1Memory Resource Controller
   3NOTE: The Memory Resource Controller has generically been referred to as the
   4      memory controller in this document. Do not confuse memory controller
   5      used here with the memory controller that is used in hardware.
   7(For editors)
   8In this document:
   9      When we mention a cgroup (cgroupfs's directory) with memory controller,
  10      we call it "memory cgroup". When you see git-log and source code, you'll
  11      see patch's title and function names tend to use "memcg".
  12      In this document, we avoid using it.
  14Benefits and Purpose of the memory controller
  16The memory controller isolates the memory behaviour of a group of tasks
  17from the rest of the system. The article on LWN [12] mentions some probable
  18uses of the memory controller. The memory controller can be used to
  20a. Isolate an application or a group of applications
  21   Memory-hungry applications can be isolated and limited to a smaller
  22   amount of memory.
  23b. Create a cgroup with a limited amount of memory; this can be used
  24   as a good alternative to booting with mem=XXXX.
  25c. Virtualization solutions can control the amount of memory they want
  26   to assign to a virtual machine instance.
  27d. A CD/DVD burner could control the amount of memory used by the
  28   rest of the system to ensure that burning does not fail due to lack
  29   of available memory.
  30e. There are several other use cases; find one or use the controller just
  31   for fun (to learn and hack on the VM subsystem).
  33Current Status: linux-2.6.34-mmotm(development version of 2010/April)
  36 - accounting anonymous pages, file caches, swap caches usage and limiting them.
  37 - pages are linked to per-memcg LRU exclusively, and there is no global LRU.
  38 - optionally, memory+swap usage can be accounted and limited.
  39 - hierarchical accounting
  40 - soft limit
  41 - moving (recharging) account at moving a task is selectable.
  42 - usage threshold notifier
  43 - memory pressure notifier
  44 - oom-killer disable knob and oom-notifier
  45 - Root cgroup has no limit controls.
  47 Kernel memory support is a work in progress, and the current version provides
  48 basically functionality. (See Section 2.7)
  50Brief summary of control files.
  52 tasks                           # attach a task(thread) and show list of threads
  53 cgroup.procs                    # show list of processes
  54 cgroup.event_control            # an interface for event_fd()
  55 memory.usage_in_bytes           # show current res_counter usage for memory
  56                                 (See 5.5 for details)
  57 memory.memsw.usage_in_bytes     # show current res_counter usage for memory+Swap
  58                                 (See 5.5 for details)
  59 memory.limit_in_bytes           # set/show limit of memory usage
  60 memory.memsw.limit_in_bytes     # set/show limit of memory+Swap usage
  61 memory.failcnt                  # show the number of memory usage hits limits
  62 memory.memsw.failcnt            # show the number of memory+Swap hits limits
  63 memory.max_usage_in_bytes       # show max memory usage recorded
  64 memory.memsw.max_usage_in_bytes # show max memory+Swap usage recorded
  65 memory.soft_limit_in_bytes      # set/show soft limit of memory usage
  66 memory.stat                     # show various statistics
  67 memory.use_hierarchy            # set/show hierarchical account enabled
  68 memory.force_empty              # trigger forced move charge to parent
  69 memory.pressure_level           # set memory pressure notifications
  70 memory.swappiness               # set/show swappiness parameter of vmscan
  71                                 (See sysctl's vm.swappiness)
  72 memory.move_charge_at_immigrate # set/show controls of moving charges
  73 memory.oom_control              # set/show oom controls.
  74 memory.numa_stat                # show the number of memory usage per numa node
  76 memory.kmem.limit_in_bytes      # set/show hard limit for kernel memory
  77 memory.kmem.usage_in_bytes      # show current kernel memory allocation
  78 memory.kmem.failcnt             # show the number of kernel memory usage hits limits
  79 memory.kmem.max_usage_in_bytes  # show max kernel memory usage recorded
  81 memory.kmem.tcp.limit_in_bytes  # set/show hard limit for tcp buf memory
  82 memory.kmem.tcp.usage_in_bytes  # show current tcp buf memory allocation
  83 memory.kmem.tcp.failcnt            # show the number of tcp buf memory usage hits limits
  84 memory.kmem.tcp.max_usage_in_bytes # show max tcp buf memory usage recorded
  861. History
  88The memory controller has a long history. A request for comments for the memory
  89controller was posted by Balbir Singh [1]. At the time the RFC was posted
  90there were several implementations for memory control. The goal of the
  91RFC was to build consensus and agreement for the minimal features required
  92for memory control. The first RSS controller was posted by Balbir Singh[2]
  93in Feb 2007. Pavel Emelianov [3][4][5] has since posted three versions of the
  94RSS controller. At OLS, at the resource management BoF, everyone suggested
  95that we handle both page cache and RSS together. Another request was raised
  96to allow user space handling of OOM. The current memory controller is
  97at version 6; it combines both mapped (RSS) and unmapped Page
  98Cache Control [11].
 1002. Memory Control
 102Memory is a unique resource in the sense that it is present in a limited
 103amount. If a task requires a lot of CPU processing, the task can spread
 104its processing over a period of hours, days, months or years, but with
 105memory, the same physical memory needs to be reused to accomplish the task.
 107The memory controller implementation has been divided into phases. These
 1101. Memory controller
 1112. mlock(2) controller
 1123. Kernel user memory accounting and slab control
 1134. user mappings length controller
 115The memory controller is the first controller developed.
 1172.1. Design
 119The core of the design is a counter called the res_counter. The res_counter
 120tracks the current memory usage and limit of the group of processes associated
 121with the controller. Each cgroup has a memory controller specific data
 122structure (mem_cgroup) associated with it.
 1242.2. Accounting
 126                +--------------------+
 127                |  mem_cgroup     |
 128                |  (res_counter)     |
 129                +--------------------+
 130                 /            ^      \
 131                /             |       \
 132           +---------------+  |        +---------------+
 133           | mm_struct     |  |....    | mm_struct     |
 134           |               |  |        |               |
 135           +---------------+  |        +---------------+
 136                              |
 137                              + --------------+
 138                                              |
 139           +---------------+           +------+--------+
 140           | page          +---------->  page_cgroup|
 141           |               |           |               |
 142           +---------------+           +---------------+
 144             (Figure 1: Hierarchy of Accounting)
 147Figure 1 shows the important aspects of the controller
 1491. Accounting happens per cgroup
 1502. Each mm_struct knows about which cgroup it belongs to
 1513. Each page has a pointer to the page_cgroup, which in turn knows the
 152   cgroup it belongs to
 154The accounting is done as follows: mem_cgroup_charge_common() is invoked to
 155set up the necessary data structures and check if the cgroup that is being
 156charged is over its limit. If it is, then reclaim is invoked on the cgroup.
 157More details can be found in the reclaim section of this document.
 158If everything goes well, a page meta-data-structure called page_cgroup is
 159updated. page_cgroup has its own LRU on cgroup.
 160(*) page_cgroup structure is allocated at boot/memory-hotplug time.
 1622.2.1 Accounting details
 164All mapped anon pages (RSS) and cache pages (Page Cache) are accounted.
 165Some pages which are never reclaimable and will not be on the LRU
 166are not accounted. We just account pages under usual VM management.
 168RSS pages are accounted at page_fault unless they've already been accounted
 169for earlier. A file page will be accounted for as Page Cache when it's
 170inserted into inode (radix-tree). While it's mapped into the page tables of
 171processes, duplicate accounting is carefully avoided.
 173An RSS page is unaccounted when it's fully unmapped. A PageCache page is
 174unaccounted when it's removed from radix-tree. Even if RSS pages are fully
 175unmapped (by kswapd), they may exist as SwapCache in the system until they
 176are really freed. Such SwapCaches are also accounted.
 177A swapped-in page is not accounted until it's mapped.
 179Note: The kernel does swapin-readahead and reads multiple swaps at once.
 180This means swapped-in pages may contain pages for other tasks than a task
 181causing page fault. So, we avoid accounting at swap-in I/O.
 183At page migration, accounting information is kept.
 185Note: we just account pages-on-LRU because our purpose is to control amount
 186of used pages; not-on-LRU pages tend to be out-of-control from VM view.
 1882.3 Shared Page Accounting
 190Shared pages are accounted on the basis of the first touch approach. The
 191cgroup that first touches a page is accounted for the page. The principle
 192behind this approach is that a cgroup that aggressively uses a shared
 193page will eventually get charged for it (once it is uncharged from
 194the cgroup that brought it in -- this will happen on memory pressure).
 196But see section 8.2: when moving a task to another cgroup, its pages may
 197be recharged to the new cgroup, if move_charge_at_immigrate has been chosen.
 199Exception: If CONFIG_MEMCG_SWAP is not used.
 200When you do swapoff and make swapped-out pages of shmem(tmpfs) to
 201be backed into memory in force, charges for pages are accounted against the
 202caller of swapoff rather than the users of shmem.
 2042.4 Swap Extension (CONFIG_MEMCG_SWAP)
 206Swap Extension allows you to record charge for swap. A swapped-in page is
 207charged back to original page allocator if possible.
 209When swap is accounted, following files are added.
 210 - memory.memsw.usage_in_bytes.
 211 - memory.memsw.limit_in_bytes.
 213memsw means memory+swap. Usage of memory+swap is limited by
 216Example: Assume a system with 4G of swap. A task which allocates 6G of memory
 217(by mistake) under 2G memory limitation will use all swap.
 218In this case, setting memsw.limit_in_bytes=3G will prevent bad use of swap.
 219By using the memsw limit, you can avoid system OOM which can be caused by swap
 222* why 'memory+swap' rather than swap.
 223The global LRU(kswapd) can swap out arbitrary pages. Swap-out means
 224to move account from memory to swap...there is no change in usage of
 225memory+swap. In other words, when we want to limit the usage of swap without
 226affecting global LRU, memory+swap limit is better than just limiting swap from
 227an OS point of view.
 229* What happens when a cgroup hits memory.memsw.limit_in_bytes
 230When a cgroup hits memory.memsw.limit_in_bytes, it's useless to do swap-out
 231in this cgroup. Then, swap-out will not be done by cgroup routine and file
 232caches are dropped. But as mentioned above, global LRU can do swapout memory
 233from it for sanity of the system's memory management state. You can't forbid
 234it by cgroup.
 2362.5 Reclaim
 238Each cgroup maintains a per cgroup LRU which has the same structure as
 239global VM. When a cgroup goes over its limit, we first try
 240to reclaim memory from the cgroup so as to make space for the new
 241pages that the cgroup has touched. If the reclaim is unsuccessful,
 242an OOM routine is invoked to select and kill the bulkiest task in the
 243cgroup. (See 10. OOM Control below.)
 245The reclaim algorithm has not been modified for cgroups, except that
 246pages that are selected for reclaiming come from the per-cgroup LRU
 249NOTE: Reclaim does not work for the root cgroup, since we cannot set any
 250limits on the root cgroup.
 252Note2: When panic_on_oom is set to "2", the whole system will panic.
 254When oom event notifier is registered, event will be delivered.
 255(See oom_control section)
 2572.6 Locking
 259   lock_page_cgroup()/unlock_page_cgroup() should not be called under
 260   mapping->tree_lock.
 262   Other lock order is following:
 263   PG_locked.
 264   mm->page_table_lock
 265       zone->lru_lock
 266          lock_page_cgroup.
 267  In many cases, just lock_page_cgroup() is called.
 268  per-zone-per-cgroup LRU (cgroup's private LRU) is just guarded by
 269  zone->lru_lock, it has no lock of its own.
 2712.7 Kernel Memory Extension (CONFIG_MEMCG_KMEM)
 273With the Kernel memory extension, the Memory Controller is able to limit
 274the amount of kernel memory used by the system. Kernel memory is fundamentally
 275different than user memory, since it can't be swapped out, which makes it
 276possible to DoS the system by consuming too much of this precious resource.
 278Kernel memory won't be accounted at all until limit on a group is set. This
 279allows for existing setups to continue working without disruption.  The limit
 280cannot be set if the cgroup have children, or if there are already tasks in the
 281cgroup. Attempting to set the limit under those conditions will return -EBUSY.
 282When use_hierarchy == 1 and a group is accounted, its children will
 283automatically be accounted regardless of their limit value.
 285After a group is first limited, it will be kept being accounted until it
 286is removed. The memory limitation itself, can of course be removed by writing
 287-1 to memory.kmem.limit_in_bytes. In this case, kmem will be accounted, but not
 290Kernel memory limits are not imposed for the root cgroup. Usage for the root
 291cgroup may or may not be accounted. The memory used is accumulated into
 292memory.kmem.usage_in_bytes, or in a separate counter when it makes sense.
 293(currently only for tcp).
 294The main "kmem" counter is fed into the main counter, so kmem charges will
 295also be visible from the user counter.
 297Currently no soft limit is implemented for kernel memory. It is future work
 298to trigger slab reclaim when those limits are reached.
 3002.7.1 Current Kernel Memory resources accounted
 302* stack pages: every process consumes some stack pages. By accounting into
 303kernel memory, we prevent new processes from being created when the kernel
 304memory usage is too high.
 306* slab pages: pages allocated by the SLAB or SLUB allocator are tracked. A copy
 307of each kmem_cache is created every time the cache is touched by the first time
 308from inside the memcg. The creation is done lazily, so some objects can still be
 309skipped while the cache is being created. All objects in a slab page should
 310belong to the same memcg. This only fails to hold when a task is migrated to a
 311different memcg during the page allocation by the cache.
 313* sockets memory pressure: some sockets protocols have memory pressure
 314thresholds. The Memory Controller allows them to be controlled individually
 315per cgroup, instead of globally.
 317* tcp memory pressure: sockets memory pressure for the tcp protocol.
 3192.7.3 Common use cases
 321Because the "kmem" counter is fed to the main user counter, kernel memory can
 322never be limited completely independently of user memory. Say "U" is the user
 323limit, and "K" the kernel limit. There are three possible ways limits can be
 326    U != 0, K = unlimited:
 327    This is the standard memcg limitation mechanism already present before kmem
 328    accounting. Kernel memory is completely ignored.
 330    U != 0, K < U:
 331    Kernel memory is a subset of the user memory. This setup is useful in
 332    deployments where the total amount of memory per-cgroup is overcommited.
 333    Overcommiting kernel memory limits is definitely not recommended, since the
 334    box can still run out of non-reclaimable memory.
 335    In this case, the admin could set up K so that the sum of all groups is
 336    never greater than the total memory, and freely set U at the cost of his
 337    QoS.
 339    U != 0, K >= U:
 340    Since kmem charges will also be fed to the user counter and reclaim will be
 341    triggered for the cgroup for both kinds of memory. This setup gives the
 342    admin a unified view of memory, and it is also useful for people who just
 343    want to track kernel memory usage.
 3453. User Interface
 3470. Configuration
 349a. Enable CONFIG_CGROUPS
 351c. Enable CONFIG_MEMCG
 352d. Enable CONFIG_MEMCG_SWAP (to use swap extension)
 353d. Enable CONFIG_MEMCG_KMEM (to use kmem extension)
 3551. Prepare the cgroups (see cgroups.txt, Why are cgroups needed?)
 356# mount -t tmpfs none /sys/fs/cgroup
 357# mkdir /sys/fs/cgroup/memory
 358# mount -t cgroup none /sys/fs/cgroup/memory -o memory
 3602. Make the new group and move bash into it
 361# mkdir /sys/fs/cgroup/memory/0
 362# echo $$ > /sys/fs/cgroup/memory/0/tasks
 364Since now we're in the 0 cgroup, we can alter the memory limit:
 365# echo 4M > /sys/fs/cgroup/memory/0/memory.limit_in_bytes
 367NOTE: We can use a suffix (k, K, m, M, g or G) to indicate values in kilo,
 368mega or gigabytes. (Here, Kilo, Mega, Giga are Kibibytes, Mebibytes, Gibibytes.)
 370NOTE: We can write "-1" to reset the *.limit_in_bytes(unlimited).
 371NOTE: We cannot set limits on the root cgroup any more.
 373# cat /sys/fs/cgroup/memory/0/memory.limit_in_bytes
 376We can check the usage:
 377# cat /sys/fs/cgroup/memory/0/memory.usage_in_bytes
 380A successful write to this file does not guarantee a successful setting of
 381this limit to the value written into the file. This can be due to a
 382number of factors, such as rounding up to page boundaries or the total
 383availability of memory on the system. The user is required to re-read
 384this file after a write to guarantee the value committed by the kernel.
 386# echo 1 > memory.limit_in_bytes
 387# cat memory.limit_in_bytes
 390The memory.failcnt field gives the number of times that the cgroup limit was
 393The memory.stat file gives accounting information. Now, the number of
 394caches, RSS and Active pages/Inactive pages are shown.
 3964. Testing
 398For testing features and implementation, see memcg_test.txt.
 400Performance test is also important. To see pure memory controller's overhead,
 401testing on tmpfs will give you good numbers of small overheads.
 402Example: do kernel make on tmpfs.
 404Page-fault scalability is also important. At measuring parallel
 405page fault test, multi-process test may be better than multi-thread
 406test because it has noise of shared objects/status.
 408But the above two are testing extreme situations.
 409Trying usual test under memory controller is always helpful.
 4114.1 Troubleshooting
 413Sometimes a user might find that the application under a cgroup is
 414terminated by the OOM killer. There are several causes for this:
 4161. The cgroup limit is too low (just too low to do anything useful)
 4172. The user is using anonymous memory and swap is turned off or too low
 419A sync followed by echo 1 > /proc/sys/vm/drop_caches will help get rid of
 420some of the pages cached in the cgroup (page cache pages).
 422To know what happens, disabling OOM_Kill as per "10. OOM Control" (below) and
 423seeing what happens will be helpful.
 4254.2 Task migration
 427When a task migrates from one cgroup to another, its charge is not
 428carried forward by default. The pages allocated from the original cgroup still
 429remain charged to it, the charge is dropped when the page is freed or
 432You can move charges of a task along with task migration.
 433See 8. "Move charges at task migration"
 4354.3 Removing a cgroup
 437A cgroup can be removed by rmdir, but as discussed in sections 4.1 and 4.2, a
 438cgroup might have some charge associated with it, even though all
 439tasks have migrated away from it. (because we charge against pages, not
 440against tasks.)
 442We move the stats to root (if use_hierarchy==0) or parent (if
 443use_hierarchy==1), and no change on the charge except uncharging
 444from the child.
 446Charges recorded in swap information is not updated at removal of cgroup.
 447Recorded information is discarded and a cgroup which uses swap (swapcache)
 448will be charged as a new owner of it.
 450About use_hierarchy, see Section 6.
 4525. Misc. interfaces.
 4545.1 force_empty
 455  memory.force_empty interface is provided to make cgroup's memory usage empty.
 456  You can use this interface only when the cgroup has no tasks.
 457  When writing anything to this
 459  # echo 0 > memory.force_empty
 461  Almost all pages tracked by this memory cgroup will be unmapped and freed.
 462  Some pages cannot be freed because they are locked or in-use. Such pages are
 463  moved to parent (if use_hierarchy==1) or root (if use_hierarchy==0) and this
 464  cgroup will be empty.
 466  The typical use case for this interface is before calling rmdir().
 467  Because rmdir() moves all pages to parent, some out-of-use page caches can be
 468  moved to the parent. If you want to avoid that, force_empty will be useful.
 470  Also, note that when memory.kmem.limit_in_bytes is set the charges due to
 471  kernel pages will still be seen. This is not considered a failure and the
 472  write will still return success. In this case, it is expected that
 473  memory.kmem.usage_in_bytes == memory.usage_in_bytes.
 475  About use_hierarchy, see Section 6.
 4775.2 stat file
 479memory.stat file includes following statistics
 481# per-memory cgroup local status
 482cache           - # of bytes of page cache memory.
 483rss             - # of bytes of anonymous and swap cache memory (includes
 484                transparent hugepages).
 485rss_huge        - # of bytes of anonymous transparent hugepages.
 486mapped_file     - # of bytes of mapped file (includes tmpfs/shmem)
 487pgpgin          - # of charging events to the memory cgroup. The charging
 488                event happens each time a page is accounted as either mapped
 489                anon page(RSS) or cache page(Page Cache) to the cgroup.
 490pgpgout         - # of uncharging events to the memory cgroup. The uncharging
 491                event happens each time a page is unaccounted from the cgroup.
 492swap            - # of bytes of swap usage
 493writeback       - # of bytes of file/anon cache that are queued for syncing to
 494                disk.
 495inactive_anon   - # of bytes of anonymous and swap cache memory on inactive
 496                LRU list.
 497active_anon     - # of bytes of anonymous and swap cache memory on active
 498                LRU list.
 499inactive_file   - # of bytes of file-backed memory on inactive LRU list.
 500active_file     - # of bytes of file-backed memory on active LRU list.
 501unevictable     - # of bytes of memory that cannot be reclaimed (mlocked etc).
 503# status considering hierarchy (see memory.use_hierarchy settings)
 505hierarchical_memory_limit - # of bytes of memory limit with regard to hierarchy
 506                        under which the memory cgroup is
 507hierarchical_memsw_limit - # of bytes of memory+swap limit with regard to
 508                        hierarchy under which memory cgroup is.
 510total_<counter>         - # hierarchical version of <counter>, which in
 511                        addition to the cgroup's own value includes the
 512                        sum of all hierarchical children's values of
 513                        <counter>, i.e. total_cache
 515# The following additional stats are dependent on CONFIG_DEBUG_VM.
 517recent_rotated_anon     - VM internal parameter. (see mm/vmscan.c)
 518recent_rotated_file     - VM internal parameter. (see mm/vmscan.c)
 519recent_scanned_anon     - VM internal parameter. (see mm/vmscan.c)
 520recent_scanned_file     - VM internal parameter. (see mm/vmscan.c)
 523        recent_rotated means recent frequency of LRU rotation.
 524        recent_scanned means recent # of scans to LRU.
 525        showing for better debug please see the code for meanings.
 528        Only anonymous and swap cache memory is listed as part of 'rss' stat.
 529        This should not be confused with the true 'resident set size' or the
 530        amount of physical memory used by the cgroup.
 531        'rss + file_mapped" will give you resident set size of cgroup.
 532        (Note: file and shmem may be shared among other cgroups. In that case,
 533         file_mapped is accounted only when the memory cgroup is owner of page
 534         cache.)
 5365.3 swappiness
 538Similar to /proc/sys/vm/swappiness, but affecting a hierarchy of groups only.
 539Please note that unlike the global swappiness, memcg knob set to 0
 540really prevents from any swapping even if there is a swap storage
 541available. This might lead to memcg OOM killer if there are no file
 542pages to reclaim.
 544Following cgroups' swappiness can't be changed.
 545- root cgroup (uses /proc/sys/vm/swappiness).
 546- a cgroup which uses hierarchy and it has other cgroup(s) below it.
 547- a cgroup which uses hierarchy and not the root of hierarchy.
 5495.4 failcnt
 551A memory cgroup provides memory.failcnt and memory.memsw.failcnt files.
 552This failcnt(== failure count) shows the number of times that a usage counter
 553hit its limit. When a memory cgroup hits a limit, failcnt increases and
 554memory under it will be reclaimed.
 556You can reset failcnt by writing 0 to failcnt file.
 557# echo 0 > .../memory.failcnt
 5595.5 usage_in_bytes
 561For efficiency, as other kernel components, memory cgroup uses some optimization
 562to avoid unnecessary cacheline false sharing. usage_in_bytes is affected by the
 563method and doesn't show 'exact' value of memory (and swap) usage, it's a fuzz
 564value for efficient access. (Of course, when necessary, it's synchronized.)
 565If you want to know more exact memory usage, you should use RSS+CACHE(+SWAP)
 566value in memory.stat(see 5.2).
 5685.6 numa_stat
 570This is similar to numa_maps but operates on a per-memcg basis.  This is
 571useful for providing visibility into the numa locality information within
 572an memcg since the pages are allowed to be allocated from any physical
 573node.  One of the use cases is evaluating application performance by
 574combining this information with the application's CPU allocation.
 576Each memcg's numa_stat file includes "total", "file", "anon" and "unevictable"
 577per-node page counts including "hierarchical_<counter>" which sums up all
 578hierarchical children's values in addition to the memcg's own value.
 580The output format of memory.numa_stat is:
 582total=<total pages> N0=<node 0 pages> N1=<node 1 pages> ...
 583file=<total file pages> N0=<node 0 pages> N1=<node 1 pages> ...
 584anon=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ...
 585unevictable=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ...
 586hierarchical_<counter>=<counter pages> N0=<node 0 pages> N1=<node 1 pages> ...
 588The "total" count is sum of file + anon + unevictable.
 5906. Hierarchy support
 592The memory controller supports a deep hierarchy and hierarchical accounting.
 593The hierarchy is created by creating the appropriate cgroups in the
 594cgroup filesystem. Consider for example, the following cgroup filesystem
 597               root
 598             /  |   \
 599            /   |    \
 600           a    b     c
 601                      | \
 602                      |  \
 603                      d   e
 605In the diagram above, with hierarchical accounting enabled, all memory
 606usage of e, is accounted to its ancestors up until the root (i.e, c and root),
 607that has memory.use_hierarchy enabled. If one of the ancestors goes over its
 608limit, the reclaim algorithm reclaims from the tasks in the ancestor and the
 609children of the ancestor.
 6116.1 Enabling hierarchical accounting and reclaim
 613A memory cgroup by default disables the hierarchy feature. Support
 614can be enabled by writing 1 to memory.use_hierarchy file of the root cgroup
 616# echo 1 > memory.use_hierarchy
 618The feature can be disabled by
 620# echo 0 > memory.use_hierarchy
 622NOTE1: Enabling/disabling will fail if either the cgroup already has other
 623       cgroups created below it, or if the parent cgroup has use_hierarchy
 624       enabled.
 626NOTE2: When panic_on_oom is set to "2", the whole system will panic in
 627       case of an OOM event in any cgroup.
 6297. Soft limits
 631Soft limits allow for greater sharing of memory. The idea behind soft limits
 632is to allow control groups to use as much of the memory as needed, provided
 634a. There is no memory contention
 635b. They do not exceed their hard limit
 637When the system detects memory contention or low memory, control groups
 638are pushed back to their soft limits. If the soft limit of each control
 639group is very high, they are pushed back as much as possible to make
 640sure that one control group does not starve the others of memory.
 642Please note that soft limits is a best-effort feature; it comes with
 643no guarantees, but it does its best to make sure that when memory is
 644heavily contended for, memory is allocated based on the soft limit
 645hints/setup. Currently soft limit based reclaim is set up such that
 646it gets invoked from balance_pgdat (kswapd).
 6487.1 Interface
 650Soft limits can be setup by using the following commands (in this example we
 651assume a soft limit of 256 MiB)
 653# echo 256M > memory.soft_limit_in_bytes
 655If we want to change this to 1G, we can at any time use
 657# echo 1G > memory.soft_limit_in_bytes
 659NOTE1: Soft limits take effect over a long period of time, since they involve
 660       reclaiming memory for balancing between memory cgroups
 661NOTE2: It is recommended to set the soft limit always below the hard limit,
 662       otherwise the hard limit will take precedence.
 6648. Move charges at task migration
 666Users can move charges associated with a task along with task migration, that
 667is, uncharge task's pages from the old cgroup and charge them to the new cgroup.
 668This feature is not supported in !CONFIG_MMU environments because of lack of
 669page tables.
 6718.1 Interface
 673This feature is disabled by default. It can be enabled (and disabled again) by
 674writing to memory.move_charge_at_immigrate of the destination cgroup.
 676If you want to enable it:
 678# echo (some positive value) > memory.move_charge_at_immigrate
 680Note: Each bits of move_charge_at_immigrate has its own meaning about what type
 681      of charges should be moved. See 8.2 for details.
 682Note: Charges are moved only when you move mm->owner, in other words,
 683      a leader of a thread group.
 684Note: If we cannot find enough space for the task in the destination cgroup, we
 685      try to make space by reclaiming memory. Task migration may fail if we
 686      cannot make enough space.
 687Note: It can take several seconds if you move charges much.
 689And if you want disable it again:
 691# echo 0 > memory.move_charge_at_immigrate
 6938.2 Type of charges which can be moved
 695Each bit in move_charge_at_immigrate has its own meaning about what type of
 696charges should be moved. But in any case, it must be noted that an account of
 697a page or a swap can be moved only when it is charged to the task's current
 698(old) memory cgroup.
 700  bit | what type of charges would be moved ?
 701 -----+------------------------------------------------------------------------
 702   0  | A charge of an anonymous page (or swap of it) used by the target task.
 703      | You must enable Swap Extension (see 2.4) to enable move of swap charges.
 704 -----+------------------------------------------------------------------------
 705   1  | A charge of file pages (normal file, tmpfs file (e.g. ipc shared memory)
 706      | and swaps of tmpfs file) mmapped by the target task. Unlike the case of
 707      | anonymous pages, file pages (and swaps) in the range mmapped by the task
 708      | will be moved even if the task hasn't done page fault, i.e. they might
 709      | not be the task's "RSS", but other task's "RSS" that maps the same file.
 710      | And mapcount of the page is ignored (the page can be moved even if
 711      | page_mapcount(page) > 1). You must enable Swap Extension (see 2.4) to
 712      | enable move of swap charges.
 7148.3 TODO
 716- All of moving charge operations are done under cgroup_mutex. It's not good
 717  behavior to hold the mutex too long, so we may need some trick.
 7199. Memory thresholds
 721Memory cgroup implements memory thresholds using the cgroups notification
 722API (see cgroups.txt). It allows to register multiple memory and memsw
 723thresholds and gets notifications when it crosses.
 725To register a threshold, an application must:
 726- create an eventfd using eventfd(2);
 727- open memory.usage_in_bytes or memory.memsw.usage_in_bytes;
 728- write string like "<event_fd> <fd of memory.usage_in_bytes> <threshold>" to
 729  cgroup.event_control.
 731Application will be notified through eventfd when memory usage crosses
 732threshold in any direction.
 734It's applicable for root and non-root cgroup.
 73610. OOM Control
 738memory.oom_control file is for OOM notification and other controls.
 740Memory cgroup implements OOM notifier using the cgroup notification
 741API (See cgroups.txt). It allows to register multiple OOM notification
 742delivery and gets notification when OOM happens.
 744To register a notifier, an application must:
 745 - create an eventfd using eventfd(2)
 746 - open memory.oom_control file
 747 - write string like "<event_fd> <fd of memory.oom_control>" to
 748   cgroup.event_control
 750The application will be notified through eventfd when OOM happens.
 751OOM notification doesn't work for the root cgroup.
 753You can disable the OOM-killer by writing "1" to memory.oom_control file, as:
 755        #echo 1 > memory.oom_control
 757This operation is only allowed to the top cgroup of a sub-hierarchy.
 758If OOM-killer is disabled, tasks under cgroup will hang/sleep
 759in memory cgroup's OOM-waitqueue when they request accountable memory.
 761For running them, you have to relax the memory cgroup's OOM status by
 762        * enlarge limit or reduce usage.
 763To reduce usage,
 764        * kill some tasks.
 765        * move some tasks to other group with account migration.
 766        * remove some files (on tmpfs?)
 768Then, stopped tasks will work again.
 770At reading, current status of OOM is shown.
 771        oom_kill_disable 0 or 1 (if 1, oom-killer is disabled)
 772        under_oom        0 or 1 (if 1, the memory cgroup is under OOM, tasks may
 773                                 be stopped.)
 77511. Memory Pressure
 777The pressure level notifications can be used to monitor the memory
 778allocation cost; based on the pressure, applications can implement
 779different strategies of managing their memory resources. The pressure
 780levels are defined as following:
 782The "low" level means that the system is reclaiming memory for new
 783allocations. Monitoring this reclaiming activity might be useful for
 784maintaining cache level. Upon notification, the program (typically
 785"Activity Manager") might analyze vmstat and act in advance (i.e.
 786prematurely shutdown unimportant services).
 788The "medium" level means that the system is experiencing medium memory
 789pressure, the system might be making swap, paging out active file caches,
 790etc. Upon this event applications may decide to further analyze
 791vmstat/zoneinfo/memcg or internal memory usage statistics and free any
 792resources that can be easily reconstructed or re-read from a disk.
 794The "critical" level means that the system is actively thrashing, it is
 795about to out of memory (OOM) or even the in-kernel OOM killer is on its
 796way to trigger. Applications should do whatever they can to help the
 797system. It might be too late to consult with vmstat or any other
 798statistics, so it's advisable to take an immediate action.
 800The events are propagated upward until the event is handled, i.e. the
 801events are not pass-through. Here is what this means: for example you have
 802three cgroups: A->B->C. Now you set up an event listener on cgroups A, B
 803and C, and suppose group C experiences some pressure. In this situation,
 804only group C will receive the notification, i.e. groups A and B will not
 805receive it. This is done to avoid excessive "broadcasting" of messages,
 806which disturbs the system and which is especially bad if we are low on
 807memory or thrashing. So, organize the cgroups wisely, or propagate the
 808events manually (or, ask us to implement the pass-through events,
 809explaining why would you need them.)
 811The file memory.pressure_level is only used to setup an eventfd. To
 812register a notification, an application must:
 814- create an eventfd using eventfd(2);
 815- open memory.pressure_level;
 816- write string like "<event_fd> <fd of memory.pressure_level> <level>"
 817  to cgroup.event_control.
 819Application will be notified through eventfd when memory pressure is at
 820the specific level (or higher). Read/write operations to
 821memory.pressure_level are no implemented.
 825   Here is a small script example that makes a new cgroup, sets up a
 826   memory limit, sets up a notification in the cgroup and then makes child
 827   cgroup experience a critical pressure:
 829   # cd /sys/fs/cgroup/memory/
 830   # mkdir foo
 831   # cd foo
 832   # cgroup_event_listener memory.pressure_level low &
 833   # echo 8000000 > memory.limit_in_bytes
 834   # echo 8000000 > memory.memsw.limit_in_bytes
 835   # echo $$ > tasks
 836   # dd if=/dev/zero | read x
 838   (Expect a bunch of notifications, and eventually, the oom-killer will
 839   trigger.)
 84112. TODO
 8431. Make per-cgroup scanner reclaim not-shared pages first
 8442. Teach controller to account for shared-pages
 8453. Start reclamation in the background when the limit is
 846   not yet hit but the usage is getting closer
 850Overall, the memory controller has been a stable controller and has been
 851commented and discussed quite extensively in the community.
 8551. Singh, Balbir. RFC: Memory Controller,
 8562. Singh, Balbir. Memory Controller (RSS Control),
 8583. Emelianov, Pavel. Resource controllers based on process cgroups
 8604. Emelianov, Pavel. RSS controller based on process cgroups (v2)
 8625. Emelianov, Pavel. RSS controller based on process cgroups (v3)
 8646. Menage, Paul. Control Groups v10,
 8657. Vaidyanathan, Srinivasan, Control Groups: Pagecache accounting and control
 866   subsystem (v3),
 8678. Singh, Balbir. RSS controller v2 test results (lmbench),
 8699. Singh, Balbir. RSS controller v2 AIM9 results
 87110. Singh, Balbir. Memory controller v6 test results,
 87311. Singh, Balbir. Memory controller introduction (v6),
 87512. Corbet, Jonathan, Controlling memory use in cgroups,
 877 kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.