |
383 | ||
384 | As a controller regulates the distribution of the target resource to | |
385 | the cgroup's children, enabling it creates the controller's interface | |
386 | files in the child cgroups. In the above example, enabling "cpu" on B | |
387 | would create the "cpu." prefixed controller interface files in C and | |
388 | D. Likewise, disabling "memory" from B would remove the "memory." | |
389 | prefixed controller interface files from C and D. This means that the | |
390 | controller interface files - anything which doesn't start with | |
391 | "cgroup." are owned by the parent rather than the cgroup itself. | |
392 | ||
393 | ||
633b11be MCC |
394 | Top-down Constraint |
395 | ~~~~~~~~~~~~~~~~~~~ | |
6c292092 TH |
396 | |
397 | Resources are distributed top-down and a cgroup can further distribute | |
398 | a resource only if the resource has been distributed to it from the | |
399 | parent. This means that all non-root "cgroup.subtree_control" files | |
400 | can only contain controllers which are enabled in the parent's | |
401 | "cgroup.subtree_control" file. A controller can be enabled only if | |
402 | the parent has the controller enabled and a controller can't be | |
403 | disabled if one or more children have it enabled. | |
404 | ||
405 | ||
633b11be MCC |
406 | No Internal Process Constraint |
407 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | |
6c292092 | 408 | |
8cfd8147 TH |
409 | Non-root cgroups can distribute domain resources to their children |
410 | only when they don't have any processes of their own. In other words, | |
411 | only domain cgroups which don't contain any processes can have domain | |
412 | controllers enabled in their "cgroup.subtree_control" files. | |
6c292092 | 413 | |
8cfd8147 TH |
414 | This guarantees that, when a domain controller is looking at the part |
415 | of the hierarchy which has it enabled, processes are always only on | |
416 | the leaves. This rules out situations where child cgroups compete | |
417 | against internal processes of the parent. | |
6c292092 TH |
418 | |
419 | The root cgroup is exempt from this restriction. Root contains | |
420 | processes and anonymous resource consumption which can't be associated | |
421 | with any other cgroups and requires special treatment from most | |
422 | controllers. How resource consumption in the root cgroup is governed | |
423 | is up to each controller. | |
424 | ||
425 | Note that the restriction doesn't get in the way if there is no | |
426 | enabled controller in the cgroup's "cgroup.subtree_control". This is | |
427 | important as otherwise it wouldn't be possible to create children of a | |
428 | populated cgroup. To control resource distribution of a cgroup, the | |
429 | cgroup must create children and transfer all its processes to the | |
430 | children before enabling controllers in its "cgroup.subtree_control" | |
431 | file. | |
432 | ||
433 | ||
633b11be MCC |
434 | Delegation |
435 | ---------- | |
6c292092 | 436 | |
633b11be MCC |
437 | Model of Delegation |
438 | ~~~~~~~~~~~~~~~~~~~ | |
6c292092 | 439 | |
5136f636 | 440 | A cgroup can be delegated in two ways. First, to a less privileged |
8cfd8147 TH |
441 | user by granting write access of the directory and its "cgroup.procs", |
442 | "cgroup.threads" and "cgroup.subtree_control" files to the user. | |
443 | Second, if the "nsdelegate" mount option is set, automatically to a | |
444 | cgroup namespace on namespace creation. | |
5136f636 TH |
445 | |
446 | Because the resource control interface files in a given directory | |
447 | control the distribution of the parent's resources, the delegatee | |
448 | shouldn't be allowed to write to them. For the first method, this is | |
449 | achieved by not granting access to these files. For the second, the | |
450 | kernel rejects writes to all files other than "cgroup.procs" and | |
451 | "cgroup.subtree_control" on a namespace root from inside the | |
452 | namespace. | |
453 | ||
454 | The end results are equivalent for both delegation types. Once | |
455 | delegated, the user can build sub-hierarchy under the directory, | |
456 | organize processes inside it as it sees fit and further distribute the | |
457 | resources it received from the parent. The limits and other settings | |
458 | of all resource controllers are hierarchical and regardless of what | |
459 | happens in the delegated sub-hierarchy, nothing can escape the | |
460 | resource restrictions imposed by the parent. | |
6c292092 TH |
461 | |
462 | Currently, cgroup doesn't impose any restrictions on the number of | |
463 | cgroups in or nesting depth of a delegated sub-hierarchy; however, | |
464 | this may be limited explicitly in the future. | |
465 | ||
466 | ||
633b11be MCC |
467 | Delegation Containment |
468 | ~~~~~~~~~~~~~~~~~~~~~~ | |
6c292092 TH |
469 | |
470 | A delegated sub-hierarchy is contained in the sense that processes | |
5136f636 TH |
471 | can't be moved into or out of the sub-hierarchy by the delegatee. |
472 | ||
473 | For delegations to a less privileged user, this is achieved by | |
474 | requiring the following conditions for a process with a non-root euid | |
475 | to migrate a target process into a cgroup by writing its PID to the | |
476 | "cgroup.procs" file. | |
6c292092 | 477 | |
6c292092 TH |
478 | - The writer must have write access to the "cgroup.procs" file. |
479 | ||
480 | - The writer must have write access to the "cgroup.procs" file of the | |
481 | common ancestor of the source and destination cgroups. | |
482 | ||
576dd464 | 483 | The above two constraints ensure that while a delegatee may migrate |
6c292092 TH |
484 | processes around freely in the delegated sub-hierarchy it can't pull |
485 | in from or push out to outside the sub-hierarchy. | |
486 | ||
487 | For an example, let's assume cgroups C0 and C1 have been delegated to | |
488 | user U0 who created C00, C01 under C0 and C10 under C1 as follows and | |
633b11be | 489 | all processes under C0 and C1 belong to U0:: |
6c292092 TH |
490 | |
491 | ~~~~~~~~~~~~~ - C0 - C00 | |
492 | ~ cgroup ~ \ C01 | |
493 | ~ hierarchy ~ | |
494 | ~~~~~~~~~~~~~ - C1 - C10 | |
495 | ||
496 | Let's also say U0 wants to write the PID of a process which is | |
497 | currently in C10 into "C00/cgroup.procs". U0 has write access to the | |
576dd464 TH |
498 | file; however, the common ancestor of the source cgroup C10 and the |
499 | destination cgroup C00 is above the points of delegation and U0 would | |
500 | not have write access to its "cgroup.procs" files and thus the write | |
501 | will be denied with -EACCES. | |
6c292092 | 502 | |
5136f636 TH |
503 | For delegations to namespaces, containment is achieved by requiring |
504 | that both the source and destination cgroups are reachable from the | |
505 | namespace of the process which is attempting the migration. If either | |
506 | is not reachable, the migration is rejected with -ENOENT. | |
507 | ||
6c292092 | 508 | |
633b11be MCC |
509 | Guidelines |
510 | ---------- | |
6c292092 | 511 | |
633b11be MCC |
512 | Organize Once and Control |
513 | ~~~~~~~~~~~~~~~~~~~~~~~~~ | |
6c292092 TH |
514 | |
515 | Migrating a process across cgroups is a relatively expensive operation | |
516 | and stateful resources such as memory are not moved together with the | |
517 | process. This is an explicit design decision as there often exist | |
518 | inherent trade-offs between migration and various hot paths in terms | |
519 | of synchronization cost. | |
520 | ||
521 | As such, migrating processes across cgroups frequently as a means to | |
522 | apply different resource restrictions is discouraged. A workload | |
523 | should be assigned to a cgroup according to the system's logical and | |
524 | resource structure once on start-up. Dynamic adjustments to resource | |
525 | distribution can be made by changing controller configuration through | |
526 | the interface files. | |
527 | ||
528 | ||
633b11be MCC |
529 | Avoid Name Collisions |
530 | ~~~~~~~~~~~~~~~~~~~~~ | |
6c292092 TH |
531 | |
532 | Interface files for a cgroup and its children cgroups occupy the same | |
533 | directory and it is possible to create children cgroups which collide | |
534 | with interface files. | |
535 | ||
536 | All cgroup core interface files are prefixed with "cgroup." and each | |
537 | controller's interface files are prefixed with the controller name and | |
538 | a dot. A controller's name is composed of lower case alphabets and | |
539 | '_'s but never begins with an '_' so it can be used as the prefix | |
540 | character for collision avoidance. Also, interface file names won't | |
541 | start or end with terms which are often used in categorizing workloads | |
542 | such as job, service, slice, unit or workload. | |
543 | ||
544 | cgroup doesn't do anything to prevent name collisions and it's the | |
545 | user's responsibility to avoid them. | |
546 | ||
547 | ||
633b11be MCC |
548 | Resource Distribution Models |
549 | ============================ | |
6c292092 TH |
550 | |
551 | cgroup controllers implement several resource distribution schemes | |
552 | depending on the resource type and expected use cases. This section | |
553 | describes major schemes in use along with their expected behaviors. | |
554 | ||
555 | ||
633b11be MCC |
556 | Weights |
557 | ------- | |
6c292092 TH |
558 | |
559 | A parent's resource is distributed by adding up the weights of all | |
560 | active children and giving each the fraction matching the ratio of its | |
561 | weight against the sum. As only children which can make use of the | |
562 | resource at the moment participate in the distribution, this is | |
563 | work-conserving. Due to the dynamic nature, this model is usually | |
564 | used for stateless resources. | |
565 | ||
566 | All weights are in the range [1, 10000] with the default at 100. This | |
567 | allows symmetric multiplicative biases in both directions at fine | |
568 | enough granularity while staying in the intuitive range. | |
569 | ||
570 | As long as the weight is in range, all configuration combinations are | |
571 | valid and there is no reason to reject configuration changes or | |
572 | process migrations. | |
573 | ||
574 | "cpu.weight" proportionally distributes CPU cycles to active children | |
575 | and is an example of this type. | |
576 | ||
577 | ||
633b11be MCC |
578 | Limits |
579 | ------ | |
6c292092 TH |
580 | |
581 | A child can only consume upto the configured amount of the resource. | |
582 | Limits can be over-committed - the sum of the limits of children can | |
583 | exceed the amount of resource available to the parent. | |
584 | ||
585 | Limits are in the range [0, max] and defaults to "max", which is noop. | |
586 | ||
587 | As limits can be over-committed, all configuration combinations are | |
588 | valid and there is no reason to reject configuration changes or | |
589 | process migrations. | |
590 | ||
591 | "io.max" limits the maximum BPS and/or IOPS that a cgroup can consume | |
592 | on an IO device and is an example of this type. | |
593 | ||
594 | ||
633b11be MCC |
595 | Protections |
596 | ----------- | |
6c292092 TH |
597 | |
598 | A cgroup is protected to be allocated upto the configured amount of | |
599 | the resource if the usages of all its ancestors are under their | |
600 | protected levels. Protections can be hard guarantees or best effort | |
601 | soft boundaries. Protections can also be over-committed in which case | |
602 | only upto the amount available to the parent is protected among | |
603 | children. | |
604 | ||
605 | Protections are in the range [0, max] and defaults to 0, which is | |
606 | noop. | |
607 | ||
608 | As protections can be over-committed, all configuration combinations | |
609 | are valid and there is no reason to reject configuration changes or | |
610 | process migrations. | |
611 | ||
612 | "memory.low" implements best-effort memory protection and is an | |
613 | example of this type. | |
614 | ||
615 | ||
633b11be MCC |
616 | Allocations |
617 | ----------- | |
6c292092 TH |
618 | |
619 | A cgroup is exclusively allocated a certain amount of a finite | |
620 | resource. Allocations can't be over-committed - the sum of the | |
621 | allocations of children can not exceed the amount of resource | |
622 | available to the parent. | |
623 | ||
624 | Allocations are in the range [0, max] and defaults to 0, which is no | |
625 | resource. | |
626 | ||
627 | As allocations can't be over-committed, some configuration | |
628 | combinations are invalid and should be rejected. Also, if the | |
629 | resource is mandatory for execution of processes, process migrations | |
630 | may be rejected. | |
631 | ||
632 | "cpu.rt.max" hard-allocates realtime slices and is an example of this | |
633 | type. | |
634 | ||
635 | ||
633b11be MCC |
636 | Interface Files |
637 | =============== | |
6c292092 | 638 | |
633b11be MCC |
639 | Format |
640 | ------ | |
6c292092 TH |
641 | |
642 | All interface files should be in one of the following formats whenever | |
633b11be | 643 | possible:: |
6c292092 TH |
644 | |
645 | New-line separated values | |
646 | (when only one value can be written at once) | |
647 | ||
648 | VAL0\n | |
649 | VAL1\n | |
650 | ... | |
651 | ||
652 | Space separated values | |
653 | (when read-only or multiple values can be written at once) | |
654 | ||
655 | VAL0 VAL1 ...\n | |
656 | ||
657 | Flat keyed | |
658 | ||
659 | KEY0 VAL0\n | |
660 | KEY1 VAL1\n | |
661 | ... | |
662 | ||
663 | Nested keyed | |
664 | ||
665 | KEY0 SUB_KEY0=VAL00 SUB_KEY1=VAL01... | |
666 | KEY1 SUB_KEY0=VAL10 SUB_KEY1=VAL11... | |
667 | ... | |
668 | ||
669 | For a writable file, the format for writing should generally match | |
670 | reading; however, controllers may allow omitting later fields or | |
671 | implement restricted shortcuts for most common use cases. | |
672 | ||
673 | For both flat and nested keyed files, only the values for a single key | |
674 | can be written at a time. For nested keyed files, the sub key pairs | |
675 | may be specified in any order and not all pairs have to be specified. | |
676 | ||
677 | ||
633b11be MCC |
678 | Conventions |
679 | ----------- | |
6c292092 TH |
680 | |
681 | - Settings for a single feature should be contained in a single file. | |
682 | ||
683 | - The root cgroup should be exempt from resource control and thus | |
684 | shouldn't have resource control interface files. Also, | |
685 | informational files on the root cgroup which end up showing global | |
686 | information available elsewhere shouldn't exist. | |
687 | ||
688 | - If a controller implements weight based resource distribution, its | |
689 | interface file should be named "weight" and have the range [1, | |
690 | 10000] with 100 as the default. The values are chosen to allow | |
691 | enough and symmetric bias in both directions while keeping it | |
692 | intuitive (the default is 100%). | |
693 | ||
694 | - If a controller implements an absolute resource guarantee and/or | |
695 | limit, the interface files should be named "min" and "max" | |
696 | respectively. If a controller implements best effort resource | |
697 | guarantee and/or limit, the interface files should be named "low" | |
698 | and "high" respectively. | |
699 | ||
700 | In the above four control files, the special token "max" should be | |
701 | used to represent upward infinity for both reading and writing. | |
702 | ||
703 | - If a setting has a configurable default value and keyed specific | |
704 | overrides, the default entry should be keyed with "default" and | |
705 | appear as the first entry in the file. | |
706 | ||
707 | The default value can be updated by writing either "default $VAL" or | |
708 | "$VAL". | |
709 | ||
710 | When writing to update a specific override, "default" can be used as | |
711 | the value to indicate removal of the override. Override entries | |
712 | with "default" as the value must not appear when read. | |
713 | ||
714 | For example, a setting which is keyed by major:minor device numbers | |
633b11be | 715 | with integer values may look like the following:: |
6c292092 TH |
716 | |
717 | # cat cgroup-example-interface-file | |
718 | default 150 | |
719 | 8:0 300 | |
720 | ||
633b11be | 721 | The default value can be updated by:: |
6c292092 TH |
722 | |
723 | # echo 125 > cgroup-example-interface-file | |
724 | ||
633b11be | 725 | or:: |
6c292092 TH |
726 | |
727 | # echo "default 125" > cgroup-example-interface-file | |
728 | ||
633b11be | 729 | An override can be set by:: |
6c292092 TH |
730 | |
731 | # echo "8:16 170" > cgroup-example-interface-file | |
732 | ||
633b11be | 733 | and cleared by:: |
6c292092 TH |
734 | |
735 | # echo "8:0 default" > cgroup-example-interface-file | |
736 | # cat cgroup-example-interface-file | |
737 | default 125 | |
738 | 8:16 170 | |
739 | ||
740 | - For events which are not very high frequency, an interface file | |
741 | "events" should be created which lists event key value pairs. | |
742 | Whenever a notifiable event happens, file modified event should be | |
743 | generated on the file. | |
744 | ||
745 | ||
633b11be MCC |
746 | Core Interface Files |
747 | -------------------- | |
6c292092 TH |
748 | |
749 | All cgroup core files are prefixed with "cgroup." | |
750 | ||
8cfd8147 TH |
751 | cgroup.type |
752 | ||
753 | A read-write single value file which exists on non-root | |
754 | cgroups. | |
755 | ||
756 | When read, it indicates the current type of the cgroup, which | |
757 | can be one of the following values. | |
758 | ||
759 | - "domain" : A normal valid domain cgroup. | |
760 | ||
761 | - "domain threaded" : A threaded domain cgroup which is | |
762 | serving as the root of a threaded subtree. | |
763 | ||
764 | - "domain invalid" : A cgroup which is in an invalid state. | |
765 | It can't be populated or have controllers enabled. It may | |
766 | be allowed to become a threaded cgroup. | |
767 | ||
768 | - "threaded" : A threaded cgroup which is a member of a | |
769 | threaded subtree. | |
770 | ||
771 | A cgroup can be turned into a threaded cgroup by writing | |
772 | "threaded" to this file. | |
773 | ||
6c292092 | 774 | cgroup.procs |
6c292092 TH |
775 | A read-write new-line separated values file which exists on |
776 | all cgroups. | |
777 | ||
778 | When read, it lists the PIDs of all processes which belong to | |
779 | the cgroup one-per-line. The PIDs are not ordered and the | |
780 | same PID may show up more than once if the process got moved | |
781 | to another cgroup and then back or the PID got recycled while | |
782 | reading. | |
783 | ||
784 | A PID can be written to migrate the process associated with | |
785 | the PID to the cgroup. The writer should match all of the | |
786 | following conditions. | |
787 | ||
6c292092 | 788 | - It must have write access to the "cgroup.procs" file. |
8cfd8147 TH |
789 | |
790 | - It must have write access to the "cgroup.procs" file of the | |
791 | common ancestor of the source and destination cgroups. | |
792 | ||
793 | When delegating a sub-hierarchy, write access to this file | |
794 | should be granted along with the containing directory. | |
795 | ||
796 | In a threaded cgroup, reading this file fails with EOPNOTSUPP | |
797 | as all the processes belong to the thread root. Writing is | |
798 | supported and moves every thread of the process to the cgroup. | |
799 | ||
800 | cgroup.threads | |
801 | A read-write new-line separated values file which exists on | |
802 | all cgroups. | |
803 | ||
804 | When read, it lists the TIDs of all threads which belong to | |
805 | the cgroup one-per-line. The TIDs are not ordered and the | |
806 | same TID may show up more than once if the thread got moved to | |
807 | another cgroup and then back or the TID got recycled while | |
808 | reading. | |
809 | ||
810 | A TID can be written to migrate the thread associated with the | |
811 | TID to the cgroup. The writer should match all of the | |
812 | following conditions. | |
813 | ||
814 | - It must have write access to the "cgroup.threads" file. | |
815 | ||
816 | - The cgroup that the thread is currently in must be in the | |
817 | same resource domain as the destination cgroup. | |
6c292092 TH |
818 | |
819 | - It must have write access to the "cgroup.procs" file of the | |
820 | common ancestor of the source and destination cgroups. | |
821 | ||
822 | When delegating a sub-hierarchy, write access to this file | |
823 | should be granted along with the containing directory. | |
824 | ||
825 | cgroup.controllers | |
6c292092 TH |
826 | A read-only space separated values file which exists on all |
827 | cgroups. | |
828 | ||
829 | It shows space separated list of all controllers available to | |
830 | the cgroup. The controllers are not ordered. | |
831 | ||
832 | cgroup.subtree_control | |
6c292092 TH |
833 | A read-write space separated values file which exists on all |
834 | cgroups. Starts out empty. | |
835 | ||
836 | When read, it shows space separated list of the controllers | |
837 | which are enabled to control resource distribution from the | |
838 | cgroup to its children. | |
839 | ||
840 | Space separated list of controllers prefixed with '+' or '-' | |
841 | can be written to enable or disable controllers. A controller | |
842 | name prefixed with '+' enables the controller and '-' | |
843 | disables. If a controller appears more than once on the list, | |
844 | the last one is effective. When multiple enable and disable | |
845 | operations are specified, either all succeed or all fail. | |
846 | ||
847 | cgroup.events | |
6c292092 TH |
848 | A read-only flat-keyed file which exists on non-root cgroups. |
849 | The following entries are defined. Unless specified | |
850 | otherwise, a value change in this file generates a file | |
851 | modified event. | |
852 | ||
853 | populated | |
6c292092 TH |
854 | 1 if the cgroup or its descendants contains any live |
855 | processes; otherwise, 0. | |
856 | ||
1a926e0b RG |
857 | cgroup.max.descendants |
858 | A read-write single value files. The default is "max". | |
859 | ||
860 | Maximum allowed number of descent cgroups. | |
861 | If the actual number of descendants is equal or larger, | |
862 | an attempt to create a new cgroup in the hierarchy will fail. | |
863 | ||
864 | cgroup.max.depth | |
865 | A read-write single value files. The default is "max". | |
866 | ||
867 | Maximum allowed descent depth below the current cgroup. | |
868 | If the actual descent depth is equal or larger, | |
869 | an attempt to create a new child cgroup will fail. | |
870 | ||
ec39225c RG |
871 | cgroup.stat |
872 | A read-only flat-keyed file with the following entries: | |
873 | ||
874 | nr_descendants | |
875 | Total number of visible descendant cgroups. | |
876 | ||
877 | nr_dying_descendants | |
878 | Total number of dying descendant cgroups. A cgroup becomes | |
879 | dying after being deleted by a user. The cgroup will remain | |
880 | in dying state for some time undefined time (which can depend | |
881 | on system load) before being completely destroyed. | |
882 | ||
883 | A process can't enter a dying cgroup under any circumstances, | |
884 | a dying cgroup can't revive. | |
885 | ||
886 | A dying cgroup can consume system resources not exceeding | |
887 | limits, which were active at the moment of cgroup deletion. | |
888 | ||
6c292092 | 889 | |
633b11be MCC |
890 | Controllers |
891 | =========== | |
6c292092 | 892 | |
633b11be MCC |
893 | CPU |
894 | --- | |
6c292092 | 895 | |
633b11be MCC |
896 | .. note:: |
897 | ||
898 | The interface for the cpu controller hasn't been merged yet | |
6c292092 TH |
899 | |
900 | The "cpu" controllers regulates distribution of CPU cycles. This | |
901 | controller implements weight and absolute bandwidth limit models for | |
902 | normal scheduling policy and absolute bandwidth allocation model for | |
903 | realtime scheduling policy. | |
904 | ||
905 | ||
633b11be MCC |
906 | CPU Interface Files |
907 | ~~~~~~~~~~~~~~~~~~~ | |
6c292092 TH |
908 | |
909 | All time durations are in microseconds. | |
910 | ||
911 | cpu.stat | |
6c292092 TH |
912 | A read-only flat-keyed file which exists on non-root cgroups. |
913 | ||
633b11be | 914 | It reports the following six stats: |
6c292092 | 915 | |
633b11be MCC |
916 | - usage_usec |
917 | - user_usec | |
918 | - system_usec | |
919 | - nr_periods | |
920 | - nr_throttled | |
921 | - throttled_usec | |
6c292092 TH |
922 | |
923 | cpu.weight | |
6c292092 TH |
924 | A read-write single value file which exists on non-root |
925 | cgroups. The default is "100". | |
926 | ||
927 | The weight in the range [1, 10000]. | |
928 | ||
929 | cpu.max | |
6c292092 TH |
930 | A read-write two value file which exists on non-root cgroups. |
931 | The default is "max 100000". | |
932 | ||
633b11be | 933 | The maximum bandwidth limit. It's in the following format:: |
6c292092 TH |
934 | |
935 | $MAX $PERIOD | |
936 | ||
937 | which indicates that the group may consume upto $MAX in each | |
938 | $PERIOD duration. "max" for $MAX indicates no limit. If only | |
939 | one number is written, $MAX is updated. | |
940 | ||
941 | cpu.rt.max | |
633b11be | 942 | .. note:: |
6c292092 | 943 | |
633b11be MCC |
944 | The semantics of this file is still under discussion and the |
945 | interface hasn't been merged yet | |
6c292092 TH |
946 | |
947 | A read-write two value file which exists on all cgroups. | |
948 | The default is "0 100000". | |
949 | ||
950 | The maximum realtime runtime allocation. Over-committing | |
951 | configurations are disallowed and process migrations are | |
952 | rejected if not enough bandwidth is available. It's in the | |
633b11be | 953 | following format:: |
6c292092 TH |
954 | |
955 | $MAX $PERIOD | |
956 | ||
957 | which indicates that the group may consume upto $MAX in each | |
958 | $PERIOD duration. If only one number is written, $MAX is | |
959 | updated. | |
960 | ||
961 | ||
633b11be MCC |
962 | Memory |
963 | ------ | |
6c292092 TH |
964 | |
965 | The "memory" controller regulates distribution of memory. Memory is | |
966 | stateful and implements both limit and protection models. Due to the | |
967 | intertwining between memory usage and reclaim pressure and the | |
968 | stateful nature of memory, the distribution model is relatively | |
969 | complex. | |
970 | ||
971 | While not completely water-tight, all major memory usages by a given | |
972 | cgroup are tracked so that the total memory consumption can be | |
973 | accounted and controlled to a reasonable extent. Currently, the | |
974 | following types of memory usages are tracked. | |
975 | ||
976 | - Userland memory - page cache and anonymous memory. | |
977 | ||
978 | - Kernel data structures such as dentries and inodes. | |
979 | ||
980 | - TCP socket buffers. | |
981 | ||
982 | The above list may expand in the future for better coverage. | |
983 | ||
984 | ||
633b11be MCC |
985 | Memory Interface Files |
986 | ~~~~~~~~~~~~~~~~~~~~~~ | |
6c292092 TH |
987 | |
988 | All memory amounts are in bytes. If a value which is not aligned to | |
989 | PAGE_SIZE is written, the value may be rounded up to the closest | |
990 | PAGE_SIZE multiple when read back. | |
991 | ||
992 | memory.current | |
6c292092 TH |
993 | A read-only single value file which exists on non-root |
994 | cgroups. | |
995 | ||
996 | The total amount of memory currently being used by the cgroup | |
997 | and its descendants. | |
998 | ||
999 | memory.low | |
6c292092 TH |
1000 | A read-write single value file which exists on non-root |
1001 | cgroups. The default is "0". | |
1002 | ||
1003 | Best-effort memory protection. If the memory usages of a | |
1004 | cgroup and all its ancestors are below their low boundaries, | |
1005 | the cgroup's memory won't be reclaimed unless memory can be | |
1006 | reclaimed from unprotected cgroups. | |
1007 | ||
1008 | Putting more memory than generally available under this | |
1009 | protection is discouraged. | |
1010 | ||
1011 | memory.high | |
6c292092 TH |
1012 | A read-write single value file which exists on non-root |
1013 | cgroups. The default is "max". | |
1014 | ||
1015 | Memory usage throttle limit. This is the main mechanism to | |
1016 | control memory usage of a cgroup. If a cgroup's usage goes | |
1017 | over the high boundary, the processes of the cgroup are | |
1018 | throttled and put under heavy reclaim pressure. | |
1019 | ||
1020 | Going over the high limit never invokes the OOM killer and | |
1021 | under extreme conditions the limit may be breached. | |
1022 | ||
1023 | memory.max | |
6c292092 TH |
1024 | A read-write single value file which exists on non-root |
1025 | cgroups. The default is "max". | |
1026 | ||
1027 | Memory usage hard limit. This is the final protection | |
1028 | mechanism. If a cgroup's memory usage reaches this limit and | |
1029 | can't be reduced, the OOM killer is invoked in the cgroup. | |
1030 | Under certain circumstances, the usage may go over the limit | |
1031 | temporarily. | |
1032 | ||
1033 | This is the ultimate protection mechanism. As long as the | |
1034 | high limit is used and monitored properly, this limit's | |
1035 | utility is limited to providing the final safety net. | |
1036 | ||
1037 | memory.events | |
6c292092 TH |
1038 | A read-only flat-keyed file which exists on non-root cgroups. |
1039 | The following entries are defined. Unless specified | |
1040 | otherwise, a value change in this file generates a file | |
1041 | modified event. | |
1042 | ||
1043 | low | |
6c292092 TH |
1044 | The number of times the cgroup is reclaimed due to |
1045 | high memory pressure even though its usage is under | |
1046 | the low boundary. This usually indicates that the low | |
1047 | boundary is over-committed. | |
1048 | ||
1049 | high | |
6c292092 TH |
1050 | The number of times processes of the cgroup are |
1051 | throttled and routed to perform direct memory reclaim | |
1052 | because the high memory boundary was exceeded. For a | |
1053 | cgroup whose memory usage is capped by the high limit | |
1054 | rather than global memory pressure, this event's | |
1055 | occurrences are expected. | |
1056 | ||
1057 | max | |
6c292092 TH |
1058 | The number of times the cgroup's memory usage was |
1059 | about to go over the max boundary. If direct reclaim | |
8e675f7a | 1060 | fails to bring it down, the cgroup goes to OOM state. |
6c292092 TH |
1061 | |
1062 | oom | |
8e675f7a KK |
1063 | The number of time the cgroup's memory usage was |
1064 | reached the limit and allocation was about to fail. | |
1065 | ||
1066 | Depending on context result could be invocation of OOM | |
1067 | killer and retrying allocation or failing alloction. | |
1068 | ||
1069 | Failed allocation in its turn could be returned into | |
1070 | userspace as -ENOMEM or siletly ignored in cases like | |
633b11be | 1071 | disk readahead. For now OOM in memory cgroup kills |
8e675f7a KK |
1072 | tasks iff shortage has happened inside page fault. |
1073 | ||
1074 | oom_kill | |
8e675f7a KK |
1075 | The number of processes belonging to this cgroup |
1076 | killed by any kind of OOM killer. | |
6c292092 | 1077 | |
587d9f72 | 1078 | memory.stat |
587d9f72 JW |
1079 | A read-only flat-keyed file which exists on non-root cgroups. |
1080 | ||
1081 | This breaks down the cgroup's memory footprint into different | |
1082 | types of memory, type-specific details, and other information | |
1083 | on the state and past events of the memory management system. | |
1084 | ||
1085 | All memory amounts are in bytes. | |
1086 | ||
1087 | The entries are ordered to be human readable, and new entries | |
1088 | can show up in the middle. Don't rely on items remaining in a | |
1089 | fixed position; use the keys to look up specific values! | |
1090 | ||
1091 | anon | |
587d9f72 JW |
1092 | Amount of memory used in anonymous mappings such as |
1093 | brk(), sbrk(), and mmap(MAP_ANONYMOUS) | |
1094 | ||
1095 | file | |
587d9f72 JW |
1096 | Amount of memory used to cache filesystem data, |
1097 | including tmpfs and shared memory. | |
1098 | ||
12580e4b | 1099 | kernel_stack |
12580e4b VD |
1100 | Amount of memory allocated to kernel stacks. |
1101 | ||
27ee57c9 | 1102 | slab |
27ee57c9 VD |
1103 | Amount of memory used for storing in-kernel data |
1104 | structures. | |
1105 | ||
4758e198 | 1106 | sock |
4758e198 JW |
1107 | Amount of memory used in network transmission buffers |
1108 | ||
9a4caf1e | 1109 | shmem |
9a4caf1e JW |
1110 | Amount of cached filesystem data that is swap-backed, |
1111 | such as tmpfs, shm segments, shared anonymous mmap()s | |
1112 | ||
587d9f72 | 1113 | file_mapped |
587d9f72 JW |
1114 | Amount of cached filesystem data mapped with mmap() |
1115 | ||
1116 | file_dirty | |
587d9f72 JW |
1117 | Amount of cached filesystem data that was modified but |
1118 | not yet written back to disk | |
1119 | ||
1120 | file_writeback | |
587d9f72 JW |
1121 | Amount of cached filesystem data that was modified and |
1122 | is currently being written back to disk | |
1123 | ||
633b11be | 1124 | inactive_anon, active_anon, inactive_file, active_file, unevictable |
587d9f72 JW |
1125 | Amount of memory, swap-backed and filesystem-backed, |
1126 | on the internal memory management lists used by the | |
1127 | page reclaim algorithm | |
1128 | ||
27ee57c9 | 1129 | slab_reclaimable |
27ee57c9 VD |
1130 | Part of "slab" that might be reclaimed, such as |
1131 | dentries and inodes. | |
1132 | ||
1133 | slab_unreclaimable | |
27ee57c9 VD |
1134 | Part of "slab" that cannot be reclaimed on memory |
1135 | pressure. | |
1136 | ||
587d9f72 | 1137 | pgfault |
587d9f72 JW |
1138 | Total number of page faults incurred |
1139 | ||
1140 | pgmajfault | |
587d9f72 JW |
1141 | Number of major page faults incurred |
1142 | ||
b340959e RG |
1143 | workingset_refault |
1144 | ||
1145 | Number of refaults of previously evicted pages | |
1146 | ||
1147 | workingset_activate | |
1148 | ||
1149 | Number of refaulted pages that were immediately activated | |
1150 | ||
1151 | workingset_nodereclaim | |
1152 | ||
1153 | Number of times a shadow node has been reclaimed | |
1154 | ||
2262185c RG |
1155 | pgrefill |
1156 | ||
1157 | Amount of scanned pages (in an active LRU list) | |
1158 | ||
1159 | pgscan | |
1160 | ||
1161 | Amount of scanned pages (in an inactive LRU list) | |
1162 | ||
1163 | pgsteal | |
1164 | ||
1165 | Amount of reclaimed pages | |
1166 | ||
1167 | pgactivate | |
1168 | ||
1169 | Amount of pages moved to the active LRU list | |
1170 | ||
1171 | pgdeactivate | |
1172 | ||
1173 | Amount of pages moved to the inactive LRU lis | |
1174 | ||
1175 | pglazyfree | |
1176 | ||
1177 | Amount of pages postponed to be freed under memory pressure | |
1178 | ||
1179 | pglazyfreed | |
1180 | ||
1181 | Amount of reclaimed lazyfree pages | |
1182 | ||
3e24b19d | 1183 | memory.swap.current |
3e24b19d VD |
1184 | A read-only single value file which exists on non-root |
1185 | cgroups. | |
1186 | ||
1187 | The total amount of swap currently being used by the cgroup | |
1188 | and its descendants. | |
1189 | ||
1190 | memory.swap.max | |
3e24b19d VD |
1191 | A read-write single value file which exists on non-root |
1192 | cgroups. The default is "max". | |
1193 | ||
1194 | Swap usage hard limit. If a cgroup's swap usage reaches this | |
1195 | limit, anonymous meomry of the cgroup will not be swapped out. | |
1196 | ||
6c292092 | 1197 | |
633b11be MCC |
1198 | Usage Guidelines |
1199 | ~~~~~~~~~~~~~~~~ | |
6c292092 TH |
1200 | |
1201 | "memory.high" is the main mechanism to control memory usage. | |
1202 | Over-committing on high limit (sum of high limits > available memory) | |
1203 | and letting global memory pressure to distribute memory according to | |
1204 | usage is a viable strategy. | |
1205 | ||
1206 | Because breach of the high limit doesn't trigger the OOM killer but | |
1207 | throttles the offending cgroup, a management agent has ample | |
1208 | opportunities to monitor and take appropriate actions such as granting | |
1209 | more memory or terminating the workload. | |
1210 | ||
1211 | Determining whether a cgroup has enough memory is not trivial as | |
1212 | memory usage doesn't indicate whether the workload can benefit from | |
1213 | more memory. For example, a workload which writes data received from | |
1214 | network to a file can use all available memory but can also operate as | |
1215 | performant with a small amount of memory. A measure of memory | |
1216 | pressure - how much the workload is being impacted due to lack of | |
1217 | memory - is necessary to determine whether a workload needs more | |
1218 | memory; unfortunately, memory pressure monitoring mechanism isn't | |
1219 | implemented yet. | |
1220 | ||
1221 | ||
633b11be MCC |
1222 | Memory Ownership |
1223 | ~~~~~~~~~~~~~~~~ | |
6c292092 TH |
1224 | |
1225 | A memory area is charged to the cgroup which instantiated it and stays | |
1226 | charged to the cgroup until the area is released. Migrating a process | |
1227 | to a different cgroup doesn't move the memory usages that it | |
1228 | instantiated while in the previous cgroup to the new cgroup. | |
1229 | ||
1230 | A memory area may be used by processes belonging to different cgroups. | |
1231 | To which cgroup the area will be charged is in-deterministic; however, | |
1232 | over time, the memory area is likely to end up in a cgroup which has | |
1233 | enough memory allowance to avoid high reclaim pressure. | |
1234 | ||
1235 | If a cgroup sweeps a considerable amount of memory which is expected | |
1236 | to be accessed repeatedly by other cgroups, it may make sense to use | |
1237 | POSIX_FADV_DONTNEED to relinquish the ownership of memory areas | |
1238 | belonging to the affected files to ensure correct memory ownership. | |
1239 | ||
1240 | ||
633b11be MCC |
1241 | IO |
1242 | -- | |
6c292092 TH |
1243 | |
1244 | The "io" controller regulates the distribution of IO resources. This | |
1245 | controller implements both weight based and absolute bandwidth or IOPS | |
1246 | limit distribution; however, weight based distribution is available | |
1247 | only if cfq-iosched is in use and neither scheme is available for | |
1248 | blk-mq devices. | |
1249 | ||
1250 | ||
633b11be MCC |
1251 | IO Interface Files |
1252 | ~~~~~~~~~~~~~~~~~~ | |
6c292092 TH |
1253 | |
1254 | io.stat | |
6c292092 TH |
1255 | A read-only nested-keyed file which exists on non-root |
1256 | cgroups. | |
1257 | ||
1258 | Lines are keyed by $MAJ:$MIN device numbers and not ordered. | |
1259 | The following nested keys are defined. | |
1260 | ||
633b11be | 1261 | ====== =================== |
6c292092 TH |
1262 | rbytes Bytes read |
1263 | wbytes Bytes written | |
1264 | rios Number of read IOs | |
1265 | wios Number of write IOs | |
633b11be | 1266 | ====== =================== |
6c292092 | 1267 | |
633b11be | 1268 | An example read output follows: |
6c292092 TH |
1269 | |
1270 | 8:16 rbytes=1459200 wbytes=314773504 rios=192 wios=353 | |
1271 | 8:0 rbytes=90430464 wbytes=299008000 rios=8950 wios=1252 | |
1272 | ||
1273 | io.weight | |
6c292092 TH |
1274 | A read-write flat-keyed file which exists on non-root cgroups. |
1275 | The default is "default 100". | |
1276 | ||
1277 | The first line is the default weight applied to devices | |
1278 | without specific override. The rest are overrides keyed by | |
1279 | $MAJ:$MIN device numbers and not ordered. The weights are in | |
1280 | the range [1, 10000] and specifies the relative amount IO time | |
1281 | the cgroup can use in relation to its siblings. | |
1282 | ||
1283 | The default weight can be updated by writing either "default | |
1284 | $WEIGHT" or simply "$WEIGHT". Overrides can be set by writing | |
1285 | "$MAJ:$MIN $WEIGHT" and unset by writing "$MAJ:$MIN default". | |
1286 | ||
633b11be | 1287 | An example read output follows:: |
6c292092 TH |
1288 | |
1289 | default 100 | |
1290 | 8:16 200 | |
1291 | 8:0 50 | |
1292 | ||
1293 | io.max | |
6c292092 TH |
1294 | A read-write nested-keyed file which exists on non-root |
1295 | cgroups. | |
1296 | ||
1297 | BPS and IOPS based IO limit. Lines are keyed by $MAJ:$MIN | |
1298 | device numbers and not ordered. The following nested keys are | |
1299 | defined. | |
1300 | ||
633b11be | 1301 | ===== ================================== |
6c292092 TH |
1302 | rbps Max read bytes per second |
1303 | wbps Max write bytes per second | |
1304 | riops Max read IO operations per second | |
1305 | wiops Max write IO operations per second | |
633b11be | 1306 | ===== ================================== |
6c292092 TH |
1307 | |
1308 | When writing, any number of nested key-value pairs can be | |
1309 | specified in any order. "max" can be specified as the value | |
1310 | to remove a specific limit. If the same key is specified | |
1311 | multiple times, the outcome is undefined. | |
1312 | ||
1313 | BPS and IOPS are measured in each IO direction and IOs are | |
1314 | delayed if limit is reached. Temporary bursts are allowed. | |
1315 | ||
633b11be | 1316 | Setting read limit at 2M BPS and write at 120 IOPS for 8:16:: |
6c292092 TH |
1317 | |
1318 | echo "8:16 rbps=2097152 wiops=120" > io.max | |
1319 | ||
633b11be | 1320 | Reading returns the following:: |
6c292092 TH |
1321 | |
1322 | 8:16 rbps=2097152 wbps=max riops=max wiops=120 | |
1323 | ||
633b11be | 1324 | Write IOPS limit can be removed by writing the following:: |
6c292092 TH |
1325 | |
1326 | echo "8:16 wiops=max" > io.max | |
1327 | ||
633b11be | 1328 | Reading now returns the following:: |
6c292092 TH |
1329 | |
1330 | 8:16 rbps=2097152 wbps=max riops=max wiops=max | |
1331 | ||
1332 | ||
633b11be MCC |
1333 | Writeback |
1334 | ~~~~~~~~~ | |
6c292092 TH |
1335 | |
1336 | Page cache is dirtied through buffered writes and shared mmaps and | |
1337 | written asynchronously to the backing filesystem by the writeback | |
1338 | mechanism. Writeback sits between the memory and IO domains and | |
1339 | regulates the proportion of dirty memory by balancing dirtying and | |
1340 | write IOs. | |
1341 | ||
1342 | The io controller, in conjunction with the memory controller, | |
1343 | implements control of page cache writeback IOs. The memory controller | |
1344 | defines the memory domain that dirty memory ratio is calculated and | |
1345 | maintained for and the io controller defines the io domain which | |
1346 | writes out dirty pages for the memory domain. Both system-wide and | |
1347 | per-cgroup dirty memory states are examined and the more restrictive | |
1348 | of the two is enforced. | |
1349 | ||
1350 | cgroup writeback requires explicit support from the underlying | |
1351 | filesystem. Currently, cgroup writeback is implemented on ext2, ext4 | |
1352 | and btrfs. On other filesystems, all writeback IOs are attributed to | |
1353 | the root cgroup. | |
1354 | ||
1355 | There are inherent differences in memory and writeback management | |
1356 | which affects how cgroup ownership is tracked. Memory is tracked per | |
1357 | page while writeback per inode. For the purpose of writeback, an | |
1358 | inode is assigned to a cgroup and all IO requests to write dirty pages | |
1359 | from the inode are attributed to that cgroup. | |
1360 | ||
1361 | As cgroup ownership for memory is tracked per page, there can be pages | |
1362 | which are associated with different cgroups than the one the inode is | |
1363 | associated with. These are called foreign pages. The writeback | |
1364 | constantly keeps track of foreign pages and, if a particular foreign | |
1365 | cgroup becomes the majority over a certain period of time, switches | |
1366 | the ownership of the inode to that cgroup. | |
1367 | ||
1368 | While this model is enough for most use cases where a given inode is | |
1369 | mostly dirtied by a single cgroup even when the main writing cgroup | |
1370 | changes over time, use cases where multiple cgroups write to a single | |
1371 | inode simultaneously are not supported well. In such circumstances, a | |
1372 | significant portion of IOs are likely to be attributed incorrectly. | |
1373 | As memory controller assigns page ownership on the first use and | |
1374 | doesn't update it until the page is released, even if writeback | |
1375 | strictly follows page ownership, multiple cgroups dirtying overlapping | |
1376 | areas wouldn't work as expected. It's recommended to avoid such usage | |
1377 | patterns. | |
1378 | ||
1379 | The sysctl knobs which affect writeback behavior are applied to cgroup | |
1380 | writeback as follows. | |
1381 | ||
633b11be | 1382 | vm.dirty_background_ratio, vm.dirty_ratio |
6c292092 TH |
1383 | These ratios apply the same to cgroup writeback with the |
1384 | amount of available memory capped by limits imposed by the | |
1385 | memory controller and system-wide clean memory. | |
1386 | ||
633b11be | 1387 | vm.dirty_background_bytes, vm.dirty_bytes |
6c292092 TH |
1388 | For cgroup writeback, this is calculated into ratio against |
1389 | total available memory and applied the same way as | |
1390 | vm.dirty[_background]_ratio. | |
1391 | ||
1392 | ||
633b11be MCC |
1393 | PID |
1394 | --- | |
20c56e59 HR |
1395 | |
1396 | The process number controller is used to allow a cgroup to stop any | |
1397 | new tasks from being fork()'d or clone()'d after a specified limit is | |
1398 | reached. | |
1399 | ||
1400 | The number of tasks in a cgroup can be exhausted in ways which other | |
1401 | controllers cannot prevent, thus warranting its own controller. For | |
1402 | example, a fork bomb is likely to exhaust the number of tasks before | |
1403 | hitting memory restrictions. | |
1404 | ||
1405 | Note that PIDs used in this controller refer to TIDs, process IDs as | |
1406 | used by the kernel. | |
1407 | ||
1408 | ||
633b11be MCC |
1409 | PID Interface Files |
1410 | ~~~~~~~~~~~~~~~~~~~ | |
20c56e59 HR |
1411 | |
1412 | pids.max | |
312eb712 TK |
1413 | A read-write single value file which exists on non-root |
1414 | cgroups. The default is "max". | |
20c56e59 | 1415 | |
312eb712 | 1416 | Hard limit of number of processes. |
20c56e59 HR |
1417 | |
1418 | pids.current | |
312eb712 | 1419 | A read-only single value file which exists on all cgroups. |
20c56e59 | 1420 | |
312eb712 TK |
1421 | The number of processes currently in the cgroup and its |
1422 | descendants. | |
20c56e59 HR |
1423 | |
1424 | Organisational operations are not blocked by cgroup policies, so it is | |
1425 | possible to have pids.current > pids.max. This can be done by either | |
1426 | setting the limit to be smaller than pids.current, or attaching enough | |
1427 | processes to the cgroup such that pids.current is larger than | |
1428 | pids.max. However, it is not possible to violate a cgroup PID policy | |
1429 | through fork() or clone(). These will return -EAGAIN if the creation | |
1430 | of a new process would cause a cgroup policy to be violated. | |
1431 | ||
1432 | ||
633b11be MCC |
1433 | RDMA |
1434 | ---- | |
968ebff1 | 1435 | |
9c1e67f9 PP |
1436 | The "rdma" controller regulates the distribution and accounting of |
1437 | of RDMA resources. | |
1438 | ||
633b11be MCC |
1439 | RDMA Interface Files |
1440 | ~~~~~~~~~~~~~~~~~~~~ | |
9c1e67f9 PP |
1441 | |
1442 | rdma.max | |
1443 | A readwrite nested-keyed file that exists for all the cgroups | |
1444 | except root that describes current configured resource limit | |
1445 | for a RDMA/IB device. | |
1446 | ||
1447 | Lines are keyed by device name and are not ordered. | |
1448 | Each line contains space separated resource name and its configured | |
1449 | limit that can be distributed. | |
1450 | ||
1451 | The following nested keys are defined. | |
1452 | ||
633b11be | 1453 | ========== ============================= |
9c1e67f9 PP |
1454 | hca_handle Maximum number of HCA Handles |
1455 | hca_object Maximum number of HCA Objects | |
633b11be | 1456 | ========== ============================= |
9c1e67f9 | 1457 | |
633b11be | 1458 | An example for mlx4 and ocrdma device follows:: |
9c1e67f9 PP |
1459 | |
1460 | mlx4_0 hca_handle=2 hca_object=2000 | |
1461 | ocrdma1 hca_handle=3 hca_object=max | |
1462 | ||
1463 | rdma.current | |
1464 | A read-only file that describes current resource usage. | |
1465 | It exists for all the cgroup except root. | |
1466 | ||
633b11be | 1467 | An example for mlx4 and ocrdma device follows:: |
9c1e67f9 PP |
1468 | |
1469 | mlx4_0 hca_handle=1 hca_object=20 | |
1470 | ocrdma1 hca_handle=1 hca_object=23 | |
1471 | ||
1472 | ||
633b11be MCC |
1473 | Misc |
1474 | ---- | |
63f1ca59 | 1475 | |
633b11be MCC |
1476 | perf_event |
1477 | ~~~~~~~~~~ | |
968ebff1 TH |
1478 | |
1479 | perf_event controller, if not mounted on a legacy hierarchy, is | |
1480 | automatically enabled on the v2 hierarchy so that perf events can | |
1481 | always be filtered by cgroup v2 path. The controller can still be | |
1482 | moved to a legacy hierarchy after v2 hierarchy is populated. | |
1483 | ||
1484 | ||
633b11be MCC |
1485 | Namespace |
1486 | ========= | |
d4021f6c | 1487 | |
633b11be MCC |
1488 | Basics |
1489 | ------ | |
d4021f6c SH |
1490 | |
1491 | cgroup namespace provides a mechanism to virtualize the view of the | |
1492 | "/proc/$PID/cgroup" file and cgroup mounts. The CLONE_NEWCGROUP clone | |
1493 | flag can be used with clone(2) and unshare(2) to create a new cgroup | |
1494 | namespace. The process running inside the cgroup namespace will have | |
1495 | its "/proc/$PID/cgroup" output restricted to cgroupns root. The | |
1496 | cgroupns root is the cgroup of the process at the time of creation of | |
1497 | the cgroup namespace. | |
1498 | ||
1499 | Without cgroup namespace, the "/proc/$PID/cgroup" file shows the | |
1500 | complete path of the cgroup of a process. In a container setup where | |
1501 | a set of cgroups and namespaces are intended to isolate processes the | |
1502 | "/proc/$PID/cgroup" file may leak potential system level information | |
633b11be | 1503 | to the isolated processes. For Example:: |
d4021f6c SH |
1504 | |
1505 | # cat /proc/self/cgroup | |
1506 | 0::/batchjobs/container_id1 | |
1507 | ||
1508 | The path '/batchjobs/container_id1' can be considered as system-data | |
1509 | and undesirable to expose to the isolated processes. cgroup namespace | |
1510 | can be used to restrict visibility of this path. For example, before | |
633b11be | 1511 | creating a cgroup namespace, one would see:: |
d4021f6c SH |
1512 | |
1513 | # ls -l /proc/self/ns/cgroup | |
1514 | lrwxrwxrwx 1 root root 0 2014-07-15 10:37 /proc/self/ns/cgroup -> cgroup:[4026531835] | |
1515 | # cat /proc/self/cgroup | |
1516 | 0::/batchjobs/container_id1 | |
1517 | ||
633b11be | 1518 | After unsharing a new namespace, the view changes:: |
d4021f6c SH |
1519 | |
1520 | # ls -l /proc/self/ns/cgroup | |
1521 | lrwxrwxrwx 1 root root 0 2014-07-15 10:35 /proc/self/ns/cgroup -> cgroup:[4026532183] | |
1522 | # cat /proc/self/cgroup | |
1523 | 0::/ | |
1524 | ||
1525 | When some thread from a multi-threaded process unshares its cgroup | |
1526 | namespace, the new cgroupns gets applied to the entire process (all | |
1527 | the threads). This is natural for the v2 hierarchy; however, for the | |
1528 | legacy hierarchies, this may be unexpected. | |
1529 | ||
1530 | A cgroup namespace is alive as long as there are processes inside or | |
1531 | mounts pinning it. When the last usage goes away, the cgroup | |
1532 | namespace is destroyed. The cgroupns root and the actual cgroups | |
1533 | remain. | |
1534 | ||
1535 | ||
633b11be MCC |
1536 | The Root and Views |
1537 | ------------------ | |
d4021f6c SH |
1538 | |
1539 | The 'cgroupns root' for a cgroup namespace is the cgroup in which the | |
1540 | process calling unshare(2) is running. For example, if a process in | |
1541 | /batchjobs/container_id1 cgroup calls unshare, cgroup | |
1542 | /batchjobs/container_id1 becomes the cgroupns root. For the | |
1543 | init_cgroup_ns, this is the real root ('/') cgroup. | |
1544 | ||
1545 | The cgroupns root cgroup does not change even if the namespace creator | |
633b11be | 1546 | process later moves to a different cgroup:: |
d4021f6c SH |
1547 | |
1548 | # ~/unshare -c # unshare cgroupns in some cgroup | |
1549 | # cat /proc/self/cgroup | |
1550 | 0::/ | |
1551 | # mkdir sub_cgrp_1 | |
1552 | # echo 0 > sub_cgrp_1/cgroup.procs | |
1553 | # cat /proc/self/cgroup | |
1554 | 0::/sub_cgrp_1 | |
1555 | ||
1556 | Each process gets its namespace-specific view of "/proc/$PID/cgroup" | |
1557 | ||
1558 | Processes running inside the cgroup namespace will be able to see | |
1559 | cgroup paths (in /proc/self/cgroup) only inside their root cgroup. | |
633b11be | 1560 | From within an unshared cgroupns:: |
d4021f6c SH |
1561 | |
1562 | # sleep 100000 & | |
1563 | [1] 7353 | |
1564 | # echo 7353 > sub_cgrp_1/cgroup.procs | |
1565 | # cat /proc/7353/cgroup | |
1566 | 0::/sub_cgrp_1 | |
1567 | ||
1568 | From the initial cgroup namespace, the real cgroup path will be | |
633b11be | 1569 | visible:: |
d4021f6c SH |
1570 | |
1571 | $ cat /proc/7353/cgroup | |
1572 | 0::/batchjobs/container_id1/sub_cgrp_1 | |
1573 | ||
1574 | From a sibling cgroup namespace (that is, a namespace rooted at a | |
1575 | different cgroup), the cgroup path relative to its own cgroup | |
1576 | namespace root will be shown. For instance, if PID 7353's cgroup | |
633b11be | 1577 | namespace root is at '/batchjobs/container_id2', then it will see:: |
d4021f6c SH |
1578 | |
1579 | # cat /proc/7353/cgroup | |
1580 | 0::/../container_id2/sub_cgrp_1 | |
1581 | ||
1582 | Note that the relative path always starts with '/' to indicate that | |
1583 | its relative to the cgroup namespace root of the caller. | |
1584 | ||
1585 | ||
633b11be MCC |
1586 | Migration and setns(2) |
1587 | ---------------------- | |
d4021f6c SH |
1588 | |
1589 | Processes inside a cgroup namespace can move into and out of the | |
1590 | namespace root if they have proper access to external cgroups. For | |
1591 | example, from inside a namespace with cgroupns root at | |
1592 | /batchjobs/container_id1, and assuming that the global hierarchy is | |
633b11be | 1593 | still accessible inside cgroupns:: |
d4021f6c SH |
1594 | |
1595 | # cat /proc/7353/cgroup | |
1596 | 0::/sub_cgrp_1 | |
1597 | # echo 7353 > batchjobs/container_id2/cgroup.procs | |
1598 | # cat /proc/7353/cgroup | |
1599 | 0::/../container_id2 | |
1600 | ||
1601 | Note that this kind of setup is not encouraged. A task inside cgroup | |
1602 | namespace should only be exposed to its own cgroupns hierarchy. | |
1603 | ||
1604 | setns(2) to another cgroup namespace is allowed when: | |
1605 | ||
1606 | (a) the process has CAP_SYS_ADMIN against its current user namespace | |
1607 | (b) the process has CAP_SYS_ADMIN against the target cgroup | |
1608 | namespace's userns | |
1609 | ||
1610 | No implicit cgroup changes happen with attaching to another cgroup | |
1611 | namespace. It is expected that the someone moves the attaching | |
1612 | process under the target cgroup namespace root. | |
1613 | ||
1614 | ||
633b11be MCC |
1615 | Interaction with Other Namespaces |
1616 | --------------------------------- | |
d4021f6c SH |
1617 | |
1618 | Namespace specific cgroup hierarchy can be mounted by a process | |
633b11be | 1619 | running inside a non-init cgroup namespace:: |
d4021f6c SH |
1620 | |
1621 | # mount -t cgroup2 none $MOUNT_POINT | |
1622 | ||
1623 | This will mount the unified cgroup hierarchy with cgroupns root as the | |
1624 | filesystem root. The process needs CAP_SYS_ADMIN against its user and | |
1625 | mount namespaces. | |
1626 | ||
1627 | The virtualization of /proc/self/cgroup file combined with restricting | |
1628 | the view of cgroup hierarchy by namespace-private cgroupfs mount | |
1629 | provides a properly isolated cgroup view inside the container. | |
1630 | ||
1631 | ||
633b11be MCC |
1632 | Information on Kernel Programming |
1633 | ================================= | |
6c292092 TH |
1634 | |
1635 | This section contains kernel programming information in the areas | |
1636 | where interacting with cgroup is necessary. cgroup core and | |
1637 | controllers are not covered. | |
1638 | ||
1639 | ||
633b11be MCC |
1640 | Filesystem Support for Writeback |
1641 | -------------------------------- | |
6c292092 TH |
1642 | |
1643 | A filesystem can support cgroup writeback by updating | |
1644 | address_space_operations->writepage[s]() to annotate bio's using the | |
1645 | following two functions. | |
1646 | ||
1647 | wbc_init_bio(@wbc, @bio) | |
6c292092 TH |
1648 | Should be called for each bio carrying writeback data and |
1649 | associates the bio with the inode's owner cgroup. Can be | |
1650 | called anytime between bio allocation and submission. | |
1651 | ||
1652 | wbc_account_io(@wbc, @page, @bytes) | |
6c292092 TH |
1653 | Should be called for each data segment being written out. |
1654 | While this function doesn't care exactly when it's called | |
1655 | during the writeback session, it's the easiest and most | |
1656 | natural to call it as data segments are added to a bio. | |
1657 | ||
1658 | With writeback bio's annotated, cgroup support can be enabled per | |
1659 | super_block by setting SB_I_CGROUPWB in ->s_iflags. This allows for | |
1660 | selective disabling of cgroup writeback support which is helpful when | |
1661 | certain filesystem features, e.g. journaled data mode, are | |
1662 | incompatible. | |
1663 | ||
1664 | wbc_init_bio() binds the specified bio to its cgroup. Depending on | |
1665 | the configuration, the bio may be executed at a lower priority and if | |
1666 | the writeback session is holding shared resources, e.g. a journal | |
1667 | entry, may lead to priority inversion. There is no one easy solution | |
1668 | for the problem. Filesystems can try to work around specific problem | |
1669 | cases by skipping wbc_init_bio() or using bio_associate_blkcg() | |
1670 | directly. | |
1671 | ||
1672 | ||
633b11be MCC |
1673 | Deprecated v1 Core Features |
1674 | =========================== | |
6c292092 TH |
1675 | |
1676 | - Multiple hierarchies including named ones are not supported. | |
1677 | ||
5136f636 | 1678 | - All v1 mount options are not supported. |
6c292092 TH |
1679 | |
1680 | - The "tasks" file is removed and "cgroup.procs" is not sorted. | |
1681 | ||
1682 | - "cgroup.clone_children" is removed. | |
1683 | ||
1684 | - /proc/cgroups is meaningless for v2. Use "cgroup.controllers" file | |
1685 | at the root instead. | |
1686 | ||
1687 | ||
633b11be MCC |
1688 | Issues with v1 and Rationales for v2 |
1689 | ==================================== | |
6c292092 | 1690 | |
633b11be MCC |
1691 | Multiple Hierarchies |
1692 | -------------------- | |
6c292092 TH |
1693 | |
1694 | cgroup v1 allowed an arbitrary number of hierarchies and each | |
1695 | hierarchy could host any number of controllers. While this seemed to | |
1696 | provide a high level of flexibility, it wasn't useful in practice. | |
1697 | ||
1698 | For example, as there is only one instance of each controller, utility | |
1699 | type controllers such as freezer which can be useful in all | |
1700 | hierarchies could only be used in one. The issue is exacerbated by | |
1701 | the fact that controllers couldn't be moved to another hierarchy once | |
1702 | hierarchies were populated. Another issue was that all controllers | |
1703 | bound to a hierarchy were forced to have exactly the same view of the | |
1704 | hierarchy. It wasn't possible to vary the granularity depending on | |
1705 | the specific controller. | |
1706 | ||
1707 | In practice, these issues heavily limited which controllers could be | |
1708 | put on the same hierarchy and most configurations resorted to putting | |
1709 | each controller on its own hierarchy. Only closely related ones, such | |
1710 | as the cpu and cpuacct controllers, made sense to be put on the same | |
1711 | hierarchy. This often meant that userland ended up managing multiple | |
1712 | similar hierarchies repeating the same steps on each hierarchy | |
1713 | whenever a hierarchy management operation was necessary. | |
1714 | ||
1715 | Furthermore, support for multiple hierarchies came at a steep cost. | |
1716 | It greatly complicated cgroup core implementation but more importantly | |
1717 | the support for multiple hierarchies restricted how cgroup could be | |
1718 | used in general and what controllers was able to do. | |
1719 | ||
1720 | There was no limit on how many hierarchies there might be, which meant | |
1721 | that a thread's cgroup membership couldn't be described in finite | |
1722 | length. The key might contain any number of entries and was unlimited | |
1723 | in length, which made it highly awkward to manipulate and led to | |
1724 | addition of controllers which existed only to identify membership, | |
1725 | which in turn exacerbated the original problem of proliferating number | |
1726 | of hierarchies. | |
1727 | ||
1728 | Also, as a controller couldn't have any expectation regarding the | |
1729 | topologies of hierarchies other controllers might be on, each | |
1730 | controller had to assume that all other controllers were attached to | |
1731 | completely orthogonal hierarchies. This made it impossible, or at | |
1732 | least very cumbersome, for controllers to cooperate with each other. | |
1733 | ||
1734 | In most use cases, putting controllers on hierarchies which are | |
1735 | completely orthogonal to each other isn't necessary. What usually is | |
1736 | called for is the ability to have differing levels of granularity | |
1737 | depending on the specific controller. In other words, hierarchy may | |
1738 | be collapsed from leaf towards root when viewed from specific | |
1739 | controllers. For example, a given configuration might not care about | |
1740 | how memory is distributed beyond a certain level while still wanting | |
1741 | to control how CPU cycles are distributed. | |
1742 | ||
1743 | ||
633b11be MCC |
1744 | Thread Granularity |
1745 | ------------------ | |
6c292092 TH |
1746 | |
1747 | cgroup v1 allowed threads of a process to belong to different cgroups. | |
1748 | This didn't make sense for some controllers and those controllers | |
1749 | ended up implementing different ways to ignore such situations but | |
1750 | much more importantly it blurred the line between API exposed to | |
1751 | individual applications and system management interface. | |
1752 | ||
1753 | Generally, in-process knowledge is available only to the process | |
1754 | itself; thus, unlike service-level organization of processes, | |
1755 | categorizing threads of a process requires active participation from | |
1756 | the application which owns the target process. | |
1757 | ||
1758 | cgroup v1 had an ambiguously defined delegation model which got abused | |
1759 | in combination with thread granularity. cgroups were delegated to | |
1760 | individual applications so that they can create and manage their own | |
1761 | sub-hierarchies and control resource distributions along them. This | |
1762 | effectively raised cgroup to the status of a syscall-like API exposed | |
1763 | to lay programs. | |
1764 | ||
1765 | First of all, cgroup has a fundamentally inadequate interface to be | |
1766 | exposed this way. For a process to access its own knobs, it has to | |
1767 | extract the path on the target hierarchy from /proc/self/cgroup, | |
1768 | construct the path by appending the name of the knob to the path, open | |
1769 | and then read and/or write to it. This is not only extremely clunky | |
1770 | and unusual but also inherently racy. There is no conventional way to | |
1771 | define transaction across the required steps and nothing can guarantee | |
1772 | that the process would actually be operating on its own sub-hierarchy. | |
1773 | ||
1774 | cgroup controllers implemented a number of knobs which would never be | |
1775 | accepted as public APIs because they were just adding control knobs to | |
1776 | system-management pseudo filesystem. cgroup ended up with interface | |
1777 | knobs which were not properly abstracted or refined and directly | |
1778 | revealed kernel internal details. These knobs got exposed to | |
1779 | individual applications through the ill-defined delegation mechanism | |
1780 | effectively abusing cgroup as a shortcut to implementing public APIs | |
1781 | without going through the required scrutiny. | |
1782 | ||
1783 | This was painful for both userland and kernel. Userland ended up with | |
1784 | misbehaving and poorly abstracted interfaces and kernel exposing and | |
1785 | locked into constructs inadvertently. | |
1786 | ||
1787 | ||
633b11be MCC |
1788 | Competition Between Inner Nodes and Threads |
1789 | ------------------------------------------- | |
6c292092 TH |
1790 | |
1791 | cgroup v1 allowed threads to be in any cgroups which created an | |
1792 | interesting problem where threads belonging to a parent cgroup and its | |
1793 | children cgroups competed for resources. This was nasty as two | |
1794 | different types of entities competed and there was no obvious way to | |
1795 | settle it. Different controllers did different things. | |
1796 | ||
1797 | The cpu controller considered threads and cgroups as equivalents and | |
1798 | mapped nice levels to cgroup weights. This worked for some cases but | |
1799 | fell flat when children wanted to be allocated specific ratios of CPU | |
1800 | cycles and the number of internal threads fluctuated - the ratios | |
1801 | constantly changed as the number of competing entities fluctuated. | |
1802 | There also were other issues. The mapping from nice level to weight | |
1803 | wasn't obvious or universal, and there were various other knobs which | |
1804 | simply weren't available for threads. | |
1805 | ||
1806 | The io controller implicitly created a hidden leaf node for each | |
1807 | cgroup to host the threads. The hidden leaf had its own copies of all | |
633b11be | 1808 | the knobs with ``leaf_`` prefixed. While this allowed equivalent |
6c292092 TH |
1809 | control over internal threads, it was with serious drawbacks. It |
1810 | always added an extra layer of nesting which wouldn't be necessary | |
1811 | otherwise, made the interface messy and significantly complicated the | |
1812 | implementation. | |
1813 | ||
1814 | The memory controller didn't have a way to control what happened | |
1815 | between internal tasks and child cgroups and the behavior was not | |
1816 | clearly defined. There were attempts to add ad-hoc behaviors and | |
1817 | knobs to tailor the behavior to specific workloads which would have | |
1818 | led to problems extremely difficult to resolve in the long term. | |
1819 | ||
1820 | Multiple controllers struggled with internal tasks and came up with | |
1821 | different ways to deal with it; unfortunately, all the approaches were | |
1822 | severely flawed and, furthermore, the widely different behaviors | |
1823 | made cgroup as a whole highly inconsistent. | |
1824 | ||
1825 | This clearly is a problem which needs to be addressed from cgroup core | |
1826 | in a uniform way. | |
1827 | ||
1828 | ||
633b11be MCC |
1829 | Other Interface Issues |
1830 | ---------------------- | |
6c292092 TH |
1831 | |
1832 | cgroup v1 grew without oversight and developed a large number of | |
1833 | idiosyncrasies and inconsistencies. One issue on the cgroup core side | |
1834 | was how an empty cgroup was notified - a userland helper binary was | |
1835 | forked and executed for each event. The event delivery wasn't | |
1836 | recursive or delegatable. The limitations of the mechanism also led | |
1837 | to in-kernel event delivery filtering mechanism further complicating | |
1838 | the interface. | |
1839 | ||
1840 | Controller interfaces were problematic too. An extreme example is | |
1841 | controllers completely ignoring hierarchical organization and treating | |
1842 | all cgroups as if they were all located directly under the root | |
1843 | cgroup. Some controllers exposed a large amount of inconsistent | |
1844 | implementation details to userland. | |
1845 | ||
1846 | There also was no consistency across controllers. When a new cgroup | |
1847 | was created, some controllers defaulted to not imposing extra | |
1848 | restrictions while others disallowed any resource usage until | |
1849 | explicitly configured. Configuration knobs for the same type of | |
1850 | control used widely differing naming schemes and formats. Statistics | |
1851 | and information knobs were named arbitrarily and used different | |
1852 | formats and units even in the same controller. | |
1853 | ||
1854 | cgroup v2 establishes common conventions where appropriate and updates | |
1855 | controllers so that they expose minimal and consistent interfaces. | |
1856 | ||
1857 | ||
633b11be MCC |
1858 | Controller Issues and Remedies |
1859 | ------------------------------ | |
6c292092 | 1860 | |
633b11be MCC |
1861 | Memory |
1862 | ~~~~~~ | |
6c292092 TH |
1863 | |
1864 | The original lower boundary, the soft limit, is defined as a limit | |
1865 | that is per default unset. As a result, the set of cgroups that | |
1866 | global reclaim prefers is opt-in, rather than opt-out. The costs for | |
1867 | optimizing these mostly negative lookups are so high that the | |
1868 | implementation, despite its enormous size, does not even provide the | |
1869 | basic desirable behavior. First off, the soft limit has no | |
1870 | hierarchical meaning. All configured groups are organized in a global | |
1871 | rbtree and treated like equal peers, regardless where they are located | |
1872 | in the hierarchy. This makes subtree delegation impossible. Second, | |
1873 | the soft limit reclaim pass is so aggressive that it not just | |
1874 | introduces high allocation latencies into the system, but also impacts | |
1875 | system performance due to overreclaim, to the point where the feature | |
1876 | becomes self-defeating. | |
1877 | ||
1878 | The memory.low boundary on the other hand is a top-down allocated | |
1879 | reserve. A cgroup enjoys reclaim protection when it and all its | |
1880 | ancestors are below their low boundaries, which makes delegation of | |
1881 | subtrees possible. Secondly, new cgroups have no reserve per default | |
1882 | and in the common case most cgroups are eligible for the preferred | |
1883 | reclaim pass. This allows the new low boundary to be efficiently | |
1884 | implemented with just a minor addition to the generic reclaim code, | |
1885 | without the need for out-of-band data structures and reclaim passes. | |
1886 | Because the generic reclaim code considers all cgroups except for the | |
1887 | ones running low in the preferred first reclaim pass, overreclaim of | |
1888 | individual groups is eliminated as well, resulting in much better | |
1889 | overall workload performance. | |
1890 | ||
1891 | The original high boundary, the hard limit, is defined as a strict | |
1892 | limit that can not budge, even if the OOM killer has to be called. | |
1893 | But this generally goes against the goal of making the most out of the | |
1894 | available memory. The memory consumption of workloads varies during | |
1895 | runtime, and that requires users to overcommit. But doing that with a | |
1896 | strict upper limit requires either a fairly accurate prediction of the | |
1897 | working set size or adding slack to the limit. Since working set size | |
1898 | estimation is hard and error prone, and getting it wrong results in | |
1899 | OOM kills, most users tend to err on the side of a looser limit and | |
1900 | end up wasting precious resources. | |
1901 | ||
1902 | The memory.high boundary on the other hand can be set much more | |
1903 | conservatively. When hit, it throttles allocations by forcing them | |
1904 | into direct reclaim to work off the excess, but it never invokes the | |
1905 | OOM killer. As a result, a high boundary that is chosen too | |
1906 | aggressively will not terminate the processes, but instead it will | |
1907 | lead to gradual performance degradation. The user can monitor this | |
1908 | and make corrections until the minimal memory footprint that still | |
1909 | gives acceptable performance is found. | |
1910 | ||
1911 | In extreme cases, with many concurrent allocations and a complete | |
1912 | breakdown of reclaim progress within the group, the high boundary can | |
1913 | be exceeded. But even then it's mostly better to satisfy the | |
1914 | allocation from the slack available in other groups or the rest of the | |
1915 | system than killing the group. Otherwise, memory.max is there to | |
1916 | limit this type of spillover and ultimately contain buggy or even | |
1917 | malicious applications. | |
3e24b19d | 1918 | |
b6e6edcf JW |
1919 | Setting the original memory.limit_in_bytes below the current usage was |
1920 | subject to a race condition, where concurrent charges could cause the | |
1921 | limit setting to fail. memory.max on the other hand will first set the | |
1922 | limit to prevent new charges, and then reclaim and OOM kill until the | |
1923 | new limit is met - or the task writing to memory.max is killed. | |
1924 | ||
3e24b19d VD |
1925 | The combined memory+swap accounting and limiting is replaced by real |
1926 | control over swap space. | |
1927 | ||
1928 | The main argument for a combined memory+swap facility in the original | |
1929 | cgroup design was that global or parental pressure would always be | |
1930 | able to swap all anonymous memory of a child group, regardless of the | |
1931 | child's own (possibly untrusted) configuration. However, untrusted | |
1932 | groups can sabotage swapping by other means - such as referencing its | |
1933 | anonymous memory in a tight loop - and an admin can not assume full | |
1934 | swappability when overcommitting untrusted jobs. | |
1935 | ||
1936 | For trusted jobs, on the other hand, a combined counter is not an | |
1937 | intuitive userspace interface, and it flies in the face of the idea | |
1938 | that cgroup controllers should account and limit specific physical | |
1939 | resources. Swap space is a resource like all others in the system, | |
1940 | and that's why unified hierarchy allows distributing it separately. |