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 | |
b351f0c7 JB |
54 | 5-3-3. IO Latency |
55 | 5-3-3-1. How IO Latency Throttling Works | |
56 | 5-3-3-2. IO Latency Interface Files | |
633b11be MCC |
57 | 5-4. PID |
58 | 5-4-1. PID Interface Files | |
4ad5a321 RG |
59 | 5-5. Device |
60 | 5-6. RDMA | |
61 | 5-6-1. RDMA Interface Files | |
62 | 5-7. Misc | |
63 | 5-7-1. perf_event | |
c4e0842b MS |
64 | 5-N. Non-normative information |
65 | 5-N-1. CPU controller root cgroup process behaviour | |
66 | 5-N-2. IO controller root cgroup process behaviour | |
633b11be MCC |
67 | 6. Namespace |
68 | 6-1. Basics | |
69 | 6-2. The Root and Views | |
70 | 6-3. Migration and setns(2) | |
71 | 6-4. Interaction with Other Namespaces | |
72 | P. Information on Kernel Programming | |
73 | P-1. Filesystem Support for Writeback | |
74 | D. Deprecated v1 Core Features | |
75 | R. Issues with v1 and Rationales for v2 | |
76 | R-1. Multiple Hierarchies | |
77 | R-2. Thread Granularity | |
78 | R-3. Competition Between Inner Nodes and Threads | |
79 | R-4. Other Interface Issues | |
80 | R-5. Controller Issues and Remedies | |
81 | R-5-1. Memory | |
82 | ||
83 | ||
84 | Introduction | |
85 | ============ | |
86 | ||
87 | Terminology | |
88 | ----------- | |
6c292092 TH |
89 | |
90 | "cgroup" stands for "control group" and is never capitalized. The | |
91 | singular form is used to designate the whole feature and also as a | |
92 | qualifier as in "cgroup controllers". When explicitly referring to | |
93 | multiple individual control groups, the plural form "cgroups" is used. | |
94 | ||
95 | ||
633b11be MCC |
96 | What is cgroup? |
97 | --------------- | |
6c292092 TH |
98 | |
99 | cgroup is a mechanism to organize processes hierarchically and | |
100 | distribute system resources along the hierarchy in a controlled and | |
101 | configurable manner. | |
102 | ||
103 | cgroup is largely composed of two parts - the core and controllers. | |
104 | cgroup core is primarily responsible for hierarchically organizing | |
105 | processes. A cgroup controller is usually responsible for | |
106 | distributing a specific type of system resource along the hierarchy | |
107 | although there are utility controllers which serve purposes other than | |
108 | resource distribution. | |
109 | ||
110 | cgroups form a tree structure and every process in the system belongs | |
111 | to one and only one cgroup. All threads of a process belong to the | |
112 | same cgroup. On creation, all processes are put in the cgroup that | |
113 | the parent process belongs to at the time. A process can be migrated | |
114 | to another cgroup. Migration of a process doesn't affect already | |
115 | existing descendant processes. | |
116 | ||
117 | Following certain structural constraints, controllers may be enabled or | |
118 | disabled selectively on a cgroup. All controller behaviors are | |
119 | hierarchical - if a controller is enabled on a cgroup, it affects all | |
120 | processes which belong to the cgroups consisting the inclusive | |
121 | sub-hierarchy of the cgroup. When a controller is enabled on a nested | |
122 | cgroup, it always restricts the resource distribution further. The | |
123 | restrictions set closer to the root in the hierarchy can not be | |
124 | overridden from further away. | |
125 | ||
126 | ||
633b11be MCC |
127 | Basic Operations |
128 | ================ | |
6c292092 | 129 | |
633b11be MCC |
130 | Mounting |
131 | -------- | |
6c292092 TH |
132 | |
133 | Unlike v1, cgroup v2 has only single hierarchy. The cgroup v2 | |
633b11be | 134 | hierarchy can be mounted with the following mount command:: |
6c292092 TH |
135 | |
136 | # mount -t cgroup2 none $MOUNT_POINT | |
137 | ||
138 | cgroup2 filesystem has the magic number 0x63677270 ("cgrp"). All | |
139 | controllers which support v2 and are not bound to a v1 hierarchy are | |
140 | automatically bound to the v2 hierarchy and show up at the root. | |
141 | Controllers which are not in active use in the v2 hierarchy can be | |
142 | bound to other hierarchies. This allows mixing v2 hierarchy with the | |
143 | legacy v1 multiple hierarchies in a fully backward compatible way. | |
144 | ||
145 | A controller can be moved across hierarchies only after the controller | |
146 | is no longer referenced in its current hierarchy. Because per-cgroup | |
147 | controller states are destroyed asynchronously and controllers may | |
148 | have lingering references, a controller may not show up immediately on | |
149 | the v2 hierarchy after the final umount of the previous hierarchy. | |
150 | Similarly, a controller should be fully disabled to be moved out of | |
151 | the unified hierarchy and it may take some time for the disabled | |
152 | controller to become available for other hierarchies; furthermore, due | |
153 | to inter-controller dependencies, other controllers may need to be | |
154 | disabled too. | |
155 | ||
156 | While useful for development and manual configurations, moving | |
157 | controllers dynamically between the v2 and other hierarchies is | |
158 | strongly discouraged for production use. It is recommended to decide | |
159 | the hierarchies and controller associations before starting using the | |
160 | controllers after system boot. | |
161 | ||
1619b6d4 JW |
162 | During transition to v2, system management software might still |
163 | automount the v1 cgroup filesystem and so hijack all controllers | |
164 | during boot, before manual intervention is possible. To make testing | |
165 | and experimenting easier, the kernel parameter cgroup_no_v1= allows | |
166 | disabling controllers in v1 and make them always available in v2. | |
167 | ||
5136f636 TH |
168 | cgroup v2 currently supports the following mount options. |
169 | ||
170 | nsdelegate | |
171 | ||
172 | Consider cgroup namespaces as delegation boundaries. This | |
173 | option is system wide and can only be set on mount or modified | |
174 | through remount from the init namespace. The mount option is | |
175 | ignored on non-init namespace mounts. Please refer to the | |
176 | Delegation section for details. | |
177 | ||
6c292092 | 178 | |
8cfd8147 TH |
179 | Organizing Processes and Threads |
180 | -------------------------------- | |
181 | ||
182 | Processes | |
183 | ~~~~~~~~~ | |
6c292092 TH |
184 | |
185 | Initially, only the root cgroup exists to which all processes belong. | |
633b11be | 186 | A child cgroup can be created by creating a sub-directory:: |
6c292092 TH |
187 | |
188 | # mkdir $CGROUP_NAME | |
189 | ||
190 | A given cgroup may have multiple child cgroups forming a tree | |
191 | structure. Each cgroup has a read-writable interface file | |
192 | "cgroup.procs". When read, it lists the PIDs of all processes which | |
193 | belong to the cgroup one-per-line. The PIDs are not ordered and the | |
194 | same PID may show up more than once if the process got moved to | |
195 | another cgroup and then back or the PID got recycled while reading. | |
196 | ||
197 | A process can be migrated into a cgroup by writing its PID to the | |
198 | target cgroup's "cgroup.procs" file. Only one process can be migrated | |
199 | on a single write(2) call. If a process is composed of multiple | |
200 | threads, writing the PID of any thread migrates all threads of the | |
201 | process. | |
202 | ||
203 | When a process forks a child process, the new process is born into the | |
204 | cgroup that the forking process belongs to at the time of the | |
205 | operation. After exit, a process stays associated with the cgroup | |
206 | that it belonged to at the time of exit until it's reaped; however, a | |
207 | zombie process does not appear in "cgroup.procs" and thus can't be | |
208 | moved to another cgroup. | |
209 | ||
210 | A cgroup which doesn't have any children or live processes can be | |
211 | destroyed by removing the directory. Note that a cgroup which doesn't | |
212 | have any children and is associated only with zombie processes is | |
633b11be | 213 | considered empty and can be removed:: |
6c292092 TH |
214 | |
215 | # rmdir $CGROUP_NAME | |
216 | ||
217 | "/proc/$PID/cgroup" lists a process's cgroup membership. If legacy | |
218 | cgroup is in use in the system, this file may contain multiple lines, | |
219 | one for each hierarchy. The entry for cgroup v2 is always in the | |
633b11be | 220 | format "0::$PATH":: |
6c292092 TH |
221 | |
222 | # cat /proc/842/cgroup | |
223 | ... | |
224 | 0::/test-cgroup/test-cgroup-nested | |
225 | ||
226 | If the process becomes a zombie and the cgroup it was associated with | |
633b11be | 227 | is removed subsequently, " (deleted)" is appended to the path:: |
6c292092 TH |
228 | |
229 | # cat /proc/842/cgroup | |
230 | ... | |
231 | 0::/test-cgroup/test-cgroup-nested (deleted) | |
232 | ||
233 | ||
8cfd8147 TH |
234 | Threads |
235 | ~~~~~~~ | |
236 | ||
237 | cgroup v2 supports thread granularity for a subset of controllers to | |
238 | support use cases requiring hierarchical resource distribution across | |
239 | the threads of a group of processes. By default, all threads of a | |
240 | process belong to the same cgroup, which also serves as the resource | |
241 | domain to host resource consumptions which are not specific to a | |
242 | process or thread. The thread mode allows threads to be spread across | |
243 | a subtree while still maintaining the common resource domain for them. | |
244 | ||
245 | Controllers which support thread mode are called threaded controllers. | |
246 | The ones which don't are called domain controllers. | |
247 | ||
248 | Marking a cgroup threaded makes it join the resource domain of its | |
249 | parent as a threaded cgroup. The parent may be another threaded | |
250 | cgroup whose resource domain is further up in the hierarchy. The root | |
251 | of a threaded subtree, that is, the nearest ancestor which is not | |
252 | threaded, is called threaded domain or thread root interchangeably and | |
253 | serves as the resource domain for the entire subtree. | |
254 | ||
255 | Inside a threaded subtree, threads of a process can be put in | |
256 | different cgroups and are not subject to the no internal process | |
257 | constraint - threaded controllers can be enabled on non-leaf cgroups | |
258 | whether they have threads in them or not. | |
259 | ||
260 | As the threaded domain cgroup hosts all the domain resource | |
261 | consumptions of the subtree, it is considered to have internal | |
262 | resource consumptions whether there are processes in it or not and | |
263 | can't have populated child cgroups which aren't threaded. Because the | |
264 | root cgroup is not subject to no internal process constraint, it can | |
265 | serve both as a threaded domain and a parent to domain cgroups. | |
266 | ||
267 | The current operation mode or type of the cgroup is shown in the | |
268 | "cgroup.type" file which indicates whether the cgroup is a normal | |
269 | domain, a domain which is serving as the domain of a threaded subtree, | |
270 | or a threaded cgroup. | |
271 | ||
272 | On creation, a cgroup is always a domain cgroup and can be made | |
273 | threaded by writing "threaded" to the "cgroup.type" file. The | |
274 | operation is single direction:: | |
275 | ||
276 | # echo threaded > cgroup.type | |
277 | ||
278 | Once threaded, the cgroup can't be made a domain again. To enable the | |
279 | thread mode, the following conditions must be met. | |
280 | ||
281 | - As the cgroup will join the parent's resource domain. The parent | |
282 | must either be a valid (threaded) domain or a threaded cgroup. | |
283 | ||
918a8c2c TH |
284 | - When the parent is an unthreaded domain, it must not have any domain |
285 | controllers enabled or populated domain children. The root is | |
286 | exempt from this requirement. | |
8cfd8147 TH |
287 | |
288 | Topology-wise, a cgroup can be in an invalid state. Please consider | |
2877cbe6 | 289 | the following topology:: |
8cfd8147 TH |
290 | |
291 | A (threaded domain) - B (threaded) - C (domain, just created) | |
292 | ||
293 | C is created as a domain but isn't connected to a parent which can | |
294 | host child domains. C can't be used until it is turned into a | |
295 | threaded cgroup. "cgroup.type" file will report "domain (invalid)" in | |
296 | these cases. Operations which fail due to invalid topology use | |
297 | EOPNOTSUPP as the errno. | |
298 | ||
299 | A domain cgroup is turned into a threaded domain when one of its child | |
300 | cgroup becomes threaded or threaded controllers are enabled in the | |
301 | "cgroup.subtree_control" file while there are processes in the cgroup. | |
302 | A threaded domain reverts to a normal domain when the conditions | |
303 | clear. | |
304 | ||
305 | When read, "cgroup.threads" contains the list of the thread IDs of all | |
306 | threads in the cgroup. Except that the operations are per-thread | |
307 | instead of per-process, "cgroup.threads" has the same format and | |
308 | behaves the same way as "cgroup.procs". While "cgroup.threads" can be | |
309 | written to in any cgroup, as it can only move threads inside the same | |
310 | threaded domain, its operations are confined inside each threaded | |
311 | subtree. | |
312 | ||
313 | The threaded domain cgroup serves as the resource domain for the whole | |
314 | subtree, and, while the threads can be scattered across the subtree, | |
315 | all the processes are considered to be in the threaded domain cgroup. | |
316 | "cgroup.procs" in a threaded domain cgroup contains the PIDs of all | |
317 | processes in the subtree and is not readable in the subtree proper. | |
318 | However, "cgroup.procs" can be written to from anywhere in the subtree | |
319 | to migrate all threads of the matching process to the cgroup. | |
320 | ||
321 | Only threaded controllers can be enabled in a threaded subtree. When | |
322 | a threaded controller is enabled inside a threaded subtree, it only | |
323 | accounts for and controls resource consumptions associated with the | |
324 | threads in the cgroup and its descendants. All consumptions which | |
325 | aren't tied to a specific thread belong to the threaded domain cgroup. | |
326 | ||
327 | Because a threaded subtree is exempt from no internal process | |
328 | constraint, a threaded controller must be able to handle competition | |
329 | between threads in a non-leaf cgroup and its child cgroups. Each | |
330 | threaded controller defines how such competitions are handled. | |
331 | ||
332 | ||
633b11be MCC |
333 | [Un]populated Notification |
334 | -------------------------- | |
6c292092 TH |
335 | |
336 | Each non-root cgroup has a "cgroup.events" file which contains | |
337 | "populated" field indicating whether the cgroup's sub-hierarchy has | |
338 | live processes in it. Its value is 0 if there is no live process in | |
339 | the cgroup and its descendants; otherwise, 1. poll and [id]notify | |
340 | events are triggered when the value changes. This can be used, for | |
341 | example, to start a clean-up operation after all processes of a given | |
342 | sub-hierarchy have exited. The populated state updates and | |
343 | notifications are recursive. Consider the following sub-hierarchy | |
344 | where the numbers in the parentheses represent the numbers of processes | |
633b11be | 345 | in each cgroup:: |
6c292092 TH |
346 | |
347 | A(4) - B(0) - C(1) | |
348 | \ D(0) | |
349 | ||
350 | A, B and C's "populated" fields would be 1 while D's 0. After the one | |
351 | process in C exits, B and C's "populated" fields would flip to "0" and | |
352 | file modified events will be generated on the "cgroup.events" files of | |
353 | both cgroups. | |
354 | ||
355 | ||
633b11be MCC |
356 | Controlling Controllers |
357 | ----------------------- | |
6c292092 | 358 | |
633b11be MCC |
359 | Enabling and Disabling |
360 | ~~~~~~~~~~~~~~~~~~~~~~ | |
6c292092 TH |
361 | |
362 | Each cgroup has a "cgroup.controllers" file which lists all | |
633b11be | 363 | controllers available for the cgroup to enable:: |
6c292092 TH |
364 | |
365 | # cat cgroup.controllers | |
366 | cpu io memory | |
367 | ||
368 | No controller is enabled by default. Controllers can be enabled and | |
633b11be | 369 | disabled by writing to the "cgroup.subtree_control" file:: |
6c292092 TH |
370 | |
371 | # echo "+cpu +memory -io" > cgroup.subtree_control | |
372 | ||
373 | Only controllers which are listed in "cgroup.controllers" can be | |
374 | enabled. When multiple operations are specified as above, either they | |
375 | all succeed or fail. If multiple operations on the same controller | |
376 | are specified, the last one is effective. | |
377 | ||
378 | Enabling a controller in a cgroup indicates that the distribution of | |
379 | the target resource across its immediate children will be controlled. | |
380 | Consider the following sub-hierarchy. The enabled controllers are | |
633b11be | 381 | listed in parentheses:: |
6c292092 TH |
382 | |
383 | A(cpu,memory) - B(memory) - C() | |
384 | \ D() | |
385 | ||
386 | As A has "cpu" and "memory" enabled, A will control the distribution | |
387 | of CPU cycles and memory to its children, in this case, B. As B has | |
388 | "memory" enabled but not "CPU", C and D will compete freely on CPU | |
389 | cycles but their division of memory available to B will be controlled. | |
390 | ||
391 | As a controller regulates the distribution of the target resource to | |
392 | the cgroup's children, enabling it creates the controller's interface | |
393 | files in the child cgroups. In the above example, enabling "cpu" on B | |
394 | would create the "cpu." prefixed controller interface files in C and | |
395 | D. Likewise, disabling "memory" from B would remove the "memory." | |
396 | prefixed controller interface files from C and D. This means that the | |
397 | controller interface files - anything which doesn't start with | |
398 | "cgroup." are owned by the parent rather than the cgroup itself. | |
399 | ||
400 | ||
633b11be MCC |
401 | Top-down Constraint |
402 | ~~~~~~~~~~~~~~~~~~~ | |
6c292092 TH |
403 | |
404 | Resources are distributed top-down and a cgroup can further distribute | |
405 | a resource only if the resource has been distributed to it from the | |
406 | parent. This means that all non-root "cgroup.subtree_control" files | |
407 | can only contain controllers which are enabled in the parent's | |
408 | "cgroup.subtree_control" file. A controller can be enabled only if | |
409 | the parent has the controller enabled and a controller can't be | |
410 | disabled if one or more children have it enabled. | |
411 | ||
412 | ||
633b11be MCC |
413 | No Internal Process Constraint |
414 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | |
6c292092 | 415 | |
8cfd8147 TH |
416 | Non-root cgroups can distribute domain resources to their children |
417 | only when they don't have any processes of their own. In other words, | |
418 | only domain cgroups which don't contain any processes can have domain | |
419 | controllers enabled in their "cgroup.subtree_control" files. | |
6c292092 | 420 | |
8cfd8147 TH |
421 | This guarantees that, when a domain controller is looking at the part |
422 | of the hierarchy which has it enabled, processes are always only on | |
423 | the leaves. This rules out situations where child cgroups compete | |
424 | against internal processes of the parent. | |
6c292092 TH |
425 | |
426 | The root cgroup is exempt from this restriction. Root contains | |
427 | processes and anonymous resource consumption which can't be associated | |
428 | with any other cgroups and requires special treatment from most | |
429 | controllers. How resource consumption in the root cgroup is governed | |
c4e0842b MS |
430 | is up to each controller (for more information on this topic please |
431 | refer to the Non-normative information section in the Controllers | |
432 | chapter). | |
6c292092 TH |
433 | |
434 | Note that the restriction doesn't get in the way if there is no | |
435 | enabled controller in the cgroup's "cgroup.subtree_control". This is | |
436 | important as otherwise it wouldn't be possible to create children of a | |
437 | populated cgroup. To control resource distribution of a cgroup, the | |
438 | cgroup must create children and transfer all its processes to the | |
439 | children before enabling controllers in its "cgroup.subtree_control" | |
440 | file. | |
441 | ||
442 | ||
633b11be MCC |
443 | Delegation |
444 | ---------- | |
6c292092 | 445 | |
633b11be MCC |
446 | Model of Delegation |
447 | ~~~~~~~~~~~~~~~~~~~ | |
6c292092 | 448 | |
5136f636 | 449 | A cgroup can be delegated in two ways. First, to a less privileged |
8cfd8147 TH |
450 | user by granting write access of the directory and its "cgroup.procs", |
451 | "cgroup.threads" and "cgroup.subtree_control" files to the user. | |
452 | Second, if the "nsdelegate" mount option is set, automatically to a | |
453 | cgroup namespace on namespace creation. | |
5136f636 TH |
454 | |
455 | Because the resource control interface files in a given directory | |
456 | control the distribution of the parent's resources, the delegatee | |
457 | shouldn't be allowed to write to them. For the first method, this is | |
458 | achieved by not granting access to these files. For the second, the | |
459 | kernel rejects writes to all files other than "cgroup.procs" and | |
460 | "cgroup.subtree_control" on a namespace root from inside the | |
461 | namespace. | |
462 | ||
463 | The end results are equivalent for both delegation types. Once | |
464 | delegated, the user can build sub-hierarchy under the directory, | |
465 | organize processes inside it as it sees fit and further distribute the | |
466 | resources it received from the parent. The limits and other settings | |
467 | of all resource controllers are hierarchical and regardless of what | |
468 | happens in the delegated sub-hierarchy, nothing can escape the | |
469 | resource restrictions imposed by the parent. | |
6c292092 TH |
470 | |
471 | Currently, cgroup doesn't impose any restrictions on the number of | |
472 | cgroups in or nesting depth of a delegated sub-hierarchy; however, | |
473 | this may be limited explicitly in the future. | |
474 | ||
475 | ||
633b11be MCC |
476 | Delegation Containment |
477 | ~~~~~~~~~~~~~~~~~~~~~~ | |
6c292092 TH |
478 | |
479 | A delegated sub-hierarchy is contained in the sense that processes | |
5136f636 TH |
480 | can't be moved into or out of the sub-hierarchy by the delegatee. |
481 | ||
482 | For delegations to a less privileged user, this is achieved by | |
483 | requiring the following conditions for a process with a non-root euid | |
484 | to migrate a target process into a cgroup by writing its PID to the | |
485 | "cgroup.procs" file. | |
6c292092 | 486 | |
6c292092 TH |
487 | - The writer must have write access to the "cgroup.procs" file. |
488 | ||
489 | - The writer must have write access to the "cgroup.procs" file of the | |
490 | common ancestor of the source and destination cgroups. | |
491 | ||
576dd464 | 492 | The above two constraints ensure that while a delegatee may migrate |
6c292092 TH |
493 | processes around freely in the delegated sub-hierarchy it can't pull |
494 | in from or push out to outside the sub-hierarchy. | |
495 | ||
496 | For an example, let's assume cgroups C0 and C1 have been delegated to | |
497 | user U0 who created C00, C01 under C0 and C10 under C1 as follows and | |
633b11be | 498 | all processes under C0 and C1 belong to U0:: |
6c292092 TH |
499 | |
500 | ~~~~~~~~~~~~~ - C0 - C00 | |
501 | ~ cgroup ~ \ C01 | |
502 | ~ hierarchy ~ | |
503 | ~~~~~~~~~~~~~ - C1 - C10 | |
504 | ||
505 | Let's also say U0 wants to write the PID of a process which is | |
506 | currently in C10 into "C00/cgroup.procs". U0 has write access to the | |
576dd464 TH |
507 | file; however, the common ancestor of the source cgroup C10 and the |
508 | destination cgroup C00 is above the points of delegation and U0 would | |
509 | not have write access to its "cgroup.procs" files and thus the write | |
510 | will be denied with -EACCES. | |
6c292092 | 511 | |
5136f636 TH |
512 | For delegations to namespaces, containment is achieved by requiring |
513 | that both the source and destination cgroups are reachable from the | |
514 | namespace of the process which is attempting the migration. If either | |
515 | is not reachable, the migration is rejected with -ENOENT. | |
516 | ||
6c292092 | 517 | |
633b11be MCC |
518 | Guidelines |
519 | ---------- | |
6c292092 | 520 | |
633b11be MCC |
521 | Organize Once and Control |
522 | ~~~~~~~~~~~~~~~~~~~~~~~~~ | |
6c292092 TH |
523 | |
524 | Migrating a process across cgroups is a relatively expensive operation | |
525 | and stateful resources such as memory are not moved together with the | |
526 | process. This is an explicit design decision as there often exist | |
527 | inherent trade-offs between migration and various hot paths in terms | |
528 | of synchronization cost. | |
529 | ||
530 | As such, migrating processes across cgroups frequently as a means to | |
531 | apply different resource restrictions is discouraged. A workload | |
532 | should be assigned to a cgroup according to the system's logical and | |
533 | resource structure once on start-up. Dynamic adjustments to resource | |
534 | distribution can be made by changing controller configuration through | |
535 | the interface files. | |
536 | ||
537 | ||
633b11be MCC |
538 | Avoid Name Collisions |
539 | ~~~~~~~~~~~~~~~~~~~~~ | |
6c292092 TH |
540 | |
541 | Interface files for a cgroup and its children cgroups occupy the same | |
542 | directory and it is possible to create children cgroups which collide | |
543 | with interface files. | |
544 | ||
545 | All cgroup core interface files are prefixed with "cgroup." and each | |
546 | controller's interface files are prefixed with the controller name and | |
547 | a dot. A controller's name is composed of lower case alphabets and | |
548 | '_'s but never begins with an '_' so it can be used as the prefix | |
549 | character for collision avoidance. Also, interface file names won't | |
550 | start or end with terms which are often used in categorizing workloads | |
551 | such as job, service, slice, unit or workload. | |
552 | ||
553 | cgroup doesn't do anything to prevent name collisions and it's the | |
554 | user's responsibility to avoid them. | |
555 | ||
556 | ||
633b11be MCC |
557 | Resource Distribution Models |
558 | ============================ | |
6c292092 TH |
559 | |
560 | cgroup controllers implement several resource distribution schemes | |
561 | depending on the resource type and expected use cases. This section | |
562 | describes major schemes in use along with their expected behaviors. | |
563 | ||
564 | ||
633b11be MCC |
565 | Weights |
566 | ------- | |
6c292092 TH |
567 | |
568 | A parent's resource is distributed by adding up the weights of all | |
569 | active children and giving each the fraction matching the ratio of its | |
570 | weight against the sum. As only children which can make use of the | |
571 | resource at the moment participate in the distribution, this is | |
572 | work-conserving. Due to the dynamic nature, this model is usually | |
573 | used for stateless resources. | |
574 | ||
575 | All weights are in the range [1, 10000] with the default at 100. This | |
576 | allows symmetric multiplicative biases in both directions at fine | |
577 | enough granularity while staying in the intuitive range. | |
578 | ||
579 | As long as the weight is in range, all configuration combinations are | |
580 | valid and there is no reason to reject configuration changes or | |
581 | process migrations. | |
582 | ||
583 | "cpu.weight" proportionally distributes CPU cycles to active children | |
584 | and is an example of this type. | |
585 | ||
586 | ||
633b11be MCC |
587 | Limits |
588 | ------ | |
6c292092 TH |
589 | |
590 | A child can only consume upto the configured amount of the resource. | |
591 | Limits can be over-committed - the sum of the limits of children can | |
592 | exceed the amount of resource available to the parent. | |
593 | ||
594 | Limits are in the range [0, max] and defaults to "max", which is noop. | |
595 | ||
596 | As limits can be over-committed, all configuration combinations are | |
597 | valid and there is no reason to reject configuration changes or | |
598 | process migrations. | |
599 | ||
600 | "io.max" limits the maximum BPS and/or IOPS that a cgroup can consume | |
601 | on an IO device and is an example of this type. | |
602 | ||
603 | ||
633b11be MCC |
604 | Protections |
605 | ----------- | |
6c292092 TH |
606 | |
607 | A cgroup is protected to be allocated upto the configured amount of | |
608 | the resource if the usages of all its ancestors are under their | |
609 | protected levels. Protections can be hard guarantees or best effort | |
610 | soft boundaries. Protections can also be over-committed in which case | |
611 | only upto the amount available to the parent is protected among | |
612 | children. | |
613 | ||
614 | Protections are in the range [0, max] and defaults to 0, which is | |
615 | noop. | |
616 | ||
617 | As protections can be over-committed, all configuration combinations | |
618 | are valid and there is no reason to reject configuration changes or | |
619 | process migrations. | |
620 | ||
621 | "memory.low" implements best-effort memory protection and is an | |
622 | example of this type. | |
623 | ||
624 | ||
633b11be MCC |
625 | Allocations |
626 | ----------- | |
6c292092 TH |
627 | |
628 | A cgroup is exclusively allocated a certain amount of a finite | |
629 | resource. Allocations can't be over-committed - the sum of the | |
630 | allocations of children can not exceed the amount of resource | |
631 | available to the parent. | |
632 | ||
633 | Allocations are in the range [0, max] and defaults to 0, which is no | |
634 | resource. | |
635 | ||
636 | As allocations can't be over-committed, some configuration | |
637 | combinations are invalid and should be rejected. Also, if the | |
638 | resource is mandatory for execution of processes, process migrations | |
639 | may be rejected. | |
640 | ||
641 | "cpu.rt.max" hard-allocates realtime slices and is an example of this | |
642 | type. | |
643 | ||
644 | ||
633b11be MCC |
645 | Interface Files |
646 | =============== | |
6c292092 | 647 | |
633b11be MCC |
648 | Format |
649 | ------ | |
6c292092 TH |
650 | |
651 | All interface files should be in one of the following formats whenever | |
633b11be | 652 | possible:: |
6c292092 TH |
653 | |
654 | New-line separated values | |
655 | (when only one value can be written at once) | |
656 | ||
657 | VAL0\n | |
658 | VAL1\n | |
659 | ... | |
660 | ||
661 | Space separated values | |
662 | (when read-only or multiple values can be written at once) | |
663 | ||
664 | VAL0 VAL1 ...\n | |
665 | ||
666 | Flat keyed | |
667 | ||
668 | KEY0 VAL0\n | |
669 | KEY1 VAL1\n | |
670 | ... | |
671 | ||
672 | Nested keyed | |
673 | ||
674 | KEY0 SUB_KEY0=VAL00 SUB_KEY1=VAL01... | |
675 | KEY1 SUB_KEY0=VAL10 SUB_KEY1=VAL11... | |
676 | ... | |
677 | ||
678 | For a writable file, the format for writing should generally match | |
679 | reading; however, controllers may allow omitting later fields or | |
680 | implement restricted shortcuts for most common use cases. | |
681 | ||
682 | For both flat and nested keyed files, only the values for a single key | |
683 | can be written at a time. For nested keyed files, the sub key pairs | |
684 | may be specified in any order and not all pairs have to be specified. | |
685 | ||
686 | ||
633b11be MCC |
687 | Conventions |
688 | ----------- | |
6c292092 TH |
689 | |
690 | - Settings for a single feature should be contained in a single file. | |
691 | ||
692 | - The root cgroup should be exempt from resource control and thus | |
693 | shouldn't have resource control interface files. Also, | |
694 | informational files on the root cgroup which end up showing global | |
695 | information available elsewhere shouldn't exist. | |
696 | ||
697 | - If a controller implements weight based resource distribution, its | |
698 | interface file should be named "weight" and have the range [1, | |
699 | 10000] with 100 as the default. The values are chosen to allow | |
700 | enough and symmetric bias in both directions while keeping it | |
701 | intuitive (the default is 100%). | |
702 | ||
703 | - If a controller implements an absolute resource guarantee and/or | |
704 | limit, the interface files should be named "min" and "max" | |
705 | respectively. If a controller implements best effort resource | |
706 | guarantee and/or limit, the interface files should be named "low" | |
707 | and "high" respectively. | |
708 | ||
709 | In the above four control files, the special token "max" should be | |
710 | used to represent upward infinity for both reading and writing. | |
711 | ||
712 | - If a setting has a configurable default value and keyed specific | |
713 | overrides, the default entry should be keyed with "default" and | |
714 | appear as the first entry in the file. | |
715 | ||
716 | The default value can be updated by writing either "default $VAL" or | |
717 | "$VAL". | |
718 | ||
719 | When writing to update a specific override, "default" can be used as | |
720 | the value to indicate removal of the override. Override entries | |
721 | with "default" as the value must not appear when read. | |
722 | ||
723 | For example, a setting which is keyed by major:minor device numbers | |
633b11be | 724 | with integer values may look like the following:: |
6c292092 TH |
725 | |
726 | # cat cgroup-example-interface-file | |
727 | default 150 | |
728 | 8:0 300 | |
729 | ||
633b11be | 730 | The default value can be updated by:: |
6c292092 TH |
731 | |
732 | # echo 125 > cgroup-example-interface-file | |
733 | ||
633b11be | 734 | or:: |
6c292092 TH |
735 | |
736 | # echo "default 125" > cgroup-example-interface-file | |
737 | ||
633b11be | 738 | An override can be set by:: |
6c292092 TH |
739 | |
740 | # echo "8:16 170" > cgroup-example-interface-file | |
741 | ||
633b11be | 742 | and cleared by:: |
6c292092 TH |
743 | |
744 | # echo "8:0 default" > cgroup-example-interface-file | |
745 | # cat cgroup-example-interface-file | |
746 | default 125 | |
747 | 8:16 170 | |
748 | ||
749 | - For events which are not very high frequency, an interface file | |
750 | "events" should be created which lists event key value pairs. | |
751 | Whenever a notifiable event happens, file modified event should be | |
752 | generated on the file. | |
753 | ||
754 | ||
633b11be MCC |
755 | Core Interface Files |
756 | -------------------- | |
6c292092 TH |
757 | |
758 | All cgroup core files are prefixed with "cgroup." | |
759 | ||
8cfd8147 TH |
760 | cgroup.type |
761 | ||
762 | A read-write single value file which exists on non-root | |
763 | cgroups. | |
764 | ||
765 | When read, it indicates the current type of the cgroup, which | |
766 | can be one of the following values. | |
767 | ||
768 | - "domain" : A normal valid domain cgroup. | |
769 | ||
770 | - "domain threaded" : A threaded domain cgroup which is | |
771 | serving as the root of a threaded subtree. | |
772 | ||
773 | - "domain invalid" : A cgroup which is in an invalid state. | |
774 | It can't be populated or have controllers enabled. It may | |
775 | be allowed to become a threaded cgroup. | |
776 | ||
777 | - "threaded" : A threaded cgroup which is a member of a | |
778 | threaded subtree. | |
779 | ||
780 | A cgroup can be turned into a threaded cgroup by writing | |
781 | "threaded" to this file. | |
782 | ||
6c292092 | 783 | cgroup.procs |
6c292092 TH |
784 | A read-write new-line separated values file which exists on |
785 | all cgroups. | |
786 | ||
787 | When read, it lists the PIDs of all processes which belong to | |
788 | the cgroup one-per-line. The PIDs are not ordered and the | |
789 | same PID may show up more than once if the process got moved | |
790 | to another cgroup and then back or the PID got recycled while | |
791 | reading. | |
792 | ||
793 | A PID can be written to migrate the process associated with | |
794 | the PID to the cgroup. The writer should match all of the | |
795 | following conditions. | |
796 | ||
6c292092 | 797 | - It must have write access to the "cgroup.procs" file. |
8cfd8147 TH |
798 | |
799 | - It must have write access to the "cgroup.procs" file of the | |
800 | common ancestor of the source and destination cgroups. | |
801 | ||
802 | When delegating a sub-hierarchy, write access to this file | |
803 | should be granted along with the containing directory. | |
804 | ||
805 | In a threaded cgroup, reading this file fails with EOPNOTSUPP | |
806 | as all the processes belong to the thread root. Writing is | |
807 | supported and moves every thread of the process to the cgroup. | |
808 | ||
809 | cgroup.threads | |
810 | A read-write new-line separated values file which exists on | |
811 | all cgroups. | |
812 | ||
813 | When read, it lists the TIDs of all threads which belong to | |
814 | the cgroup one-per-line. The TIDs are not ordered and the | |
815 | same TID may show up more than once if the thread got moved to | |
816 | another cgroup and then back or the TID got recycled while | |
817 | reading. | |
818 | ||
819 | A TID can be written to migrate the thread associated with the | |
820 | TID to the cgroup. The writer should match all of the | |
821 | following conditions. | |
822 | ||
823 | - It must have write access to the "cgroup.threads" file. | |
824 | ||
825 | - The cgroup that the thread is currently in must be in the | |
826 | same resource domain as the destination cgroup. | |
6c292092 TH |
827 | |
828 | - It must have write access to the "cgroup.procs" file of the | |
829 | common ancestor of the source and destination cgroups. | |
830 | ||
831 | When delegating a sub-hierarchy, write access to this file | |
832 | should be granted along with the containing directory. | |
833 | ||
834 | cgroup.controllers | |
6c292092 TH |
835 | A read-only space separated values file which exists on all |
836 | cgroups. | |
837 | ||
838 | It shows space separated list of all controllers available to | |
839 | the cgroup. The controllers are not ordered. | |
840 | ||
841 | cgroup.subtree_control | |
6c292092 TH |
842 | A read-write space separated values file which exists on all |
843 | cgroups. Starts out empty. | |
844 | ||
845 | When read, it shows space separated list of the controllers | |
846 | which are enabled to control resource distribution from the | |
847 | cgroup to its children. | |
848 | ||
849 | Space separated list of controllers prefixed with '+' or '-' | |
850 | can be written to enable or disable controllers. A controller | |
851 | name prefixed with '+' enables the controller and '-' | |
852 | disables. If a controller appears more than once on the list, | |
853 | the last one is effective. When multiple enable and disable | |
854 | operations are specified, either all succeed or all fail. | |
855 | ||
856 | cgroup.events | |
6c292092 TH |
857 | A read-only flat-keyed file which exists on non-root cgroups. |
858 | The following entries are defined. Unless specified | |
859 | otherwise, a value change in this file generates a file | |
860 | modified event. | |
861 | ||
862 | populated | |
6c292092 TH |
863 | 1 if the cgroup or its descendants contains any live |
864 | processes; otherwise, 0. | |
865 | ||
1a926e0b RG |
866 | cgroup.max.descendants |
867 | A read-write single value files. The default is "max". | |
868 | ||
869 | Maximum allowed number of descent cgroups. | |
870 | If the actual number of descendants is equal or larger, | |
871 | an attempt to create a new cgroup in the hierarchy will fail. | |
872 | ||
873 | cgroup.max.depth | |
874 | A read-write single value files. The default is "max". | |
875 | ||
876 | Maximum allowed descent depth below the current cgroup. | |
877 | If the actual descent depth is equal or larger, | |
878 | an attempt to create a new child cgroup will fail. | |
879 | ||
ec39225c RG |
880 | cgroup.stat |
881 | A read-only flat-keyed file with the following entries: | |
882 | ||
883 | nr_descendants | |
884 | Total number of visible descendant cgroups. | |
885 | ||
886 | nr_dying_descendants | |
887 | Total number of dying descendant cgroups. A cgroup becomes | |
888 | dying after being deleted by a user. The cgroup will remain | |
889 | in dying state for some time undefined time (which can depend | |
890 | on system load) before being completely destroyed. | |
891 | ||
892 | A process can't enter a dying cgroup under any circumstances, | |
893 | a dying cgroup can't revive. | |
894 | ||
895 | A dying cgroup can consume system resources not exceeding | |
896 | limits, which were active at the moment of cgroup deletion. | |
897 | ||
6c292092 | 898 | |
633b11be MCC |
899 | Controllers |
900 | =========== | |
6c292092 | 901 | |
633b11be MCC |
902 | CPU |
903 | --- | |
6c292092 | 904 | |
6c292092 TH |
905 | The "cpu" controllers regulates distribution of CPU cycles. This |
906 | controller implements weight and absolute bandwidth limit models for | |
907 | normal scheduling policy and absolute bandwidth allocation model for | |
908 | realtime scheduling policy. | |
909 | ||
c2f31b79 TH |
910 | WARNING: cgroup2 doesn't yet support control of realtime processes and |
911 | the cpu controller can only be enabled when all RT processes are in | |
912 | the root cgroup. Be aware that system management software may already | |
913 | have placed RT processes into nonroot cgroups during the system boot | |
914 | process, and these processes may need to be moved to the root cgroup | |
915 | before the cpu controller can be enabled. | |
916 | ||
6c292092 | 917 | |
633b11be MCC |
918 | CPU Interface Files |
919 | ~~~~~~~~~~~~~~~~~~~ | |
6c292092 TH |
920 | |
921 | All time durations are in microseconds. | |
922 | ||
923 | cpu.stat | |
6c292092 | 924 | A read-only flat-keyed file which exists on non-root cgroups. |
d41bf8c9 | 925 | This file exists whether the controller is enabled or not. |
6c292092 | 926 | |
d41bf8c9 | 927 | It always reports the following three stats: |
6c292092 | 928 | |
633b11be MCC |
929 | - usage_usec |
930 | - user_usec | |
931 | - system_usec | |
d41bf8c9 TH |
932 | |
933 | and the following three when the controller is enabled: | |
934 | ||
633b11be MCC |
935 | - nr_periods |
936 | - nr_throttled | |
937 | - throttled_usec | |
6c292092 TH |
938 | |
939 | cpu.weight | |
6c292092 TH |
940 | A read-write single value file which exists on non-root |
941 | cgroups. The default is "100". | |
942 | ||
943 | The weight in the range [1, 10000]. | |
944 | ||
0d593634 TH |
945 | cpu.weight.nice |
946 | A read-write single value file which exists on non-root | |
947 | cgroups. The default is "0". | |
948 | ||
949 | The nice value is in the range [-20, 19]. | |
950 | ||
951 | This interface file is an alternative interface for | |
952 | "cpu.weight" and allows reading and setting weight using the | |
953 | same values used by nice(2). Because the range is smaller and | |
954 | granularity is coarser for the nice values, the read value is | |
955 | the closest approximation of the current weight. | |
956 | ||
6c292092 | 957 | cpu.max |
6c292092 TH |
958 | A read-write two value file which exists on non-root cgroups. |
959 | The default is "max 100000". | |
960 | ||
633b11be | 961 | The maximum bandwidth limit. It's in the following format:: |
6c292092 TH |
962 | |
963 | $MAX $PERIOD | |
964 | ||
965 | which indicates that the group may consume upto $MAX in each | |
966 | $PERIOD duration. "max" for $MAX indicates no limit. If only | |
967 | one number is written, $MAX is updated. | |
968 | ||
6c292092 | 969 | |
633b11be MCC |
970 | Memory |
971 | ------ | |
6c292092 TH |
972 | |
973 | The "memory" controller regulates distribution of memory. Memory is | |
974 | stateful and implements both limit and protection models. Due to the | |
975 | intertwining between memory usage and reclaim pressure and the | |
976 | stateful nature of memory, the distribution model is relatively | |
977 | complex. | |
978 | ||
979 | While not completely water-tight, all major memory usages by a given | |
980 | cgroup are tracked so that the total memory consumption can be | |
981 | accounted and controlled to a reasonable extent. Currently, the | |
982 | following types of memory usages are tracked. | |
983 | ||
984 | - Userland memory - page cache and anonymous memory. | |
985 | ||
986 | - Kernel data structures such as dentries and inodes. | |
987 | ||
988 | - TCP socket buffers. | |
989 | ||
990 | The above list may expand in the future for better coverage. | |
991 | ||
992 | ||
633b11be MCC |
993 | Memory Interface Files |
994 | ~~~~~~~~~~~~~~~~~~~~~~ | |
6c292092 TH |
995 | |
996 | All memory amounts are in bytes. If a value which is not aligned to | |
997 | PAGE_SIZE is written, the value may be rounded up to the closest | |
998 | PAGE_SIZE multiple when read back. | |
999 | ||
1000 | memory.current | |
6c292092 TH |
1001 | A read-only single value file which exists on non-root |
1002 | cgroups. | |
1003 | ||
1004 | The total amount of memory currently being used by the cgroup | |
1005 | and its descendants. | |
1006 | ||
bf8d5d52 RG |
1007 | memory.min |
1008 | A read-write single value file which exists on non-root | |
1009 | cgroups. The default is "0". | |
1010 | ||
1011 | Hard memory protection. If the memory usage of a cgroup | |
1012 | is within its effective min boundary, the cgroup's memory | |
1013 | won't be reclaimed under any conditions. If there is no | |
1014 | unprotected reclaimable memory available, OOM killer | |
1015 | is invoked. | |
1016 | ||
1017 | Effective min boundary is limited by memory.min values of | |
1018 | all ancestor cgroups. If there is memory.min overcommitment | |
1019 | (child cgroup or cgroups are requiring more protected memory | |
1020 | than parent will allow), then each child cgroup will get | |
1021 | the part of parent's protection proportional to its | |
1022 | actual memory usage below memory.min. | |
1023 | ||
1024 | Putting more memory than generally available under this | |
1025 | protection is discouraged and may lead to constant OOMs. | |
1026 | ||
1027 | If a memory cgroup is not populated with processes, | |
1028 | its memory.min is ignored. | |
1029 | ||
6c292092 | 1030 | memory.low |
6c292092 TH |
1031 | A read-write single value file which exists on non-root |
1032 | cgroups. The default is "0". | |
1033 | ||
7854207f RG |
1034 | Best-effort memory protection. If the memory usage of a |
1035 | cgroup is within its effective low boundary, the cgroup's | |
1036 | memory won't be reclaimed unless memory can be reclaimed | |
1037 | from unprotected cgroups. | |
1038 | ||
1039 | Effective low boundary is limited by memory.low values of | |
1040 | all ancestor cgroups. If there is memory.low overcommitment | |
bf8d5d52 | 1041 | (child cgroup or cgroups are requiring more protected memory |
7854207f | 1042 | than parent will allow), then each child cgroup will get |
bf8d5d52 | 1043 | the part of parent's protection proportional to its |
7854207f | 1044 | actual memory usage below memory.low. |
6c292092 TH |
1045 | |
1046 | Putting more memory than generally available under this | |
1047 | protection is discouraged. | |
1048 | ||
1049 | memory.high | |
6c292092 TH |
1050 | A read-write single value file which exists on non-root |
1051 | cgroups. The default is "max". | |
1052 | ||
1053 | Memory usage throttle limit. This is the main mechanism to | |
1054 | control memory usage of a cgroup. If a cgroup's usage goes | |
1055 | over the high boundary, the processes of the cgroup are | |
1056 | throttled and put under heavy reclaim pressure. | |
1057 | ||
1058 | Going over the high limit never invokes the OOM killer and | |
1059 | under extreme conditions the limit may be breached. | |
1060 | ||
1061 | memory.max | |
6c292092 TH |
1062 | A read-write single value file which exists on non-root |
1063 | cgroups. The default is "max". | |
1064 | ||
1065 | Memory usage hard limit. This is the final protection | |
1066 | mechanism. If a cgroup's memory usage reaches this limit and | |
1067 | can't be reduced, the OOM killer is invoked in the cgroup. | |
1068 | Under certain circumstances, the usage may go over the limit | |
1069 | temporarily. | |
1070 | ||
1071 | This is the ultimate protection mechanism. As long as the | |
1072 | high limit is used and monitored properly, this limit's | |
1073 | utility is limited to providing the final safety net. | |
1074 | ||
1075 | memory.events | |
6c292092 TH |
1076 | A read-only flat-keyed file which exists on non-root cgroups. |
1077 | The following entries are defined. Unless specified | |
1078 | otherwise, a value change in this file generates a file | |
1079 | modified event. | |
1080 | ||
1081 | low | |
6c292092 TH |
1082 | The number of times the cgroup is reclaimed due to |
1083 | high memory pressure even though its usage is under | |
1084 | the low boundary. This usually indicates that the low | |
1085 | boundary is over-committed. | |
1086 | ||
1087 | high | |
6c292092 TH |
1088 | The number of times processes of the cgroup are |
1089 | throttled and routed to perform direct memory reclaim | |
1090 | because the high memory boundary was exceeded. For a | |
1091 | cgroup whose memory usage is capped by the high limit | |
1092 | rather than global memory pressure, this event's | |
1093 | occurrences are expected. | |
1094 | ||
1095 | max | |
6c292092 TH |
1096 | The number of times the cgroup's memory usage was |
1097 | about to go over the max boundary. If direct reclaim | |
8e675f7a | 1098 | fails to bring it down, the cgroup goes to OOM state. |
6c292092 TH |
1099 | |
1100 | oom | |
8e675f7a KK |
1101 | The number of time the cgroup's memory usage was |
1102 | reached the limit and allocation was about to fail. | |
1103 | ||
1104 | Depending on context result could be invocation of OOM | |
2877cbe6 | 1105 | killer and retrying allocation or failing allocation. |
8e675f7a KK |
1106 | |
1107 | Failed allocation in its turn could be returned into | |
2877cbe6 | 1108 | userspace as -ENOMEM or silently ignored in cases like |
633b11be | 1109 | disk readahead. For now OOM in memory cgroup kills |
8e675f7a KK |
1110 | tasks iff shortage has happened inside page fault. |
1111 | ||
1112 | oom_kill | |
8e675f7a KK |
1113 | The number of processes belonging to this cgroup |
1114 | killed by any kind of OOM killer. | |
6c292092 | 1115 | |
587d9f72 | 1116 | memory.stat |
587d9f72 JW |
1117 | A read-only flat-keyed file which exists on non-root cgroups. |
1118 | ||
1119 | This breaks down the cgroup's memory footprint into different | |
1120 | types of memory, type-specific details, and other information | |
1121 | on the state and past events of the memory management system. | |
1122 | ||
1123 | All memory amounts are in bytes. | |
1124 | ||
1125 | The entries are ordered to be human readable, and new entries | |
1126 | can show up in the middle. Don't rely on items remaining in a | |
1127 | fixed position; use the keys to look up specific values! | |
1128 | ||
1129 | anon | |
587d9f72 JW |
1130 | Amount of memory used in anonymous mappings such as |
1131 | brk(), sbrk(), and mmap(MAP_ANONYMOUS) | |
1132 | ||
1133 | file | |
587d9f72 JW |
1134 | Amount of memory used to cache filesystem data, |
1135 | including tmpfs and shared memory. | |
1136 | ||
12580e4b | 1137 | kernel_stack |
12580e4b VD |
1138 | Amount of memory allocated to kernel stacks. |
1139 | ||
27ee57c9 | 1140 | slab |
27ee57c9 VD |
1141 | Amount of memory used for storing in-kernel data |
1142 | structures. | |
1143 | ||
4758e198 | 1144 | sock |
4758e198 JW |
1145 | Amount of memory used in network transmission buffers |
1146 | ||
9a4caf1e | 1147 | shmem |
9a4caf1e JW |
1148 | Amount of cached filesystem data that is swap-backed, |
1149 | such as tmpfs, shm segments, shared anonymous mmap()s | |
1150 | ||
587d9f72 | 1151 | file_mapped |
587d9f72 JW |
1152 | Amount of cached filesystem data mapped with mmap() |
1153 | ||
1154 | file_dirty | |
587d9f72 JW |
1155 | Amount of cached filesystem data that was modified but |
1156 | not yet written back to disk | |
1157 | ||
1158 | file_writeback | |
587d9f72 JW |
1159 | Amount of cached filesystem data that was modified and |
1160 | is currently being written back to disk | |
1161 | ||
633b11be | 1162 | inactive_anon, active_anon, inactive_file, active_file, unevictable |
587d9f72 JW |
1163 | Amount of memory, swap-backed and filesystem-backed, |
1164 | on the internal memory management lists used by the | |
1165 | page reclaim algorithm | |
1166 | ||
27ee57c9 | 1167 | slab_reclaimable |
27ee57c9 VD |
1168 | Part of "slab" that might be reclaimed, such as |
1169 | dentries and inodes. | |
1170 | ||
1171 | slab_unreclaimable | |
27ee57c9 VD |
1172 | Part of "slab" that cannot be reclaimed on memory |
1173 | pressure. | |
1174 | ||
587d9f72 | 1175 | pgfault |
587d9f72 JW |
1176 | Total number of page faults incurred |
1177 | ||
1178 | pgmajfault | |
587d9f72 JW |
1179 | Number of major page faults incurred |
1180 | ||
b340959e RG |
1181 | workingset_refault |
1182 | ||
1183 | Number of refaults of previously evicted pages | |
1184 | ||
1185 | workingset_activate | |
1186 | ||
1187 | Number of refaulted pages that were immediately activated | |
1188 | ||
1189 | workingset_nodereclaim | |
1190 | ||
1191 | Number of times a shadow node has been reclaimed | |
1192 | ||
2262185c RG |
1193 | pgrefill |
1194 | ||
1195 | Amount of scanned pages (in an active LRU list) | |
1196 | ||
1197 | pgscan | |
1198 | ||
1199 | Amount of scanned pages (in an inactive LRU list) | |
1200 | ||
1201 | pgsteal | |
1202 | ||
1203 | Amount of reclaimed pages | |
1204 | ||
1205 | pgactivate | |
1206 | ||
1207 | Amount of pages moved to the active LRU list | |
1208 | ||
1209 | pgdeactivate | |
1210 | ||
1211 | Amount of pages moved to the inactive LRU lis | |
1212 | ||
1213 | pglazyfree | |
1214 | ||
1215 | Amount of pages postponed to be freed under memory pressure | |
1216 | ||
1217 | pglazyfreed | |
1218 | ||
1219 | Amount of reclaimed lazyfree pages | |
1220 | ||
3e24b19d | 1221 | memory.swap.current |
3e24b19d VD |
1222 | A read-only single value file which exists on non-root |
1223 | cgroups. | |
1224 | ||
1225 | The total amount of swap currently being used by the cgroup | |
1226 | and its descendants. | |
1227 | ||
1228 | memory.swap.max | |
3e24b19d VD |
1229 | A read-write single value file which exists on non-root |
1230 | cgroups. The default is "max". | |
1231 | ||
1232 | Swap usage hard limit. If a cgroup's swap usage reaches this | |
2877cbe6 | 1233 | limit, anonymous memory of the cgroup will not be swapped out. |
3e24b19d | 1234 | |
f3a53a3a TH |
1235 | memory.swap.events |
1236 | A read-only flat-keyed file which exists on non-root cgroups. | |
1237 | The following entries are defined. Unless specified | |
1238 | otherwise, a value change in this file generates a file | |
1239 | modified event. | |
1240 | ||
1241 | max | |
1242 | The number of times the cgroup's swap usage was about | |
1243 | to go over the max boundary and swap allocation | |
1244 | failed. | |
1245 | ||
1246 | fail | |
1247 | The number of times swap allocation failed either | |
1248 | because of running out of swap system-wide or max | |
1249 | limit. | |
1250 | ||
be09102b TH |
1251 | When reduced under the current usage, the existing swap |
1252 | entries are reclaimed gradually and the swap usage may stay | |
1253 | higher than the limit for an extended period of time. This | |
1254 | reduces the impact on the workload and memory management. | |
1255 | ||
6c292092 | 1256 | |
633b11be MCC |
1257 | Usage Guidelines |
1258 | ~~~~~~~~~~~~~~~~ | |
6c292092 TH |
1259 | |
1260 | "memory.high" is the main mechanism to control memory usage. | |
1261 | Over-committing on high limit (sum of high limits > available memory) | |
1262 | and letting global memory pressure to distribute memory according to | |
1263 | usage is a viable strategy. | |
1264 | ||
1265 | Because breach of the high limit doesn't trigger the OOM killer but | |
1266 | throttles the offending cgroup, a management agent has ample | |
1267 | opportunities to monitor and take appropriate actions such as granting | |
1268 | more memory or terminating the workload. | |
1269 | ||
1270 | Determining whether a cgroup has enough memory is not trivial as | |
1271 | memory usage doesn't indicate whether the workload can benefit from | |
1272 | more memory. For example, a workload which writes data received from | |
1273 | network to a file can use all available memory but can also operate as | |
1274 | performant with a small amount of memory. A measure of memory | |
1275 | pressure - how much the workload is being impacted due to lack of | |
1276 | memory - is necessary to determine whether a workload needs more | |
1277 | memory; unfortunately, memory pressure monitoring mechanism isn't | |
1278 | implemented yet. | |
1279 | ||
1280 | ||
633b11be MCC |
1281 | Memory Ownership |
1282 | ~~~~~~~~~~~~~~~~ | |
6c292092 TH |
1283 | |
1284 | A memory area is charged to the cgroup which instantiated it and stays | |
1285 | charged to the cgroup until the area is released. Migrating a process | |
1286 | to a different cgroup doesn't move the memory usages that it | |
1287 | instantiated while in the previous cgroup to the new cgroup. | |
1288 | ||
1289 | A memory area may be used by processes belonging to different cgroups. | |
1290 | To which cgroup the area will be charged is in-deterministic; however, | |
1291 | over time, the memory area is likely to end up in a cgroup which has | |
1292 | enough memory allowance to avoid high reclaim pressure. | |
1293 | ||
1294 | If a cgroup sweeps a considerable amount of memory which is expected | |
1295 | to be accessed repeatedly by other cgroups, it may make sense to use | |
1296 | POSIX_FADV_DONTNEED to relinquish the ownership of memory areas | |
1297 | belonging to the affected files to ensure correct memory ownership. | |
1298 | ||
1299 | ||
633b11be MCC |
1300 | IO |
1301 | -- | |
6c292092 TH |
1302 | |
1303 | The "io" controller regulates the distribution of IO resources. This | |
1304 | controller implements both weight based and absolute bandwidth or IOPS | |
1305 | limit distribution; however, weight based distribution is available | |
1306 | only if cfq-iosched is in use and neither scheme is available for | |
1307 | blk-mq devices. | |
1308 | ||
1309 | ||
633b11be MCC |
1310 | IO Interface Files |
1311 | ~~~~~~~~~~~~~~~~~~ | |
6c292092 TH |
1312 | |
1313 | io.stat | |
6c292092 TH |
1314 | A read-only nested-keyed file which exists on non-root |
1315 | cgroups. | |
1316 | ||
1317 | Lines are keyed by $MAJ:$MIN device numbers and not ordered. | |
1318 | The following nested keys are defined. | |
1319 | ||
636620b6 | 1320 | ====== ===================== |
6c292092 TH |
1321 | rbytes Bytes read |
1322 | wbytes Bytes written | |
1323 | rios Number of read IOs | |
1324 | wios Number of write IOs | |
636620b6 TH |
1325 | dbytes Bytes discarded |
1326 | dios Number of discard IOs | |
1327 | ====== ===================== | |
6c292092 | 1328 | |
633b11be | 1329 | An example read output follows: |
6c292092 | 1330 | |
636620b6 TH |
1331 | 8:16 rbytes=1459200 wbytes=314773504 rios=192 wios=353 dbytes=0 dios=0 |
1332 | 8:0 rbytes=90430464 wbytes=299008000 rios=8950 wios=1252 dbytes=50331648 dios=3021 | |
6c292092 TH |
1333 | |
1334 | io.weight | |
6c292092 TH |
1335 | A read-write flat-keyed file which exists on non-root cgroups. |
1336 | The default is "default 100". | |
1337 | ||
1338 | The first line is the default weight applied to devices | |
1339 | without specific override. The rest are overrides keyed by | |
1340 | $MAJ:$MIN device numbers and not ordered. The weights are in | |
1341 | the range [1, 10000] and specifies the relative amount IO time | |
1342 | the cgroup can use in relation to its siblings. | |
1343 | ||
1344 | The default weight can be updated by writing either "default | |
1345 | $WEIGHT" or simply "$WEIGHT". Overrides can be set by writing | |
1346 | "$MAJ:$MIN $WEIGHT" and unset by writing "$MAJ:$MIN default". | |
1347 | ||
633b11be | 1348 | An example read output follows:: |
6c292092 TH |
1349 | |
1350 | default 100 | |
1351 | 8:16 200 | |
1352 | 8:0 50 | |
1353 | ||
1354 | io.max | |
6c292092 TH |
1355 | A read-write nested-keyed file which exists on non-root |
1356 | cgroups. | |
1357 | ||
1358 | BPS and IOPS based IO limit. Lines are keyed by $MAJ:$MIN | |
1359 | device numbers and not ordered. The following nested keys are | |
1360 | defined. | |
1361 | ||
633b11be | 1362 | ===== ================================== |
6c292092 TH |
1363 | rbps Max read bytes per second |
1364 | wbps Max write bytes per second | |
1365 | riops Max read IO operations per second | |
1366 | wiops Max write IO operations per second | |
633b11be | 1367 | ===== ================================== |
6c292092 TH |
1368 | |
1369 | When writing, any number of nested key-value pairs can be | |
1370 | specified in any order. "max" can be specified as the value | |
1371 | to remove a specific limit. If the same key is specified | |
1372 | multiple times, the outcome is undefined. | |
1373 | ||
1374 | BPS and IOPS are measured in each IO direction and IOs are | |
1375 | delayed if limit is reached. Temporary bursts are allowed. | |
1376 | ||
633b11be | 1377 | Setting read limit at 2M BPS and write at 120 IOPS for 8:16:: |
6c292092 TH |
1378 | |
1379 | echo "8:16 rbps=2097152 wiops=120" > io.max | |
1380 | ||
633b11be | 1381 | Reading returns the following:: |
6c292092 TH |
1382 | |
1383 | 8:16 rbps=2097152 wbps=max riops=max wiops=120 | |
1384 | ||
633b11be | 1385 | Write IOPS limit can be removed by writing the following:: |
6c292092 TH |
1386 | |
1387 | echo "8:16 wiops=max" > io.max | |
1388 | ||
633b11be | 1389 | Reading now returns the following:: |
6c292092 TH |
1390 | |
1391 | 8:16 rbps=2097152 wbps=max riops=max wiops=max | |
1392 | ||
1393 | ||
633b11be MCC |
1394 | Writeback |
1395 | ~~~~~~~~~ | |
6c292092 TH |
1396 | |
1397 | Page cache is dirtied through buffered writes and shared mmaps and | |
1398 | written asynchronously to the backing filesystem by the writeback | |
1399 | mechanism. Writeback sits between the memory and IO domains and | |
1400 | regulates the proportion of dirty memory by balancing dirtying and | |
1401 | write IOs. | |
1402 | ||
1403 | The io controller, in conjunction with the memory controller, | |
1404 | implements control of page cache writeback IOs. The memory controller | |
1405 | defines the memory domain that dirty memory ratio is calculated and | |
1406 | maintained for and the io controller defines the io domain which | |
1407 | writes out dirty pages for the memory domain. Both system-wide and | |
1408 | per-cgroup dirty memory states are examined and the more restrictive | |
1409 | of the two is enforced. | |
1410 | ||
1411 | cgroup writeback requires explicit support from the underlying | |
1412 | filesystem. Currently, cgroup writeback is implemented on ext2, ext4 | |
1413 | and btrfs. On other filesystems, all writeback IOs are attributed to | |
1414 | the root cgroup. | |
1415 | ||
1416 | There are inherent differences in memory and writeback management | |
1417 | which affects how cgroup ownership is tracked. Memory is tracked per | |
1418 | page while writeback per inode. For the purpose of writeback, an | |
1419 | inode is assigned to a cgroup and all IO requests to write dirty pages | |
1420 | from the inode are attributed to that cgroup. | |
1421 | ||
1422 | As cgroup ownership for memory is tracked per page, there can be pages | |
1423 | which are associated with different cgroups than the one the inode is | |
1424 | associated with. These are called foreign pages. The writeback | |
1425 | constantly keeps track of foreign pages and, if a particular foreign | |
1426 | cgroup becomes the majority over a certain period of time, switches | |
1427 | the ownership of the inode to that cgroup. | |
1428 | ||
1429 | While this model is enough for most use cases where a given inode is | |
1430 | mostly dirtied by a single cgroup even when the main writing cgroup | |
1431 | changes over time, use cases where multiple cgroups write to a single | |
1432 | inode simultaneously are not supported well. In such circumstances, a | |
1433 | significant portion of IOs are likely to be attributed incorrectly. | |
1434 | As memory controller assigns page ownership on the first use and | |
1435 | doesn't update it until the page is released, even if writeback | |
1436 | strictly follows page ownership, multiple cgroups dirtying overlapping | |
1437 | areas wouldn't work as expected. It's recommended to avoid such usage | |
1438 | patterns. | |
1439 | ||
1440 | The sysctl knobs which affect writeback behavior are applied to cgroup | |
1441 | writeback as follows. | |
1442 | ||
633b11be | 1443 | vm.dirty_background_ratio, vm.dirty_ratio |
6c292092 TH |
1444 | These ratios apply the same to cgroup writeback with the |
1445 | amount of available memory capped by limits imposed by the | |
1446 | memory controller and system-wide clean memory. | |
1447 | ||
633b11be | 1448 | vm.dirty_background_bytes, vm.dirty_bytes |
6c292092 TH |
1449 | For cgroup writeback, this is calculated into ratio against |
1450 | total available memory and applied the same way as | |
1451 | vm.dirty[_background]_ratio. | |
1452 | ||
1453 | ||
b351f0c7 JB |
1454 | IO Latency |
1455 | ~~~~~~~~~~ | |
1456 | ||
1457 | This is a cgroup v2 controller for IO workload protection. You provide a group | |
1458 | with a latency target, and if the average latency exceeds that target the | |
1459 | controller will throttle any peers that have a lower latency target than the | |
1460 | protected workload. | |
1461 | ||
1462 | The limits are only applied at the peer level in the hierarchy. This means that | |
1463 | in the diagram below, only groups A, B, and C will influence each other, and | |
1464 | groups D and F will influence each other. Group G will influence nobody. | |
1465 | ||
1466 | [root] | |
1467 | / | \ | |
1468 | A B C | |
1469 | / \ | | |
1470 | D F G | |
1471 | ||
1472 | ||
1473 | So the ideal way to configure this is to set io.latency in groups A, B, and C. | |
1474 | Generally you do not want to set a value lower than the latency your device | |
1475 | supports. Experiment to find the value that works best for your workload. | |
1476 | Start at higher than the expected latency for your device and watch the | |
1477 | total_lat_avg value in io.stat for your workload group to get an idea of the | |
1478 | latency you see during normal operation. Use this value as a basis for your | |
1479 | real setting, setting at 10-15% higher than the value in io.stat. | |
1480 | Experimentation is key here because total_lat_avg is a running total, so is the | |
1481 | "statistics" portion of "lies, damned lies, and statistics." | |
1482 | ||
1483 | How IO Latency Throttling Works | |
1484 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | |
1485 | ||
1486 | io.latency is work conserving; so as long as everybody is meeting their latency | |
1487 | target the controller doesn't do anything. Once a group starts missing its | |
1488 | target it begins throttling any peer group that has a higher target than itself. | |
1489 | This throttling takes 2 forms: | |
1490 | ||
1491 | - Queue depth throttling. This is the number of outstanding IO's a group is | |
1492 | allowed to have. We will clamp down relatively quickly, starting at no limit | |
1493 | and going all the way down to 1 IO at a time. | |
1494 | ||
1495 | - Artificial delay induction. There are certain types of IO that cannot be | |
1496 | throttled without possibly adversely affecting higher priority groups. This | |
1497 | includes swapping and metadata IO. These types of IO are allowed to occur | |
1498 | normally, however they are "charged" to the originating group. If the | |
1499 | originating group is being throttled you will see the use_delay and delay | |
1500 | fields in io.stat increase. The delay value is how many microseconds that are | |
1501 | being added to any process that runs in this group. Because this number can | |
1502 | grow quite large if there is a lot of swapping or metadata IO occurring we | |
1503 | limit the individual delay events to 1 second at a time. | |
1504 | ||
1505 | Once the victimized group starts meeting its latency target again it will start | |
1506 | unthrottling any peer groups that were throttled previously. If the victimized | |
1507 | group simply stops doing IO the global counter will unthrottle appropriately. | |
1508 | ||
1509 | IO Latency Interface Files | |
1510 | ~~~~~~~~~~~~~~~~~~~~~~~~~~ | |
1511 | ||
1512 | io.latency | |
1513 | This takes a similar format as the other controllers. | |
1514 | ||
1515 | "MAJOR:MINOR target=<target time in microseconds" | |
1516 | ||
1517 | io.stat | |
1518 | If the controller is enabled you will see extra stats in io.stat in | |
1519 | addition to the normal ones. | |
1520 | ||
1521 | depth | |
1522 | This is the current queue depth for the group. | |
1523 | ||
1524 | avg_lat | |
1525 | The running average IO latency for this group in microseconds. | |
1526 | Running average is generally flawed, but will give an | |
1527 | administrator a general idea of the overall latency they can | |
1528 | expect for their workload on the given disk. | |
1529 | ||
633b11be MCC |
1530 | PID |
1531 | --- | |
20c56e59 HR |
1532 | |
1533 | The process number controller is used to allow a cgroup to stop any | |
1534 | new tasks from being fork()'d or clone()'d after a specified limit is | |
1535 | reached. | |
1536 | ||
1537 | The number of tasks in a cgroup can be exhausted in ways which other | |
1538 | controllers cannot prevent, thus warranting its own controller. For | |
1539 | example, a fork bomb is likely to exhaust the number of tasks before | |
1540 | hitting memory restrictions. | |
1541 | ||
1542 | Note that PIDs used in this controller refer to TIDs, process IDs as | |
1543 | used by the kernel. | |
1544 | ||
1545 | ||
633b11be MCC |
1546 | PID Interface Files |
1547 | ~~~~~~~~~~~~~~~~~~~ | |
20c56e59 HR |
1548 | |
1549 | pids.max | |
312eb712 TK |
1550 | A read-write single value file which exists on non-root |
1551 | cgroups. The default is "max". | |
20c56e59 | 1552 | |
312eb712 | 1553 | Hard limit of number of processes. |
20c56e59 HR |
1554 | |
1555 | pids.current | |
312eb712 | 1556 | A read-only single value file which exists on all cgroups. |
20c56e59 | 1557 | |
312eb712 TK |
1558 | The number of processes currently in the cgroup and its |
1559 | descendants. | |
20c56e59 HR |
1560 | |
1561 | Organisational operations are not blocked by cgroup policies, so it is | |
1562 | possible to have pids.current > pids.max. This can be done by either | |
1563 | setting the limit to be smaller than pids.current, or attaching enough | |
1564 | processes to the cgroup such that pids.current is larger than | |
1565 | pids.max. However, it is not possible to violate a cgroup PID policy | |
1566 | through fork() or clone(). These will return -EAGAIN if the creation | |
1567 | of a new process would cause a cgroup policy to be violated. | |
1568 | ||
1569 | ||
4ad5a321 RG |
1570 | Device controller |
1571 | ----------------- | |
1572 | ||
1573 | Device controller manages access to device files. It includes both | |
1574 | creation of new device files (using mknod), and access to the | |
1575 | existing device files. | |
1576 | ||
1577 | Cgroup v2 device controller has no interface files and is implemented | |
1578 | on top of cgroup BPF. To control access to device files, a user may | |
1579 | create bpf programs of the BPF_CGROUP_DEVICE type and attach them | |
1580 | to cgroups. On an attempt to access a device file, corresponding | |
1581 | BPF programs will be executed, and depending on the return value | |
1582 | the attempt will succeed or fail with -EPERM. | |
1583 | ||
1584 | A BPF_CGROUP_DEVICE program takes a pointer to the bpf_cgroup_dev_ctx | |
1585 | structure, which describes the device access attempt: access type | |
1586 | (mknod/read/write) and device (type, major and minor numbers). | |
1587 | If the program returns 0, the attempt fails with -EPERM, otherwise | |
1588 | it succeeds. | |
1589 | ||
1590 | An example of BPF_CGROUP_DEVICE program may be found in the kernel | |
1591 | source tree in the tools/testing/selftests/bpf/dev_cgroup.c file. | |
1592 | ||
1593 | ||
633b11be MCC |
1594 | RDMA |
1595 | ---- | |
968ebff1 | 1596 | |
9c1e67f9 PP |
1597 | The "rdma" controller regulates the distribution and accounting of |
1598 | of RDMA resources. | |
1599 | ||
633b11be MCC |
1600 | RDMA Interface Files |
1601 | ~~~~~~~~~~~~~~~~~~~~ | |
9c1e67f9 PP |
1602 | |
1603 | rdma.max | |
1604 | A readwrite nested-keyed file that exists for all the cgroups | |
1605 | except root that describes current configured resource limit | |
1606 | for a RDMA/IB device. | |
1607 | ||
1608 | Lines are keyed by device name and are not ordered. | |
1609 | Each line contains space separated resource name and its configured | |
1610 | limit that can be distributed. | |
1611 | ||
1612 | The following nested keys are defined. | |
1613 | ||
633b11be | 1614 | ========== ============================= |
9c1e67f9 PP |
1615 | hca_handle Maximum number of HCA Handles |
1616 | hca_object Maximum number of HCA Objects | |
633b11be | 1617 | ========== ============================= |
9c1e67f9 | 1618 | |
633b11be | 1619 | An example for mlx4 and ocrdma device follows:: |
9c1e67f9 PP |
1620 | |
1621 | mlx4_0 hca_handle=2 hca_object=2000 | |
1622 | ocrdma1 hca_handle=3 hca_object=max | |
1623 | ||
1624 | rdma.current | |
1625 | A read-only file that describes current resource usage. | |
1626 | It exists for all the cgroup except root. | |
1627 | ||
633b11be | 1628 | An example for mlx4 and ocrdma device follows:: |
9c1e67f9 PP |
1629 | |
1630 | mlx4_0 hca_handle=1 hca_object=20 | |
1631 | ocrdma1 hca_handle=1 hca_object=23 | |
1632 | ||
1633 | ||
633b11be MCC |
1634 | Misc |
1635 | ---- | |
63f1ca59 | 1636 | |
633b11be MCC |
1637 | perf_event |
1638 | ~~~~~~~~~~ | |
968ebff1 TH |
1639 | |
1640 | perf_event controller, if not mounted on a legacy hierarchy, is | |
1641 | automatically enabled on the v2 hierarchy so that perf events can | |
1642 | always be filtered by cgroup v2 path. The controller can still be | |
1643 | moved to a legacy hierarchy after v2 hierarchy is populated. | |
1644 | ||
1645 | ||
c4e0842b MS |
1646 | Non-normative information |
1647 | ------------------------- | |
1648 | ||
1649 | This section contains information that isn't considered to be a part of | |
1650 | the stable kernel API and so is subject to change. | |
1651 | ||
1652 | ||
1653 | CPU controller root cgroup process behaviour | |
1654 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | |
1655 | ||
1656 | When distributing CPU cycles in the root cgroup each thread in this | |
1657 | cgroup is treated as if it was hosted in a separate child cgroup of the | |
1658 | root cgroup. This child cgroup weight is dependent on its thread nice | |
1659 | level. | |
1660 | ||
1661 | For details of this mapping see sched_prio_to_weight array in | |
1662 | kernel/sched/core.c file (values from this array should be scaled | |
1663 | appropriately so the neutral - nice 0 - value is 100 instead of 1024). | |
1664 | ||
1665 | ||
1666 | IO controller root cgroup process behaviour | |
1667 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | |
1668 | ||
1669 | Root cgroup processes are hosted in an implicit leaf child node. | |
1670 | When distributing IO resources this implicit child node is taken into | |
1671 | account as if it was a normal child cgroup of the root cgroup with a | |
1672 | weight value of 200. | |
1673 | ||
1674 | ||
633b11be MCC |
1675 | Namespace |
1676 | ========= | |
d4021f6c | 1677 | |
633b11be MCC |
1678 | Basics |
1679 | ------ | |
d4021f6c SH |
1680 | |
1681 | cgroup namespace provides a mechanism to virtualize the view of the | |
1682 | "/proc/$PID/cgroup" file and cgroup mounts. The CLONE_NEWCGROUP clone | |
1683 | flag can be used with clone(2) and unshare(2) to create a new cgroup | |
1684 | namespace. The process running inside the cgroup namespace will have | |
1685 | its "/proc/$PID/cgroup" output restricted to cgroupns root. The | |
1686 | cgroupns root is the cgroup of the process at the time of creation of | |
1687 | the cgroup namespace. | |
1688 | ||
1689 | Without cgroup namespace, the "/proc/$PID/cgroup" file shows the | |
1690 | complete path of the cgroup of a process. In a container setup where | |
1691 | a set of cgroups and namespaces are intended to isolate processes the | |
1692 | "/proc/$PID/cgroup" file may leak potential system level information | |
633b11be | 1693 | to the isolated processes. For Example:: |
d4021f6c SH |
1694 | |
1695 | # cat /proc/self/cgroup | |
1696 | 0::/batchjobs/container_id1 | |
1697 | ||
1698 | The path '/batchjobs/container_id1' can be considered as system-data | |
1699 | and undesirable to expose to the isolated processes. cgroup namespace | |
1700 | can be used to restrict visibility of this path. For example, before | |
633b11be | 1701 | creating a cgroup namespace, one would see:: |
d4021f6c SH |
1702 | |
1703 | # ls -l /proc/self/ns/cgroup | |
1704 | lrwxrwxrwx 1 root root 0 2014-07-15 10:37 /proc/self/ns/cgroup -> cgroup:[4026531835] | |
1705 | # cat /proc/self/cgroup | |
1706 | 0::/batchjobs/container_id1 | |
1707 | ||
633b11be | 1708 | After unsharing a new namespace, the view changes:: |
d4021f6c SH |
1709 | |
1710 | # ls -l /proc/self/ns/cgroup | |
1711 | lrwxrwxrwx 1 root root 0 2014-07-15 10:35 /proc/self/ns/cgroup -> cgroup:[4026532183] | |
1712 | # cat /proc/self/cgroup | |
1713 | 0::/ | |
1714 | ||
1715 | When some thread from a multi-threaded process unshares its cgroup | |
1716 | namespace, the new cgroupns gets applied to the entire process (all | |
1717 | the threads). This is natural for the v2 hierarchy; however, for the | |
1718 | legacy hierarchies, this may be unexpected. | |
1719 | ||
1720 | A cgroup namespace is alive as long as there are processes inside or | |
1721 | mounts pinning it. When the last usage goes away, the cgroup | |
1722 | namespace is destroyed. The cgroupns root and the actual cgroups | |
1723 | remain. | |
1724 | ||
1725 | ||
633b11be MCC |
1726 | The Root and Views |
1727 | ------------------ | |
d4021f6c SH |
1728 | |
1729 | The 'cgroupns root' for a cgroup namespace is the cgroup in which the | |
1730 | process calling unshare(2) is running. For example, if a process in | |
1731 | /batchjobs/container_id1 cgroup calls unshare, cgroup | |
1732 | /batchjobs/container_id1 becomes the cgroupns root. For the | |
1733 | init_cgroup_ns, this is the real root ('/') cgroup. | |
1734 | ||
1735 | The cgroupns root cgroup does not change even if the namespace creator | |
633b11be | 1736 | process later moves to a different cgroup:: |
d4021f6c SH |
1737 | |
1738 | # ~/unshare -c # unshare cgroupns in some cgroup | |
1739 | # cat /proc/self/cgroup | |
1740 | 0::/ | |
1741 | # mkdir sub_cgrp_1 | |
1742 | # echo 0 > sub_cgrp_1/cgroup.procs | |
1743 | # cat /proc/self/cgroup | |
1744 | 0::/sub_cgrp_1 | |
1745 | ||
1746 | Each process gets its namespace-specific view of "/proc/$PID/cgroup" | |
1747 | ||
1748 | Processes running inside the cgroup namespace will be able to see | |
1749 | cgroup paths (in /proc/self/cgroup) only inside their root cgroup. | |
633b11be | 1750 | From within an unshared cgroupns:: |
d4021f6c SH |
1751 | |
1752 | # sleep 100000 & | |
1753 | [1] 7353 | |
1754 | # echo 7353 > sub_cgrp_1/cgroup.procs | |
1755 | # cat /proc/7353/cgroup | |
1756 | 0::/sub_cgrp_1 | |
1757 | ||
1758 | From the initial cgroup namespace, the real cgroup path will be | |
633b11be | 1759 | visible:: |
d4021f6c SH |
1760 | |
1761 | $ cat /proc/7353/cgroup | |
1762 | 0::/batchjobs/container_id1/sub_cgrp_1 | |
1763 | ||
1764 | From a sibling cgroup namespace (that is, a namespace rooted at a | |
1765 | different cgroup), the cgroup path relative to its own cgroup | |
1766 | namespace root will be shown. For instance, if PID 7353's cgroup | |
633b11be | 1767 | namespace root is at '/batchjobs/container_id2', then it will see:: |
d4021f6c SH |
1768 | |
1769 | # cat /proc/7353/cgroup | |
1770 | 0::/../container_id2/sub_cgrp_1 | |
1771 | ||
1772 | Note that the relative path always starts with '/' to indicate that | |
1773 | its relative to the cgroup namespace root of the caller. | |
1774 | ||
1775 | ||
633b11be MCC |
1776 | Migration and setns(2) |
1777 | ---------------------- | |
d4021f6c SH |
1778 | |
1779 | Processes inside a cgroup namespace can move into and out of the | |
1780 | namespace root if they have proper access to external cgroups. For | |
1781 | example, from inside a namespace with cgroupns root at | |
1782 | /batchjobs/container_id1, and assuming that the global hierarchy is | |
633b11be | 1783 | still accessible inside cgroupns:: |
d4021f6c SH |
1784 | |
1785 | # cat /proc/7353/cgroup | |
1786 | 0::/sub_cgrp_1 | |
1787 | # echo 7353 > batchjobs/container_id2/cgroup.procs | |
1788 | # cat /proc/7353/cgroup | |
1789 | 0::/../container_id2 | |
1790 | ||
1791 | Note that this kind of setup is not encouraged. A task inside cgroup | |
1792 | namespace should only be exposed to its own cgroupns hierarchy. | |
1793 | ||
1794 | setns(2) to another cgroup namespace is allowed when: | |
1795 | ||
1796 | (a) the process has CAP_SYS_ADMIN against its current user namespace | |
1797 | (b) the process has CAP_SYS_ADMIN against the target cgroup | |
1798 | namespace's userns | |
1799 | ||
1800 | No implicit cgroup changes happen with attaching to another cgroup | |
1801 | namespace. It is expected that the someone moves the attaching | |
1802 | process under the target cgroup namespace root. | |
1803 | ||
1804 | ||
633b11be MCC |
1805 | Interaction with Other Namespaces |
1806 | --------------------------------- | |
d4021f6c SH |
1807 | |
1808 | Namespace specific cgroup hierarchy can be mounted by a process | |
633b11be | 1809 | running inside a non-init cgroup namespace:: |
d4021f6c SH |
1810 | |
1811 | # mount -t cgroup2 none $MOUNT_POINT | |
1812 | ||
1813 | This will mount the unified cgroup hierarchy with cgroupns root as the | |
1814 | filesystem root. The process needs CAP_SYS_ADMIN against its user and | |
1815 | mount namespaces. | |
1816 | ||
1817 | The virtualization of /proc/self/cgroup file combined with restricting | |
1818 | the view of cgroup hierarchy by namespace-private cgroupfs mount | |
1819 | provides a properly isolated cgroup view inside the container. | |
1820 | ||
1821 | ||
633b11be MCC |
1822 | Information on Kernel Programming |
1823 | ================================= | |
6c292092 TH |
1824 | |
1825 | This section contains kernel programming information in the areas | |
1826 | where interacting with cgroup is necessary. cgroup core and | |
1827 | controllers are not covered. | |
1828 | ||
1829 | ||
633b11be MCC |
1830 | Filesystem Support for Writeback |
1831 | -------------------------------- | |
6c292092 TH |
1832 | |
1833 | A filesystem can support cgroup writeback by updating | |
1834 | address_space_operations->writepage[s]() to annotate bio's using the | |
1835 | following two functions. | |
1836 | ||
1837 | wbc_init_bio(@wbc, @bio) | |
6c292092 TH |
1838 | Should be called for each bio carrying writeback data and |
1839 | associates the bio with the inode's owner cgroup. Can be | |
1840 | called anytime between bio allocation and submission. | |
1841 | ||
1842 | wbc_account_io(@wbc, @page, @bytes) | |
6c292092 TH |
1843 | Should be called for each data segment being written out. |
1844 | While this function doesn't care exactly when it's called | |
1845 | during the writeback session, it's the easiest and most | |
1846 | natural to call it as data segments are added to a bio. | |
1847 | ||
1848 | With writeback bio's annotated, cgroup support can be enabled per | |
1849 | super_block by setting SB_I_CGROUPWB in ->s_iflags. This allows for | |
1850 | selective disabling of cgroup writeback support which is helpful when | |
1851 | certain filesystem features, e.g. journaled data mode, are | |
1852 | incompatible. | |
1853 | ||
1854 | wbc_init_bio() binds the specified bio to its cgroup. Depending on | |
1855 | the configuration, the bio may be executed at a lower priority and if | |
1856 | the writeback session is holding shared resources, e.g. a journal | |
1857 | entry, may lead to priority inversion. There is no one easy solution | |
1858 | for the problem. Filesystems can try to work around specific problem | |
1859 | cases by skipping wbc_init_bio() or using bio_associate_blkcg() | |
1860 | directly. | |
1861 | ||
1862 | ||
633b11be MCC |
1863 | Deprecated v1 Core Features |
1864 | =========================== | |
6c292092 TH |
1865 | |
1866 | - Multiple hierarchies including named ones are not supported. | |
1867 | ||
5136f636 | 1868 | - All v1 mount options are not supported. |
6c292092 TH |
1869 | |
1870 | - The "tasks" file is removed and "cgroup.procs" is not sorted. | |
1871 | ||
1872 | - "cgroup.clone_children" is removed. | |
1873 | ||
1874 | - /proc/cgroups is meaningless for v2. Use "cgroup.controllers" file | |
1875 | at the root instead. | |
1876 | ||
1877 | ||
633b11be MCC |
1878 | Issues with v1 and Rationales for v2 |
1879 | ==================================== | |
6c292092 | 1880 | |
633b11be MCC |
1881 | Multiple Hierarchies |
1882 | -------------------- | |
6c292092 TH |
1883 | |
1884 | cgroup v1 allowed an arbitrary number of hierarchies and each | |
1885 | hierarchy could host any number of controllers. While this seemed to | |
1886 | provide a high level of flexibility, it wasn't useful in practice. | |
1887 | ||
1888 | For example, as there is only one instance of each controller, utility | |
1889 | type controllers such as freezer which can be useful in all | |
1890 | hierarchies could only be used in one. The issue is exacerbated by | |
1891 | the fact that controllers couldn't be moved to another hierarchy once | |
1892 | hierarchies were populated. Another issue was that all controllers | |
1893 | bound to a hierarchy were forced to have exactly the same view of the | |
1894 | hierarchy. It wasn't possible to vary the granularity depending on | |
1895 | the specific controller. | |
1896 | ||
1897 | In practice, these issues heavily limited which controllers could be | |
1898 | put on the same hierarchy and most configurations resorted to putting | |
1899 | each controller on its own hierarchy. Only closely related ones, such | |
1900 | as the cpu and cpuacct controllers, made sense to be put on the same | |
1901 | hierarchy. This often meant that userland ended up managing multiple | |
1902 | similar hierarchies repeating the same steps on each hierarchy | |
1903 | whenever a hierarchy management operation was necessary. | |
1904 | ||
1905 | Furthermore, support for multiple hierarchies came at a steep cost. | |
1906 | It greatly complicated cgroup core implementation but more importantly | |
1907 | the support for multiple hierarchies restricted how cgroup could be | |
1908 | used in general and what controllers was able to do. | |
1909 | ||
1910 | There was no limit on how many hierarchies there might be, which meant | |
1911 | that a thread's cgroup membership couldn't be described in finite | |
1912 | length. The key might contain any number of entries and was unlimited | |
1913 | in length, which made it highly awkward to manipulate and led to | |
1914 | addition of controllers which existed only to identify membership, | |
1915 | which in turn exacerbated the original problem of proliferating number | |
1916 | of hierarchies. | |
1917 | ||
1918 | Also, as a controller couldn't have any expectation regarding the | |
1919 | topologies of hierarchies other controllers might be on, each | |
1920 | controller had to assume that all other controllers were attached to | |
1921 | completely orthogonal hierarchies. This made it impossible, or at | |
1922 | least very cumbersome, for controllers to cooperate with each other. | |
1923 | ||
1924 | In most use cases, putting controllers on hierarchies which are | |
1925 | completely orthogonal to each other isn't necessary. What usually is | |
1926 | called for is the ability to have differing levels of granularity | |
1927 | depending on the specific controller. In other words, hierarchy may | |
1928 | be collapsed from leaf towards root when viewed from specific | |
1929 | controllers. For example, a given configuration might not care about | |
1930 | how memory is distributed beyond a certain level while still wanting | |
1931 | to control how CPU cycles are distributed. | |
1932 | ||
1933 | ||
633b11be MCC |
1934 | Thread Granularity |
1935 | ------------------ | |
6c292092 TH |
1936 | |
1937 | cgroup v1 allowed threads of a process to belong to different cgroups. | |
1938 | This didn't make sense for some controllers and those controllers | |
1939 | ended up implementing different ways to ignore such situations but | |
1940 | much more importantly it blurred the line between API exposed to | |
1941 | individual applications and system management interface. | |
1942 | ||
1943 | Generally, in-process knowledge is available only to the process | |
1944 | itself; thus, unlike service-level organization of processes, | |
1945 | categorizing threads of a process requires active participation from | |
1946 | the application which owns the target process. | |
1947 | ||
1948 | cgroup v1 had an ambiguously defined delegation model which got abused | |
1949 | in combination with thread granularity. cgroups were delegated to | |
1950 | individual applications so that they can create and manage their own | |
1951 | sub-hierarchies and control resource distributions along them. This | |
1952 | effectively raised cgroup to the status of a syscall-like API exposed | |
1953 | to lay programs. | |
1954 | ||
1955 | First of all, cgroup has a fundamentally inadequate interface to be | |
1956 | exposed this way. For a process to access its own knobs, it has to | |
1957 | extract the path on the target hierarchy from /proc/self/cgroup, | |
1958 | construct the path by appending the name of the knob to the path, open | |
1959 | and then read and/or write to it. This is not only extremely clunky | |
1960 | and unusual but also inherently racy. There is no conventional way to | |
1961 | define transaction across the required steps and nothing can guarantee | |
1962 | that the process would actually be operating on its own sub-hierarchy. | |
1963 | ||
1964 | cgroup controllers implemented a number of knobs which would never be | |
1965 | accepted as public APIs because they were just adding control knobs to | |
1966 | system-management pseudo filesystem. cgroup ended up with interface | |
1967 | knobs which were not properly abstracted or refined and directly | |
1968 | revealed kernel internal details. These knobs got exposed to | |
1969 | individual applications through the ill-defined delegation mechanism | |
1970 | effectively abusing cgroup as a shortcut to implementing public APIs | |
1971 | without going through the required scrutiny. | |
1972 | ||
1973 | This was painful for both userland and kernel. Userland ended up with | |
1974 | misbehaving and poorly abstracted interfaces and kernel exposing and | |
1975 | locked into constructs inadvertently. | |
1976 | ||
1977 | ||
633b11be MCC |
1978 | Competition Between Inner Nodes and Threads |
1979 | ------------------------------------------- | |
6c292092 TH |
1980 | |
1981 | cgroup v1 allowed threads to be in any cgroups which created an | |
1982 | interesting problem where threads belonging to a parent cgroup and its | |
1983 | children cgroups competed for resources. This was nasty as two | |
1984 | different types of entities competed and there was no obvious way to | |
1985 | settle it. Different controllers did different things. | |
1986 | ||
1987 | The cpu controller considered threads and cgroups as equivalents and | |
1988 | mapped nice levels to cgroup weights. This worked for some cases but | |
1989 | fell flat when children wanted to be allocated specific ratios of CPU | |
1990 | cycles and the number of internal threads fluctuated - the ratios | |
1991 | constantly changed as the number of competing entities fluctuated. | |
1992 | There also were other issues. The mapping from nice level to weight | |
1993 | wasn't obvious or universal, and there were various other knobs which | |
1994 | simply weren't available for threads. | |
1995 | ||
1996 | The io controller implicitly created a hidden leaf node for each | |
1997 | cgroup to host the threads. The hidden leaf had its own copies of all | |
633b11be | 1998 | the knobs with ``leaf_`` prefixed. While this allowed equivalent |
6c292092 TH |
1999 | control over internal threads, it was with serious drawbacks. It |
2000 | always added an extra layer of nesting which wouldn't be necessary | |
2001 | otherwise, made the interface messy and significantly complicated the | |
2002 | implementation. | |
2003 | ||
2004 | The memory controller didn't have a way to control what happened | |
2005 | between internal tasks and child cgroups and the behavior was not | |
2006 | clearly defined. There were attempts to add ad-hoc behaviors and | |
2007 | knobs to tailor the behavior to specific workloads which would have | |
2008 | led to problems extremely difficult to resolve in the long term. | |
2009 | ||
2010 | Multiple controllers struggled with internal tasks and came up with | |
2011 | different ways to deal with it; unfortunately, all the approaches were | |
2012 | severely flawed and, furthermore, the widely different behaviors | |
2013 | made cgroup as a whole highly inconsistent. | |
2014 | ||
2015 | This clearly is a problem which needs to be addressed from cgroup core | |
2016 | in a uniform way. | |
2017 | ||
2018 | ||
633b11be MCC |
2019 | Other Interface Issues |
2020 | ---------------------- | |
6c292092 TH |
2021 | |
2022 | cgroup v1 grew without oversight and developed a large number of | |
2023 | idiosyncrasies and inconsistencies. One issue on the cgroup core side | |
2024 | was how an empty cgroup was notified - a userland helper binary was | |
2025 | forked and executed for each event. The event delivery wasn't | |
2026 | recursive or delegatable. The limitations of the mechanism also led | |
2027 | to in-kernel event delivery filtering mechanism further complicating | |
2028 | the interface. | |
2029 | ||
2030 | Controller interfaces were problematic too. An extreme example is | |
2031 | controllers completely ignoring hierarchical organization and treating | |
2032 | all cgroups as if they were all located directly under the root | |
2033 | cgroup. Some controllers exposed a large amount of inconsistent | |
2034 | implementation details to userland. | |
2035 | ||
2036 | There also was no consistency across controllers. When a new cgroup | |
2037 | was created, some controllers defaulted to not imposing extra | |
2038 | restrictions while others disallowed any resource usage until | |
2039 | explicitly configured. Configuration knobs for the same type of | |
2040 | control used widely differing naming schemes and formats. Statistics | |
2041 | and information knobs were named arbitrarily and used different | |
2042 | formats and units even in the same controller. | |
2043 | ||
2044 | cgroup v2 establishes common conventions where appropriate and updates | |
2045 | controllers so that they expose minimal and consistent interfaces. | |
2046 | ||
2047 | ||
633b11be MCC |
2048 | Controller Issues and Remedies |
2049 | ------------------------------ | |
6c292092 | 2050 | |
633b11be MCC |
2051 | Memory |
2052 | ~~~~~~ | |
6c292092 TH |
2053 | |
2054 | The original lower boundary, the soft limit, is defined as a limit | |
2055 | that is per default unset. As a result, the set of cgroups that | |
2056 | global reclaim prefers is opt-in, rather than opt-out. The costs for | |
2057 | optimizing these mostly negative lookups are so high that the | |
2058 | implementation, despite its enormous size, does not even provide the | |
2059 | basic desirable behavior. First off, the soft limit has no | |
2060 | hierarchical meaning. All configured groups are organized in a global | |
2061 | rbtree and treated like equal peers, regardless where they are located | |
2062 | in the hierarchy. This makes subtree delegation impossible. Second, | |
2063 | the soft limit reclaim pass is so aggressive that it not just | |
2064 | introduces high allocation latencies into the system, but also impacts | |
2065 | system performance due to overreclaim, to the point where the feature | |
2066 | becomes self-defeating. | |
2067 | ||
2068 | The memory.low boundary on the other hand is a top-down allocated | |
7854207f RG |
2069 | reserve. A cgroup enjoys reclaim protection when it's within its low, |
2070 | which makes delegation of subtrees possible. | |
6c292092 TH |
2071 | |
2072 | The original high boundary, the hard limit, is defined as a strict | |
2073 | limit that can not budge, even if the OOM killer has to be called. | |
2074 | But this generally goes against the goal of making the most out of the | |
2075 | available memory. The memory consumption of workloads varies during | |
2076 | runtime, and that requires users to overcommit. But doing that with a | |
2077 | strict upper limit requires either a fairly accurate prediction of the | |
2078 | working set size or adding slack to the limit. Since working set size | |
2079 | estimation is hard and error prone, and getting it wrong results in | |
2080 | OOM kills, most users tend to err on the side of a looser limit and | |
2081 | end up wasting precious resources. | |
2082 | ||
2083 | The memory.high boundary on the other hand can be set much more | |
2084 | conservatively. When hit, it throttles allocations by forcing them | |
2085 | into direct reclaim to work off the excess, but it never invokes the | |
2086 | OOM killer. As a result, a high boundary that is chosen too | |
2087 | aggressively will not terminate the processes, but instead it will | |
2088 | lead to gradual performance degradation. The user can monitor this | |
2089 | and make corrections until the minimal memory footprint that still | |
2090 | gives acceptable performance is found. | |
2091 | ||
2092 | In extreme cases, with many concurrent allocations and a complete | |
2093 | breakdown of reclaim progress within the group, the high boundary can | |
2094 | be exceeded. But even then it's mostly better to satisfy the | |
2095 | allocation from the slack available in other groups or the rest of the | |
2096 | system than killing the group. Otherwise, memory.max is there to | |
2097 | limit this type of spillover and ultimately contain buggy or even | |
2098 | malicious applications. | |
3e24b19d | 2099 | |
b6e6edcf JW |
2100 | Setting the original memory.limit_in_bytes below the current usage was |
2101 | subject to a race condition, where concurrent charges could cause the | |
2102 | limit setting to fail. memory.max on the other hand will first set the | |
2103 | limit to prevent new charges, and then reclaim and OOM kill until the | |
2104 | new limit is met - or the task writing to memory.max is killed. | |
2105 | ||
3e24b19d VD |
2106 | The combined memory+swap accounting and limiting is replaced by real |
2107 | control over swap space. | |
2108 | ||
2109 | The main argument for a combined memory+swap facility in the original | |
2110 | cgroup design was that global or parental pressure would always be | |
2111 | able to swap all anonymous memory of a child group, regardless of the | |
2112 | child's own (possibly untrusted) configuration. However, untrusted | |
2113 | groups can sabotage swapping by other means - such as referencing its | |
2114 | anonymous memory in a tight loop - and an admin can not assume full | |
2115 | swappability when overcommitting untrusted jobs. | |
2116 | ||
2117 | For trusted jobs, on the other hand, a combined counter is not an | |
2118 | intuitive userspace interface, and it flies in the face of the idea | |
2119 | that cgroup controllers should account and limit specific physical | |
2120 | resources. Swap space is a resource like all others in the system, | |
2121 | and that's why unified hierarchy allows distributing it separately. |