1 ==========================
2 Memory Resource Controller
3 ==========================
6 This document is hopelessly outdated and it asks for a complete
7 rewrite. It still contains a useful information so we are keeping it
8 here but make sure to check the current code if you need a deeper
12 The Memory Resource Controller has generically been referred to as the
13 memory controller in this document. Do not confuse memory controller
14 used here with the memory controller that is used in hardware.
16 (For editors) In this document:
17 When we mention a cgroup (cgroupfs's directory) with memory controller,
18 we call it "memory cgroup". When you see git-log and source code, you'll
19 see patch's title and function names tend to use "memcg".
20 In this document, we avoid using it.
22 Benefits and Purpose of the memory controller
23 =============================================
25 The memory controller isolates the memory behaviour of a group of tasks
26 from the rest of the system. The article on LWN [12] mentions some probable
27 uses of the memory controller. The memory controller can be used to
29 a. Isolate an application or a group of applications
30 Memory-hungry applications can be isolated and limited to a smaller
32 b. Create a cgroup with a limited amount of memory; this can be used
33 as a good alternative to booting with mem=XXXX.
34 c. Virtualization solutions can control the amount of memory they want
35 to assign to a virtual machine instance.
36 d. A CD/DVD burner could control the amount of memory used by the
37 rest of the system to ensure that burning does not fail due to lack
39 e. There are several other use cases; find one or use the controller just
40 for fun (to learn and hack on the VM subsystem).
42 Current Status: linux-2.6.34-mmotm(development version of 2010/April)
46 - accounting anonymous pages, file caches, swap caches usage and limiting them.
47 - pages are linked to per-memcg LRU exclusively, and there is no global LRU.
48 - optionally, memory+swap usage can be accounted and limited.
49 - hierarchical accounting
51 - moving (recharging) account at moving a task is selectable.
52 - usage threshold notifier
53 - memory pressure notifier
54 - oom-killer disable knob and oom-notifier
55 - Root cgroup has no limit controls.
57 Kernel memory support is a work in progress, and the current version provides
58 basically functionality. (See Section 2.7)
60 Brief summary of control files.
62 ==================================== ==========================================
63 tasks attach a task(thread) and show list of
65 cgroup.procs show list of processes
66 cgroup.event_control an interface for event_fd()
67 memory.usage_in_bytes show current usage for memory
69 memory.memsw.usage_in_bytes show current usage for memory+Swap
71 memory.limit_in_bytes set/show limit of memory usage
72 memory.memsw.limit_in_bytes set/show limit of memory+Swap usage
73 memory.failcnt show the number of memory usage hits limits
74 memory.memsw.failcnt show the number of memory+Swap hits limits
75 memory.max_usage_in_bytes show max memory usage recorded
76 memory.memsw.max_usage_in_bytes show max memory+Swap usage recorded
77 memory.soft_limit_in_bytes set/show soft limit of memory usage
78 memory.stat show various statistics
79 memory.use_hierarchy set/show hierarchical account enabled
80 memory.force_empty trigger forced page reclaim
81 memory.pressure_level set memory pressure notifications
82 memory.swappiness set/show swappiness parameter of vmscan
83 (See sysctl's vm.swappiness)
84 memory.move_charge_at_immigrate set/show controls of moving charges
85 This knob is deprecated and shouldn't be
87 memory.oom_control set/show oom controls.
88 memory.numa_stat show the number of memory usage per numa
90 memory.kmem.limit_in_bytes set/show hard limit for kernel memory
91 This knob is deprecated and shouldn't be
92 used. It is planned that this be removed in
93 the foreseeable future.
94 memory.kmem.usage_in_bytes show current kernel memory allocation
95 memory.kmem.failcnt show the number of kernel memory usage
97 memory.kmem.max_usage_in_bytes show max kernel memory usage recorded
99 memory.kmem.tcp.limit_in_bytes set/show hard limit for tcp buf memory
100 memory.kmem.tcp.usage_in_bytes show current tcp buf memory allocation
101 memory.kmem.tcp.failcnt show the number of tcp buf memory usage
103 memory.kmem.tcp.max_usage_in_bytes show max tcp buf memory usage recorded
104 ==================================== ==========================================
109 The memory controller has a long history. A request for comments for the memory
110 controller was posted by Balbir Singh [1]. At the time the RFC was posted
111 there were several implementations for memory control. The goal of the
112 RFC was to build consensus and agreement for the minimal features required
113 for memory control. The first RSS controller was posted by Balbir Singh[2]
114 in Feb 2007. Pavel Emelianov [3][4][5] has since posted three versions of the
115 RSS controller. At OLS, at the resource management BoF, everyone suggested
116 that we handle both page cache and RSS together. Another request was raised
117 to allow user space handling of OOM. The current memory controller is
118 at version 6; it combines both mapped (RSS) and unmapped Page
124 Memory is a unique resource in the sense that it is present in a limited
125 amount. If a task requires a lot of CPU processing, the task can spread
126 its processing over a period of hours, days, months or years, but with
127 memory, the same physical memory needs to be reused to accomplish the task.
129 The memory controller implementation has been divided into phases. These
133 2. mlock(2) controller
134 3. Kernel user memory accounting and slab control
135 4. user mappings length controller
137 The memory controller is the first controller developed.
142 The core of the design is a counter called the page_counter. The
143 page_counter tracks the current memory usage and limit of the group of
144 processes associated with the controller. Each cgroup has a memory controller
145 specific data structure (mem_cgroup) associated with it.
152 +--------------------+
155 +--------------------+
158 +---------------+ | +---------------+
159 | mm_struct | |.... | mm_struct |
161 +---------------+ | +---------------+
165 +---------------+ +------+--------+
166 | page +----------> page_cgroup|
168 +---------------+ +---------------+
170 (Figure 1: Hierarchy of Accounting)
173 Figure 1 shows the important aspects of the controller
175 1. Accounting happens per cgroup
176 2. Each mm_struct knows about which cgroup it belongs to
177 3. Each page has a pointer to the page_cgroup, which in turn knows the
180 The accounting is done as follows: mem_cgroup_charge_common() is invoked to
181 set up the necessary data structures and check if the cgroup that is being
182 charged is over its limit. If it is, then reclaim is invoked on the cgroup.
183 More details can be found in the reclaim section of this document.
184 If everything goes well, a page meta-data-structure called page_cgroup is
185 updated. page_cgroup has its own LRU on cgroup.
186 (*) page_cgroup structure is allocated at boot/memory-hotplug time.
188 2.2.1 Accounting details
189 ------------------------
191 All mapped anon pages (RSS) and cache pages (Page Cache) are accounted.
192 Some pages which are never reclaimable and will not be on the LRU
193 are not accounted. We just account pages under usual VM management.
195 RSS pages are accounted at page_fault unless they've already been accounted
196 for earlier. A file page will be accounted for as Page Cache when it's
197 inserted into inode (radix-tree). While it's mapped into the page tables of
198 processes, duplicate accounting is carefully avoided.
200 An RSS page is unaccounted when it's fully unmapped. A PageCache page is
201 unaccounted when it's removed from radix-tree. Even if RSS pages are fully
202 unmapped (by kswapd), they may exist as SwapCache in the system until they
203 are really freed. Such SwapCaches are also accounted.
204 A swapped-in page is accounted after adding into swapcache.
206 Note: The kernel does swapin-readahead and reads multiple swaps at once.
207 Since page's memcg recorded into swap whatever memsw enabled, the page will
208 be accounted after swapin.
210 At page migration, accounting information is kept.
212 Note: we just account pages-on-LRU because our purpose is to control amount
213 of used pages; not-on-LRU pages tend to be out-of-control from VM view.
215 2.3 Shared Page Accounting
216 --------------------------
218 Shared pages are accounted on the basis of the first touch approach. The
219 cgroup that first touches a page is accounted for the page. The principle
220 behind this approach is that a cgroup that aggressively uses a shared
221 page will eventually get charged for it (once it is uncharged from
222 the cgroup that brought it in -- this will happen on memory pressure).
224 But see section 8.2: when moving a task to another cgroup, its pages may
225 be recharged to the new cgroup, if move_charge_at_immigrate has been chosen.
228 --------------------------------------
230 Swap usage is always recorded for each of cgroup. Swap Extension allows you to
233 When CONFIG_SWAP is enabled, following files are added.
235 - memory.memsw.usage_in_bytes.
236 - memory.memsw.limit_in_bytes.
238 memsw means memory+swap. Usage of memory+swap is limited by
239 memsw.limit_in_bytes.
241 Example: Assume a system with 4G of swap. A task which allocates 6G of memory
242 (by mistake) under 2G memory limitation will use all swap.
243 In this case, setting memsw.limit_in_bytes=3G will prevent bad use of swap.
244 By using the memsw limit, you can avoid system OOM which can be caused by swap
247 **why 'memory+swap' rather than swap**
249 The global LRU(kswapd) can swap out arbitrary pages. Swap-out means
250 to move account from memory to swap...there is no change in usage of
251 memory+swap. In other words, when we want to limit the usage of swap without
252 affecting global LRU, memory+swap limit is better than just limiting swap from
255 **What happens when a cgroup hits memory.memsw.limit_in_bytes**
257 When a cgroup hits memory.memsw.limit_in_bytes, it's useless to do swap-out
258 in this cgroup. Then, swap-out will not be done by cgroup routine and file
259 caches are dropped. But as mentioned above, global LRU can do swapout memory
260 from it for sanity of the system's memory management state. You can't forbid
266 Each cgroup maintains a per cgroup LRU which has the same structure as
267 global VM. When a cgroup goes over its limit, we first try
268 to reclaim memory from the cgroup so as to make space for the new
269 pages that the cgroup has touched. If the reclaim is unsuccessful,
270 an OOM routine is invoked to select and kill the bulkiest task in the
271 cgroup. (See 10. OOM Control below.)
273 The reclaim algorithm has not been modified for cgroups, except that
274 pages that are selected for reclaiming come from the per-cgroup LRU
278 Reclaim does not work for the root cgroup, since we cannot set any
279 limits on the root cgroup.
282 When panic_on_oom is set to "2", the whole system will panic.
284 When oom event notifier is registered, event will be delivered.
285 (See oom_control section)
290 lock_page_cgroup()/unlock_page_cgroup() should not be called under
293 Other lock order is following:
300 In many cases, just lock_page_cgroup() is called.
302 per-zone-per-cgroup LRU (cgroup's private LRU) is just guarded by
303 pgdat->lru_lock, it has no lock of its own.
305 2.7 Kernel Memory Extension (CONFIG_MEMCG_KMEM)
306 -----------------------------------------------
308 With the Kernel memory extension, the Memory Controller is able to limit
309 the amount of kernel memory used by the system. Kernel memory is fundamentally
310 different than user memory, since it can't be swapped out, which makes it
311 possible to DoS the system by consuming too much of this precious resource.
313 Kernel memory accounting is enabled for all memory cgroups by default. But
314 it can be disabled system-wide by passing cgroup.memory=nokmem to the kernel
315 at boot time. In this case, kernel memory will not be accounted at all.
317 Kernel memory limits are not imposed for the root cgroup. Usage for the root
318 cgroup may or may not be accounted. The memory used is accumulated into
319 memory.kmem.usage_in_bytes, or in a separate counter when it makes sense.
320 (currently only for tcp).
322 The main "kmem" counter is fed into the main counter, so kmem charges will
323 also be visible from the user counter.
325 Currently no soft limit is implemented for kernel memory. It is future work
326 to trigger slab reclaim when those limits are reached.
328 2.7.1 Current Kernel Memory resources accounted
329 -----------------------------------------------
332 every process consumes some stack pages. By accounting into
333 kernel memory, we prevent new processes from being created when the kernel
334 memory usage is too high.
337 pages allocated by the SLAB or SLUB allocator are tracked. A copy
338 of each kmem_cache is created every time the cache is touched by the first time
339 from inside the memcg. The creation is done lazily, so some objects can still be
340 skipped while the cache is being created. All objects in a slab page should
341 belong to the same memcg. This only fails to hold when a task is migrated to a
342 different memcg during the page allocation by the cache.
344 sockets memory pressure:
345 some sockets protocols have memory pressure
346 thresholds. The Memory Controller allows them to be controlled individually
347 per cgroup, instead of globally.
350 sockets memory pressure for the tcp protocol.
352 2.7.2 Common use cases
353 ----------------------
355 Because the "kmem" counter is fed to the main user counter, kernel memory can
356 never be limited completely independently of user memory. Say "U" is the user
357 limit, and "K" the kernel limit. There are three possible ways limits can be
360 U != 0, K = unlimited:
361 This is the standard memcg limitation mechanism already present before kmem
362 accounting. Kernel memory is completely ignored.
365 Kernel memory is a subset of the user memory. This setup is useful in
366 deployments where the total amount of memory per-cgroup is overcommited.
367 Overcommiting kernel memory limits is definitely not recommended, since the
368 box can still run out of non-reclaimable memory.
369 In this case, the admin could set up K so that the sum of all groups is
370 never greater than the total memory, and freely set U at the cost of his
374 In the current implementation, memory reclaim will NOT be
375 triggered for a cgroup when it hits K while staying below U, which makes
376 this setup impractical.
379 Since kmem charges will also be fed to the user counter and reclaim will be
380 triggered for the cgroup for both kinds of memory. This setup gives the
381 admin a unified view of memory, and it is also useful for people who just
382 want to track kernel memory usage.
390 a. Enable CONFIG_CGROUPS
391 b. Enable CONFIG_MEMCG
392 c. Enable CONFIG_MEMCG_SWAP (to use swap extension)
393 d. Enable CONFIG_MEMCG_KMEM (to use kmem extension)
395 3.1. Prepare the cgroups (see cgroups.txt, Why are cgroups needed?)
396 -------------------------------------------------------------------
400 # mount -t tmpfs none /sys/fs/cgroup
401 # mkdir /sys/fs/cgroup/memory
402 # mount -t cgroup none /sys/fs/cgroup/memory -o memory
404 3.2. Make the new group and move bash into it::
406 # mkdir /sys/fs/cgroup/memory/0
407 # echo $$ > /sys/fs/cgroup/memory/0/tasks
409 Since now we're in the 0 cgroup, we can alter the memory limit::
411 # echo 4M > /sys/fs/cgroup/memory/0/memory.limit_in_bytes
414 We can use a suffix (k, K, m, M, g or G) to indicate values in kilo,
415 mega or gigabytes. (Here, Kilo, Mega, Giga are Kibibytes, Mebibytes,
419 We can write "-1" to reset the ``*.limit_in_bytes(unlimited)``.
422 We cannot set limits on the root cgroup any more.
426 # cat /sys/fs/cgroup/memory/0/memory.limit_in_bytes
429 We can check the usage::
431 # cat /sys/fs/cgroup/memory/0/memory.usage_in_bytes
434 A successful write to this file does not guarantee a successful setting of
435 this limit to the value written into the file. This can be due to a
436 number of factors, such as rounding up to page boundaries or the total
437 availability of memory on the system. The user is required to re-read
438 this file after a write to guarantee the value committed by the kernel::
440 # echo 1 > memory.limit_in_bytes
441 # cat memory.limit_in_bytes
444 The memory.failcnt field gives the number of times that the cgroup limit was
447 The memory.stat file gives accounting information. Now, the number of
448 caches, RSS and Active pages/Inactive pages are shown.
453 For testing features and implementation, see memcg_test.txt.
455 Performance test is also important. To see pure memory controller's overhead,
456 testing on tmpfs will give you good numbers of small overheads.
457 Example: do kernel make on tmpfs.
459 Page-fault scalability is also important. At measuring parallel
460 page fault test, multi-process test may be better than multi-thread
461 test because it has noise of shared objects/status.
463 But the above two are testing extreme situations.
464 Trying usual test under memory controller is always helpful.
469 Sometimes a user might find that the application under a cgroup is
470 terminated by the OOM killer. There are several causes for this:
472 1. The cgroup limit is too low (just too low to do anything useful)
473 2. The user is using anonymous memory and swap is turned off or too low
475 A sync followed by echo 1 > /proc/sys/vm/drop_caches will help get rid of
476 some of the pages cached in the cgroup (page cache pages).
478 To know what happens, disabling OOM_Kill as per "10. OOM Control" (below) and
479 seeing what happens will be helpful.
484 When a task migrates from one cgroup to another, its charge is not
485 carried forward by default. The pages allocated from the original cgroup still
486 remain charged to it, the charge is dropped when the page is freed or
489 You can move charges of a task along with task migration.
490 See 8. "Move charges at task migration"
492 4.3 Removing a cgroup
493 ---------------------
495 A cgroup can be removed by rmdir, but as discussed in sections 4.1 and 4.2, a
496 cgroup might have some charge associated with it, even though all
497 tasks have migrated away from it. (because we charge against pages, not
500 We move the stats to root (if use_hierarchy==0) or parent (if
501 use_hierarchy==1), and no change on the charge except uncharging
504 Charges recorded in swap information is not updated at removal of cgroup.
505 Recorded information is discarded and a cgroup which uses swap (swapcache)
506 will be charged as a new owner of it.
508 About use_hierarchy, see Section 6.
515 memory.force_empty interface is provided to make cgroup's memory usage empty.
516 When writing anything to this::
518 # echo 0 > memory.force_empty
520 the cgroup will be reclaimed and as many pages reclaimed as possible.
522 The typical use case for this interface is before calling rmdir().
523 Though rmdir() offlines memcg, but the memcg may still stay there due to
524 charged file caches. Some out-of-use page caches may keep charged until
525 memory pressure happens. If you want to avoid that, force_empty will be useful.
527 Also, note that when memory.kmem.limit_in_bytes is set the charges due to
528 kernel pages will still be seen. This is not considered a failure and the
529 write will still return success. In this case, it is expected that
530 memory.kmem.usage_in_bytes == memory.usage_in_bytes.
532 About use_hierarchy, see Section 6.
537 memory.stat file includes following statistics
539 per-memory cgroup local status
540 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
542 =============== ===============================================================
543 cache # of bytes of page cache memory.
544 rss # of bytes of anonymous and swap cache memory (includes
545 transparent hugepages).
546 rss_huge # of bytes of anonymous transparent hugepages.
547 mapped_file # of bytes of mapped file (includes tmpfs/shmem)
548 pgpgin # of charging events to the memory cgroup. The charging
549 event happens each time a page is accounted as either mapped
550 anon page(RSS) or cache page(Page Cache) to the cgroup.
551 pgpgout # of uncharging events to the memory cgroup. The uncharging
552 event happens each time a page is unaccounted from the cgroup.
553 swap # of bytes of swap usage
554 dirty # of bytes that are waiting to get written back to the disk.
555 writeback # of bytes of file/anon cache that are queued for syncing to
557 inactive_anon # of bytes of anonymous and swap cache memory on inactive
559 active_anon # of bytes of anonymous and swap cache memory on active
561 inactive_file # of bytes of file-backed memory on inactive LRU list.
562 active_file # of bytes of file-backed memory on active LRU list.
563 unevictable # of bytes of memory that cannot be reclaimed (mlocked etc).
564 =============== ===============================================================
566 status considering hierarchy (see memory.use_hierarchy settings)
567 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
569 ========================= ===================================================
570 hierarchical_memory_limit # of bytes of memory limit with regard to hierarchy
571 under which the memory cgroup is
572 hierarchical_memsw_limit # of bytes of memory+swap limit with regard to
573 hierarchy under which memory cgroup is.
575 total_<counter> # hierarchical version of <counter>, which in
576 addition to the cgroup's own value includes the
577 sum of all hierarchical children's values of
578 <counter>, i.e. total_cache
579 ========================= ===================================================
581 The following additional stats are dependent on CONFIG_DEBUG_VM
582 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
584 ========================= ========================================
585 recent_rotated_anon VM internal parameter. (see mm/vmscan.c)
586 recent_rotated_file VM internal parameter. (see mm/vmscan.c)
587 recent_scanned_anon VM internal parameter. (see mm/vmscan.c)
588 recent_scanned_file VM internal parameter. (see mm/vmscan.c)
589 ========================= ========================================
592 recent_rotated means recent frequency of LRU rotation.
593 recent_scanned means recent # of scans to LRU.
594 showing for better debug please see the code for meanings.
597 Only anonymous and swap cache memory is listed as part of 'rss' stat.
598 This should not be confused with the true 'resident set size' or the
599 amount of physical memory used by the cgroup.
601 'rss + mapped_file" will give you resident set size of cgroup.
603 (Note: file and shmem may be shared among other cgroups. In that case,
604 mapped_file is accounted only when the memory cgroup is owner of page
610 Overrides /proc/sys/vm/swappiness for the particular group. The tunable
611 in the root cgroup corresponds to the global swappiness setting.
613 Please note that unlike during the global reclaim, limit reclaim
614 enforces that 0 swappiness really prevents from any swapping even if
615 there is a swap storage available. This might lead to memcg OOM killer
616 if there are no file pages to reclaim.
621 A memory cgroup provides memory.failcnt and memory.memsw.failcnt files.
622 This failcnt(== failure count) shows the number of times that a usage counter
623 hit its limit. When a memory cgroup hits a limit, failcnt increases and
624 memory under it will be reclaimed.
626 You can reset failcnt by writing 0 to failcnt file::
628 # echo 0 > .../memory.failcnt
633 For efficiency, as other kernel components, memory cgroup uses some optimization
634 to avoid unnecessary cacheline false sharing. usage_in_bytes is affected by the
635 method and doesn't show 'exact' value of memory (and swap) usage, it's a fuzz
636 value for efficient access. (Of course, when necessary, it's synchronized.)
637 If you want to know more exact memory usage, you should use RSS+CACHE(+SWAP)
638 value in memory.stat(see 5.2).
643 This is similar to numa_maps but operates on a per-memcg basis. This is
644 useful for providing visibility into the numa locality information within
645 an memcg since the pages are allowed to be allocated from any physical
646 node. One of the use cases is evaluating application performance by
647 combining this information with the application's CPU allocation.
649 Each memcg's numa_stat file includes "total", "file", "anon" and "unevictable"
650 per-node page counts including "hierarchical_<counter>" which sums up all
651 hierarchical children's values in addition to the memcg's own value.
653 The output format of memory.numa_stat is::
655 total=<total pages> N0=<node 0 pages> N1=<node 1 pages> ...
656 file=<total file pages> N0=<node 0 pages> N1=<node 1 pages> ...
657 anon=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ...
658 unevictable=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ...
659 hierarchical_<counter>=<counter pages> N0=<node 0 pages> N1=<node 1 pages> ...
661 The "total" count is sum of file + anon + unevictable.
666 The memory controller supports a deep hierarchy and hierarchical accounting.
667 The hierarchy is created by creating the appropriate cgroups in the
668 cgroup filesystem. Consider for example, the following cgroup filesystem
679 In the diagram above, with hierarchical accounting enabled, all memory
680 usage of e, is accounted to its ancestors up until the root (i.e, c and root),
681 that has memory.use_hierarchy enabled. If one of the ancestors goes over its
682 limit, the reclaim algorithm reclaims from the tasks in the ancestor and the
683 children of the ancestor.
685 6.1 Enabling hierarchical accounting and reclaim
686 ------------------------------------------------
688 A memory cgroup by default disables the hierarchy feature. Support
689 can be enabled by writing 1 to memory.use_hierarchy file of the root cgroup::
691 # echo 1 > memory.use_hierarchy
693 The feature can be disabled by::
695 # echo 0 > memory.use_hierarchy
698 Enabling/disabling will fail if either the cgroup already has other
699 cgroups created below it, or if the parent cgroup has use_hierarchy
703 When panic_on_oom is set to "2", the whole system will panic in
704 case of an OOM event in any cgroup.
709 Soft limits allow for greater sharing of memory. The idea behind soft limits
710 is to allow control groups to use as much of the memory as needed, provided
712 a. There is no memory contention
713 b. They do not exceed their hard limit
715 When the system detects memory contention or low memory, control groups
716 are pushed back to their soft limits. If the soft limit of each control
717 group is very high, they are pushed back as much as possible to make
718 sure that one control group does not starve the others of memory.
720 Please note that soft limits is a best-effort feature; it comes with
721 no guarantees, but it does its best to make sure that when memory is
722 heavily contended for, memory is allocated based on the soft limit
723 hints/setup. Currently soft limit based reclaim is set up such that
724 it gets invoked from balance_pgdat (kswapd).
729 Soft limits can be setup by using the following commands (in this example we
730 assume a soft limit of 256 MiB)::
732 # echo 256M > memory.soft_limit_in_bytes
734 If we want to change this to 1G, we can at any time use::
736 # echo 1G > memory.soft_limit_in_bytes
739 Soft limits take effect over a long period of time, since they involve
740 reclaiming memory for balancing between memory cgroups
742 It is recommended to set the soft limit always below the hard limit,
743 otherwise the hard limit will take precedence.
745 8. Move charges at task migration (DEPRECATED!)
746 ===============================================
750 It's expensive and unreliable! It's better practice to launch workload
751 tasks directly from inside their target cgroup. Use dedicated workload
752 cgroups to allow fine-grained policy adjustments without having to
753 move physical pages between control domains.
755 Users can move charges associated with a task along with task migration, that
756 is, uncharge task's pages from the old cgroup and charge them to the new cgroup.
757 This feature is not supported in !CONFIG_MMU environments because of lack of
763 This feature is disabled by default. It can be enabled (and disabled again) by
764 writing to memory.move_charge_at_immigrate of the destination cgroup.
766 If you want to enable it::
768 # echo (some positive value) > memory.move_charge_at_immigrate
771 Each bits of move_charge_at_immigrate has its own meaning about what type
772 of charges should be moved. See 8.2 for details.
774 Charges are moved only when you move mm->owner, in other words,
775 a leader of a thread group.
777 If we cannot find enough space for the task in the destination cgroup, we
778 try to make space by reclaiming memory. Task migration may fail if we
779 cannot make enough space.
781 It can take several seconds if you move charges much.
783 And if you want disable it again::
785 # echo 0 > memory.move_charge_at_immigrate
787 8.2 Type of charges which can be moved
788 --------------------------------------
790 Each bit in move_charge_at_immigrate has its own meaning about what type of
791 charges should be moved. But in any case, it must be noted that an account of
792 a page or a swap can be moved only when it is charged to the task's current
795 +---+--------------------------------------------------------------------------+
796 |bit| what type of charges would be moved ? |
797 +===+==========================================================================+
798 | 0 | A charge of an anonymous page (or swap of it) used by the target task. |
799 | | You must enable Swap Extension (see 2.4) to enable move of swap charges. |
800 +---+--------------------------------------------------------------------------+
801 | 1 | A charge of file pages (normal file, tmpfs file (e.g. ipc shared memory) |
802 | | and swaps of tmpfs file) mmapped by the target task. Unlike the case of |
803 | | anonymous pages, file pages (and swaps) in the range mmapped by the task |
804 | | will be moved even if the task hasn't done page fault, i.e. they might |
805 | | not be the task's "RSS", but other task's "RSS" that maps the same file. |
806 | | And mapcount of the page is ignored (the page can be moved even if |
807 | | page_mapcount(page) > 1). You must enable Swap Extension (see 2.4) to |
808 | | enable move of swap charges. |
809 +---+--------------------------------------------------------------------------+
814 - All of moving charge operations are done under cgroup_mutex. It's not good
815 behavior to hold the mutex too long, so we may need some trick.
820 Memory cgroup implements memory thresholds using the cgroups notification
821 API (see cgroups.txt). It allows to register multiple memory and memsw
822 thresholds and gets notifications when it crosses.
824 To register a threshold, an application must:
826 - create an eventfd using eventfd(2);
827 - open memory.usage_in_bytes or memory.memsw.usage_in_bytes;
828 - write string like "<event_fd> <fd of memory.usage_in_bytes> <threshold>" to
829 cgroup.event_control.
831 Application will be notified through eventfd when memory usage crosses
832 threshold in any direction.
834 It's applicable for root and non-root cgroup.
839 memory.oom_control file is for OOM notification and other controls.
841 Memory cgroup implements OOM notifier using the cgroup notification
842 API (See cgroups.txt). It allows to register multiple OOM notification
843 delivery and gets notification when OOM happens.
845 To register a notifier, an application must:
847 - create an eventfd using eventfd(2)
848 - open memory.oom_control file
849 - write string like "<event_fd> <fd of memory.oom_control>" to
852 The application will be notified through eventfd when OOM happens.
853 OOM notification doesn't work for the root cgroup.
855 You can disable the OOM-killer by writing "1" to memory.oom_control file, as:
857 #echo 1 > memory.oom_control
859 If OOM-killer is disabled, tasks under cgroup will hang/sleep
860 in memory cgroup's OOM-waitqueue when they request accountable memory.
862 For running them, you have to relax the memory cgroup's OOM status by
864 * enlarge limit or reduce usage.
869 * move some tasks to other group with account migration.
870 * remove some files (on tmpfs?)
872 Then, stopped tasks will work again.
874 At reading, current status of OOM is shown.
876 - oom_kill_disable 0 or 1
877 (if 1, oom-killer is disabled)
879 (if 1, the memory cgroup is under OOM, tasks may be stopped.)
884 The pressure level notifications can be used to monitor the memory
885 allocation cost; based on the pressure, applications can implement
886 different strategies of managing their memory resources. The pressure
887 levels are defined as following:
889 The "low" level means that the system is reclaiming memory for new
890 allocations. Monitoring this reclaiming activity might be useful for
891 maintaining cache level. Upon notification, the program (typically
892 "Activity Manager") might analyze vmstat and act in advance (i.e.
893 prematurely shutdown unimportant services).
895 The "medium" level means that the system is experiencing medium memory
896 pressure, the system might be making swap, paging out active file caches,
897 etc. Upon this event applications may decide to further analyze
898 vmstat/zoneinfo/memcg or internal memory usage statistics and free any
899 resources that can be easily reconstructed or re-read from a disk.
901 The "critical" level means that the system is actively thrashing, it is
902 about to out of memory (OOM) or even the in-kernel OOM killer is on its
903 way to trigger. Applications should do whatever they can to help the
904 system. It might be too late to consult with vmstat or any other
905 statistics, so it's advisable to take an immediate action.
907 By default, events are propagated upward until the event is handled, i.e. the
908 events are not pass-through. For example, you have three cgroups: A->B->C. Now
909 you set up an event listener on cgroups A, B and C, and suppose group C
910 experiences some pressure. In this situation, only group C will receive the
911 notification, i.e. groups A and B will not receive it. This is done to avoid
912 excessive "broadcasting" of messages, which disturbs the system and which is
913 especially bad if we are low on memory or thrashing. Group B, will receive
914 notification only if there are no event listers for group C.
916 There are three optional modes that specify different propagation behavior:
918 - "default": this is the default behavior specified above. This mode is the
919 same as omitting the optional mode parameter, preserved by backwards
922 - "hierarchy": events always propagate up to the root, similar to the default
923 behavior, except that propagation continues regardless of whether there are
924 event listeners at each level, with the "hierarchy" mode. In the above
925 example, groups A, B, and C will receive notification of memory pressure.
927 - "local": events are pass-through, i.e. they only receive notifications when
928 memory pressure is experienced in the memcg for which the notification is
929 registered. In the above example, group C will receive notification if
930 registered for "local" notification and the group experiences memory
931 pressure. However, group B will never receive notification, regardless if
932 there is an event listener for group C or not, if group B is registered for
935 The level and event notification mode ("hierarchy" or "local", if necessary) are
936 specified by a comma-delimited string, i.e. "low,hierarchy" specifies
937 hierarchical, pass-through, notification for all ancestor memcgs. Notification
938 that is the default, non pass-through behavior, does not specify a mode.
939 "medium,local" specifies pass-through notification for the medium level.
941 The file memory.pressure_level is only used to setup an eventfd. To
942 register a notification, an application must:
944 - create an eventfd using eventfd(2);
945 - open memory.pressure_level;
946 - write string as "<event_fd> <fd of memory.pressure_level> <level[,mode]>"
947 to cgroup.event_control.
949 Application will be notified through eventfd when memory pressure is at
950 the specific level (or higher). Read/write operations to
951 memory.pressure_level are no implemented.
955 Here is a small script example that makes a new cgroup, sets up a
956 memory limit, sets up a notification in the cgroup and then makes child
957 cgroup experience a critical pressure::
959 # cd /sys/fs/cgroup/memory/
962 # cgroup_event_listener memory.pressure_level low,hierarchy &
963 # echo 8000000 > memory.limit_in_bytes
964 # echo 8000000 > memory.memsw.limit_in_bytes
966 # dd if=/dev/zero | read x
968 (Expect a bunch of notifications, and eventually, the oom-killer will
974 1. Make per-cgroup scanner reclaim not-shared pages first
975 2. Teach controller to account for shared-pages
976 3. Start reclamation in the background when the limit is
977 not yet hit but the usage is getting closer
982 Overall, the memory controller has been a stable controller and has been
983 commented and discussed quite extensively in the community.
988 1. Singh, Balbir. RFC: Memory Controller, http://lwn.net/Articles/206697/
989 2. Singh, Balbir. Memory Controller (RSS Control),
990 http://lwn.net/Articles/222762/
991 3. Emelianov, Pavel. Resource controllers based on process cgroups
992 http://lkml.org/lkml/2007/3/6/198
993 4. Emelianov, Pavel. RSS controller based on process cgroups (v2)
994 http://lkml.org/lkml/2007/4/9/78
995 5. Emelianov, Pavel. RSS controller based on process cgroups (v3)
996 http://lkml.org/lkml/2007/5/30/244
997 6. Menage, Paul. Control Groups v10, http://lwn.net/Articles/236032/
998 7. Vaidyanathan, Srinivasan, Control Groups: Pagecache accounting and control
999 subsystem (v3), http://lwn.net/Articles/235534/
1000 8. Singh, Balbir. RSS controller v2 test results (lmbench),
1001 http://lkml.org/lkml/2007/5/17/232
1002 9. Singh, Balbir. RSS controller v2 AIM9 results
1003 http://lkml.org/lkml/2007/5/18/1
1004 10. Singh, Balbir. Memory controller v6 test results,
1005 http://lkml.org/lkml/2007/8/19/36
1006 11. Singh, Balbir. Memory controller introduction (v6),
1007 http://lkml.org/lkml/2007/8/17/69
1008 12. Corbet, Jonathan, Controlling memory use in cgroups,
1009 http://lwn.net/Articles/243795/