Control Group v2 October, 2015 Tejun Heo This is the authoritative documentation on the design, interface and conventions of cgroup v2. It describes all userland-visible aspects of cgroup including core and specific controller behaviors. All future changes must be reflected in this document. Documentation for v1 is available under Documentation/cgroup-legacy/. CONTENTS 1. Introduction 1-1. Terminology 1-2. What is cgroup? 2. Basic Operations 2-1. Mounting 2-2. Organizing Processes 2-3. [Un]populated Notification 2-4. Controlling Controllers 2-4-1. Enabling and Disabling 2-4-2. Top-down Constraint 2-4-3. No Internal Process Constraint 2-5. Delegation 2-5-1. Model of Delegation 2-5-2. Delegation Containment 2-6. Guidelines 2-6-1. Organize Once and Control 2-6-2. Avoid Name Collisions 3. Resource Distribution Models 3-1. Weights 3-2. Limits 3-3. Protections 3-4. Allocations 4. Interface Files 4-1. Format 4-2. Conventions 4-3. Core Interface Files 5. Controllers 5-1. CPU 5-1-1. CPU Interface Files 5-2. Memory 5-2-1. Memory Interface Files 5-2-2. Usage Guidelines 5-2-3. Memory Ownership 5-3. IO 5-3-1. IO Interface Files 5-3-2. Writeback P. Information on Kernel Programming P-1. Filesystem Support for Writeback D. Deprecated v1 Core Features R. Issues with v1 and Rationales for v2 R-1. Multiple Hierarchies R-2. Thread Granularity R-3. Competition Between Inner Nodes and Threads R-4. Other Interface Issues R-5. Controller Issues and Remedies R-5-1. Memory 1. Introduction 1-1. Terminology "cgroup" stands for "control group" and is never capitalized. The singular form is used to designate the whole feature and also as a qualifier as in "cgroup controllers". When explicitly referring to multiple individual control groups, the plural form "cgroups" is used. 1-2. What is cgroup? cgroup is a mechanism to organize processes hierarchically and distribute system resources along the hierarchy in a controlled and configurable manner. cgroup is largely composed of two parts - the core and controllers. cgroup core is primarily responsible for hierarchically organizing processes. A cgroup controller is usually responsible for distributing a specific type of system resource along the hierarchy although there are utility controllers which serve purposes other than resource distribution. cgroups form a tree structure and every process in the system belongs to one and only one cgroup. All threads of a process belong to the same cgroup. On creation, all processes are put in the cgroup that the parent process belongs to at the time. A process can be migrated to another cgroup. Migration of a process doesn't affect already existing descendant processes. Following certain structural constraints, controllers may be enabled or disabled selectively on a cgroup. All controller behaviors are hierarchical - if a controller is enabled on a cgroup, it affects all processes which belong to the cgroups consisting the inclusive sub-hierarchy of the cgroup. When a controller is enabled on a nested cgroup, it always restricts the resource distribution further. The restrictions set closer to the root in the hierarchy can not be overridden from further away. 2. Basic Operations 2-1. Mounting Unlike v1, cgroup v2 has only single hierarchy. The cgroup v2 hierarchy can be mounted with the following mount command. # mount -t cgroup2 none $MOUNT_POINT cgroup2 filesystem has the magic number 0x63677270 ("cgrp"). All controllers which support v2 and are not bound to a v1 hierarchy are automatically bound to the v2 hierarchy and show up at the root. Controllers which are not in active use in the v2 hierarchy can be bound to other hierarchies. This allows mixing v2 hierarchy with the legacy v1 multiple hierarchies in a fully backward compatible way. A controller can be moved across hierarchies only after the controller is no longer referenced in its current hierarchy. Because per-cgroup controller states are destroyed asynchronously and controllers may have lingering references, a controller may not show up immediately on the v2 hierarchy after the final umount of the previous hierarchy. Similarly, a controller should be fully disabled to be moved out of the unified hierarchy and it may take some time for the disabled controller to become available for other hierarchies; furthermore, due to inter-controller dependencies, other controllers may need to be disabled too. While useful for development and manual configurations, moving controllers dynamically between the v2 and other hierarchies is strongly discouraged for production use. It is recommended to decide the hierarchies and controller associations before starting using the controllers after system boot. 2-2. Organizing Processes Initially, only the root cgroup exists to which all processes belong. A child cgroup can be created by creating a sub-directory. # mkdir $CGROUP_NAME A given cgroup may have multiple child cgroups forming a tree structure. Each cgroup has a read-writable interface file "cgroup.procs". When read, it lists the PIDs of all processes which belong to the cgroup one-per-line. The PIDs are not ordered and the same PID may show up more than once if the process got moved to another cgroup and then back or the PID got recycled while reading. A process can be migrated into a cgroup by writing its PID to the target cgroup's "cgroup.procs" file. Only one process can be migrated on a single write(2) call. If a process is composed of multiple threads, writing the PID of any thread migrates all threads of the process. When a process forks a child process, the new process is born into the cgroup that the forking process belongs to at the time of the operation. After exit, a process stays associated with the cgroup that it belonged to at the time of exit until it's reaped; however, a zombie process does not appear in "cgroup.procs" and thus can't be moved to another cgroup. A cgroup which doesn't have any children or live processes can be destroyed by removing the directory. Note that a cgroup which doesn't have any children and is associated only with zombie processes is considered empty and can be removed. # rmdir $CGROUP_NAME "/proc/$PID/cgroup" lists a process's cgroup membership. If legacy cgroup is in use in the system, this file may contain multiple lines, one for each hierarchy. The entry for cgroup v2 is always in the format "0::$PATH". # cat /proc/842/cgroup ... 0::/test-cgroup/test-cgroup-nested If the process becomes a zombie and the cgroup it was associated with is removed subsequently, " (deleted)" is appended to the path. # cat /proc/842/cgroup ... 0::/test-cgroup/test-cgroup-nested (deleted) 2-3. [Un]populated Notification Each non-root cgroup has a "cgroup.events" file which contains "populated" field indicating whether the cgroup's sub-hierarchy has live processes in it. Its value is 0 if there is no live process in the cgroup and its descendants; otherwise, 1. poll and [id]notify events are triggered when the value changes. This can be used, for example, to start a clean-up operation after all processes of a given sub-hierarchy have exited. The populated state updates and notifications are recursive. Consider the following sub-hierarchy where the numbers in the parentheses represent the numbers of processes in each cgroup. A(4) - B(0) - C(1) \ D(0) A, B and C's "populated" fields would be 1 while D's 0. After the one process in C exits, B and C's "populated" fields would flip to "0" and file modified events will be generated on the "cgroup.events" files of both cgroups. 2-4. Controlling Controllers 2-4-1. Enabling and Disabling Each cgroup has a "cgroup.controllers" file which lists all controllers available for the cgroup to enable. # cat cgroup.controllers cpu io memory No controller is enabled by default. Controllers can be enabled and disabled by writing to the "cgroup.subtree_control" file. # echo "+cpu +memory -io" > cgroup.subtree_control Only controllers which are listed in "cgroup.controllers" can be enabled. When multiple operations are specified as above, either they all succeed or fail. If multiple operations on the same controller are specified, the last one is effective. Enabling a controller in a cgroup indicates that the distribution of the target resource across its immediate children will be controlled. Consider the following sub-hierarchy. The enabled controllers are listed in parentheses. A(cpu,memory) - B(memory) - C() \ D() As A has "cpu" and "memory" enabled, A will control the distribution of CPU cycles and memory to its children, in this case, B. As B has "memory" enabled but not "CPU", C and D will compete freely on CPU cycles but their division of memory available to B will be controlled. As a controller regulates the distribution of the target resource to the cgroup's children, enabling it creates the controller's interface files in the child cgroups. In the above example, enabling "cpu" on B would create the "cpu." prefixed controller interface files in C and D. Likewise, disabling "memory" from B would remove the "memory." prefixed controller interface files from C and D. This means that the controller interface files - anything which doesn't start with "cgroup." are owned by the parent rather than the cgroup itself. 2-4-2. Top-down Constraint Resources are distributed top-down and a cgroup can further distribute a resource only if the resource has been distributed to it from the parent. This means that all non-root "cgroup.subtree_control" files can only contain controllers which are enabled in the parent's "cgroup.subtree_control" file. A controller can be enabled only if the parent has the controller enabled and a controller can't be disabled if one or more children have it enabled. 2-4-3. No Internal Process Constraint Non-root cgroups can only distribute resources to their children when they don't have any processes of their own. In other words, only cgroups which don't contain any processes can have controllers enabled in their "cgroup.subtree_control" files. This guarantees that, when a controller is looking at the part of the hierarchy which has it enabled, processes are always only on the leaves. This rules out situations where child cgroups compete against internal processes of the parent. The root cgroup is exempt from this restriction. Root contains processes and anonymous resource consumption which can't be associated with any other cgroups and requires special treatment from most controllers. How resource consumption in the root cgroup is governed is up to each controller. Note that the restriction doesn't get in the way if there is no enabled controller in the cgroup's "cgroup.subtree_control". This is important as otherwise it wouldn't be possible to create children of a populated cgroup. To control resource distribution of a cgroup, the cgroup must create children and transfer all its processes to the children before enabling controllers in its "cgroup.subtree_control" file. 2-5. Delegation 2-5-1. Model of Delegation A cgroup can be delegated to a less privileged user by granting write access of the directory and its "cgroup.procs" file to the user. Note that resource control interface files in a given directory control the distribution of the parent's resources and thus must not be delegated along with the directory. Once delegated, the user can build sub-hierarchy under the directory, organize processes as it sees fit and further distribute the resources it received from the parent. The limits and other settings of all resource controllers are hierarchical and regardless of what happens in the delegated sub-hierarchy, nothing can escape the resource restrictions imposed by the parent. Currently, cgroup doesn't impose any restrictions on the number of cgroups in or nesting depth of a delegated sub-hierarchy; however, this may be limited explicitly in the future. 2-5-2. Delegation Containment A delegated sub-hierarchy is contained in the sense that processes can't be moved into or out of the sub-hierarchy by the delegatee. For a process with a non-root euid to migrate a target process into a cgroup by writing its PID to the "cgroup.procs" file, the following conditions must be met. - The writer's euid must match either uid or suid of the target process. - The writer must have write access to the "cgroup.procs" file. - The writer must have write access to the "cgroup.procs" file of the common ancestor of the source and destination cgroups. The above three constraints ensure that while a delegatee may migrate processes around freely in the delegated sub-hierarchy it can't pull in from or push out to outside the sub-hierarchy. For an example, let's assume cgroups C0 and C1 have been delegated to user U0 who created C00, C01 under C0 and C10 under C1 as follows and all processes under C0 and C1 belong to U0. ~~~~~~~~~~~~~ - C0 - C00 ~ cgroup ~ \ C01 ~ hierarchy ~ ~~~~~~~~~~~~~ - C1 - C10 Let's also say U0 wants to write the PID of a process which is currently in C10 into "C00/cgroup.procs". U0 has write access to the file and uid match on the process; however, the common ancestor of the source cgroup C10 and the destination cgroup C00 is above the points of delegation and U0 would not have write access to its "cgroup.procs" files and thus the write will be denied with -EACCES. 2-6. Guidelines 2-6-1. Organize Once and Control Migrating a process across cgroups is a relatively expensive operation and stateful resources such as memory are not moved together with the process. This is an explicit design decision as there often exist inherent trade-offs between migration and various hot paths in terms of synchronization cost. As such, migrating processes across cgroups frequently as a means to apply different resource restrictions is discouraged. A workload should be assigned to a cgroup according to the system's logical and resource structure once on start-up. Dynamic adjustments to resource distribution can be made by changing controller configuration through the interface files. 2-6-2. Avoid Name Collisions Interface files for a cgroup and its children cgroups occupy the same directory and it is possible to create children cgroups which collide with interface files. All cgroup core interface files are prefixed with "cgroup." and each controller's interface files are prefixed with the controller name and a dot. A controller's name is composed of lower case alphabets and '_'s but never begins with an '_' so it can be used as the prefix character for collision avoidance. Also, interface file names won't start or end with terms which are often used in categorizing workloads such as job, service, slice, unit or workload. cgroup doesn't do anything to prevent name collisions and it's the user's responsibility to avoid them. 3. Resource Distribution Models cgroup controllers implement several resource distribution schemes depending on the resource type and expected use cases. This section describes major schemes in use along with their expected behaviors. 3-1. Weights A parent's resource is distributed by adding up the weights of all active children and giving each the fraction matching the ratio of its weight against the sum. As only children which can make use of the resource at the moment participate in the distribution, this is work-conserving. Due to the dynamic nature, this model is usually used for stateless resources. All weights are in the range [1, 10000] with the default at 100. This allows symmetric multiplicative biases in both directions at fine enough granularity while staying in the intuitive range. As long as the weight is in range, all configuration combinations are valid and there is no reason to reject configuration changes or process migrations. "cpu.weight" proportionally distributes CPU cycles to active children and is an example of this type. 3-2. Limits A child can only consume upto the configured amount of the resource. Limits can be over-committed - the sum of the limits of children can exceed the amount of resource available to the parent. Limits are in the range [0, max] and defaults to "max", which is noop. As limits can be over-committed, all configuration combinations are valid and there is no reason to reject configuration changes or process migrations. "io.max" limits the maximum BPS and/or IOPS that a cgroup can consume on an IO device and is an example of this type. 3-3. Protections A cgroup is protected to be allocated upto the configured amount of the resource if the usages of all its ancestors are under their protected levels. Protections can be hard guarantees or best effort soft boundaries. Protections can also be over-committed in which case only upto the amount available to the parent is protected among children. Protections are in the range [0, max] and defaults to 0, which is noop. As protections can be over-committed, all configuration combinations are valid and there is no reason to reject configuration changes or process migrations. "memory.low" implements best-effort memory protection and is an example of this type. 3-4. Allocations A cgroup is exclusively allocated a certain amount of a finite resource. Allocations can't be over-committed - the sum of the allocations of children can not exceed the amount of resource available to the parent. Allocations are in the range [0, max] and defaults to 0, which is no resource. As allocations can't be over-committed, some configuration combinations are invalid and should be rejected. Also, if the resource is mandatory for execution of processes, process migrations may be rejected. "cpu.rt.max" hard-allocates realtime slices and is an example of this type. 4. Interface Files 4-1. Format All interface files should be in one of the following formats whenever possible. New-line separated values (when only one value can be written at once) VAL0\n VAL1\n ... Space separated values (when read-only or multiple values can be written at once) VAL0 VAL1 ...\n Flat keyed KEY0 VAL0\n KEY1 VAL1\n ... Nested keyed KEY0 SUB_KEY0=VAL00 SUB_KEY1=VAL01... KEY1 SUB_KEY0=VAL10 SUB_KEY1=VAL11... ... For a writable file, the format for writing should generally match reading; however, controllers may allow omitting later fields or implement restricted shortcuts for most common use cases. For both flat and nested keyed files, only the values for a single key can be written at a time. For nested keyed files, the sub key pairs may be specified in any order and not all pairs have to be specified. 4-2. Conventions - Settings for a single feature should be contained in a single file. - The root cgroup should be exempt from resource control and thus shouldn't have resource control interface files. Also, informational files on the root cgroup which end up showing global information available elsewhere shouldn't exist. - If a controller implements weight based resource distribution, its interface file should be named "weight" and have the range [1, 10000] with 100 as the default. The values are chosen to allow enough and symmetric bias in both directions while keeping it intuitive (the default is 100%). - If a controller implements an absolute resource guarantee and/or limit, the interface files should be named "min" and "max" respectively. If a controller implements best effort resource guarantee and/or limit, the interface files should be named "low" and "high" respectively. In the above four control files, the special token "max" should be used to represent upward infinity for both reading and writing. - If a setting has a configurable default value and keyed specific overrides, the default entry should be keyed with "default" and appear as the first entry in the file. The default value can be updated by writing either "default $VAL" or "$VAL". When writing to update a specific override, "default" can be used as the value to indicate removal of the override. Override entries with "default" as the value must not appear when read. For example, a setting which is keyed by major:minor device numbers with integer values may look like the following. # cat cgroup-example-interface-file default 150 8:0 300 The default value can be updated by # echo 125 > cgroup-example-interface-file or # echo "default 125" > cgroup-example-interface-file An override can be set by # echo "8:16 170" > cgroup-example-interface-file and cleared by # echo "8:0 default" > cgroup-example-interface-file # cat cgroup-example-interface-file default 125 8:16 170 - For events which are not very high frequency, an interface file "events" should be created which lists event key value pairs. Whenever a notifiable event happens, file modified event should be generated on the file. 4-3. Core Interface Files All cgroup core files are prefixed with "cgroup." cgroup.procs A read-write new-line separated values file which exists on all cgroups. When read, it lists the PIDs of all processes which belong to the cgroup one-per-line. The PIDs are not ordered and the same PID may show up more than once if the process got moved to another cgroup and then back or the PID got recycled while reading. A PID can be written to migrate the process associated with the PID to the cgroup. The writer should match all of the following conditions. - Its euid is either root or must match either uid or suid of the target process. - It must have write access to the "cgroup.procs" file. - It must have write access to the "cgroup.procs" file of the common ancestor of the source and destination cgroups. When delegating a sub-hierarchy, write access to this file should be granted along with the containing directory. cgroup.controllers A read-only space separated values file which exists on all cgroups. It shows space separated list of all controllers available to the cgroup. The controllers are not ordered. cgroup.subtree_control A read-write space separated values file which exists on all cgroups. Starts out empty. When read, it shows space separated list of the controllers which are enabled to control resource distribution from the cgroup to its children. Space separated list of controllers prefixed with '+' or '-' can be written to enable or disable controllers. A controller name prefixed with '+' enables the controller and '-' disables. If a controller appears more than once on the list, the last one is effective. When multiple enable and disable operations are specified, either all succeed or all fail. cgroup.events A read-only flat-keyed file which exists on non-root cgroups. The following entries are defined. Unless specified otherwise, a value change in this file generates a file modified event. populated 1 if the cgroup or its descendants contains any live processes; otherwise, 0. 5. Controllers 5-1. CPU [NOTE: The interface for the cpu controller hasn't been merged yet] The "cpu" controllers regulates distribution of CPU cycles. This controller implements weight and absolute bandwidth limit models for normal scheduling policy and absolute bandwidth allocation model for realtime scheduling policy. 5-1-1. CPU Interface Files All time durations are in microseconds. cpu.stat A read-only flat-keyed file which exists on non-root cgroups. It reports the following six stats. usage_usec user_usec system_usec nr_periods nr_throttled throttled_usec cpu.weight A read-write single value file which exists on non-root cgroups. The default is "100". The weight in the range [1, 10000]. cpu.max A read-write two value file which exists on non-root cgroups. The default is "max 100000". The maximum bandwidth limit. It's in the following format. $MAX $PERIOD which indicates that the group may consume upto $MAX in each $PERIOD duration. "max" for $MAX indicates no limit. If only one number is written, $MAX is updated. cpu.rt.max [NOTE: The semantics of this file is still under discussion and the interface hasn't been merged yet] A read-write two value file which exists on all cgroups. The default is "0 100000". The maximum realtime runtime allocation. Over-committing configurations are disallowed and process migrations are rejected if not enough bandwidth is available. It's in the following format. $MAX $PERIOD which indicates that the group may consume upto $MAX in each $PERIOD duration. If only one number is written, $MAX is updated. 5-2. Memory The "memory" controller regulates distribution of memory. Memory is stateful and implements both limit and protection models. Due to the intertwining between memory usage and reclaim pressure and the stateful nature of memory, the distribution model is relatively complex. While not completely water-tight, all major memory usages by a given cgroup are tracked so that the total memory consumption can be accounted and controlled to a reasonable extent. Currently, the following types of memory usages are tracked. - Userland memory - page cache and anonymous memory. - Kernel data structures such as dentries and inodes. - TCP socket buffers. The above list may expand in the future for better coverage. 5-2-1. Memory Interface Files All memory amounts are in bytes. If a value which is not aligned to PAGE_SIZE is written, the value may be rounded up to the closest PAGE_SIZE multiple when read back. memory.current A read-only single value file which exists on non-root cgroups. The total amount of memory currently being used by the cgroup and its descendants. memory.low A read-write single value file which exists on non-root cgroups. The default is "0". Best-effort memory protection. If the memory usages of a cgroup and all its ancestors are below their low boundaries, the cgroup's memory won't be reclaimed unless memory can be reclaimed from unprotected cgroups. Putting more memory than generally available under this protection is discouraged. memory.high A read-write single value file which exists on non-root cgroups. The default is "max". Memory usage throttle limit. This is the main mechanism to control memory usage of a cgroup. If a cgroup's usage goes over the high boundary, the processes of the cgroup are throttled and put under heavy reclaim pressure. Going over the high limit never invokes the OOM killer and under extreme conditions the limit may be breached. memory.max A read-write single value file which exists on non-root cgroups. The default is "max". Memory usage hard limit. This is the final protection mechanism. If a cgroup's memory usage reaches this limit and can't be reduced, the OOM killer is invoked in the cgroup. Under certain circumstances, the usage may go over the limit temporarily. This is the ultimate protection mechanism. As long as the high limit is used and monitored properly, this limit's utility is limited to providing the final safety net. memory.events A read-only flat-keyed file which exists on non-root cgroups. The following entries are defined. Unless specified otherwise, a value change in this file generates a file modified event. low The number of times the cgroup is reclaimed due to high memory pressure even though its usage is under the low boundary. This usually indicates that the low boundary is over-committed. high The number of times processes of the cgroup are throttled and routed to perform direct memory reclaim because the high memory boundary was exceeded. For a cgroup whose memory usage is capped by the high limit rather than global memory pressure, this event's occurrences are expected. max The number of times the cgroup's memory usage was about to go over the max boundary. If direct reclaim fails to bring it down, the OOM killer is invoked. oom The number of times the OOM killer has been invoked in the cgroup. This may not exactly match the number of processes killed but should generally be close. memory.stat A read-only flat-keyed file which exists on non-root cgroups. This breaks down the cgroup's memory footprint into different types of memory, type-specific details, and other information on the state and past events of the memory management system. All memory amounts are in bytes. The entries are ordered to be human readable, and new entries can show up in the middle. Don't rely on items remaining in a fixed position; use the keys to look up specific values! anon Amount of memory used in anonymous mappings such as brk(), sbrk(), and mmap(MAP_ANONYMOUS) file Amount of memory used to cache filesystem data, including tmpfs and shared memory. file_mapped Amount of cached filesystem data mapped with mmap() file_dirty Amount of cached filesystem data that was modified but not yet written back to disk file_writeback Amount of cached filesystem data that was modified and is currently being written back to disk inactive_anon active_anon inactive_file active_file unevictable Amount of memory, swap-backed and filesystem-backed, on the internal memory management lists used by the page reclaim algorithm pgfault Total number of page faults incurred pgmajfault Number of major page faults incurred memory.swap.current A read-only single value file which exists on non-root cgroups. The total amount of swap currently being used by the cgroup and its descendants. memory.swap.max A read-write single value file which exists on non-root cgroups. The default is "max". Swap usage hard limit. If a cgroup's swap usage reaches this limit, anonymous meomry of the cgroup will not be swapped out. 5-2-2. General Usage "memory.high" is the main mechanism to control memory usage. Over-committing on high limit (sum of high limits > available memory) and letting global memory pressure to distribute memory according to usage is a viable strategy. Because breach of the high limit doesn't trigger the OOM killer but throttles the offending cgroup, a management agent has ample opportunities to monitor and take appropriate actions such as granting more memory or terminating the workload. Determining whether a cgroup has enough memory is not trivial as memory usage doesn't indicate whether the workload can benefit from more memory. For example, a workload which writes data received from network to a file can use all available memory but can also operate as performant with a small amount of memory. A measure of memory pressure - how much the workload is being impacted due to lack of memory - is necessary to determine whether a workload needs more memory; unfortunately, memory pressure monitoring mechanism isn't implemented yet. 5-2-3. Memory Ownership A memory area is charged to the cgroup which instantiated it and stays charged to the cgroup until the area is released. Migrating a process to a different cgroup doesn't move the memory usages that it instantiated while in the previous cgroup to the new cgroup. A memory area may be used by processes belonging to different cgroups. To which cgroup the area will be charged is in-deterministic; however, over time, the memory area is likely to end up in a cgroup which has enough memory allowance to avoid high reclaim pressure. If a cgroup sweeps a considerable amount of memory which is expected to be accessed repeatedly by other cgroups, it may make sense to use POSIX_FADV_DONTNEED to relinquish the ownership of memory areas belonging to the affected files to ensure correct memory ownership. 5-3. IO The "io" controller regulates the distribution of IO resources. This controller implements both weight based and absolute bandwidth or IOPS limit distribution; however, weight based distribution is available only if cfq-iosched is in use and neither scheme is available for blk-mq devices. 5-3-1. IO Interface Files io.stat A read-only nested-keyed file which exists on non-root cgroups. Lines are keyed by $MAJ:$MIN device numbers and not ordered. The following nested keys are defined. rbytes Bytes read wbytes Bytes written rios Number of read IOs wios Number of write IOs An example read output follows. 8:16 rbytes=1459200 wbytes=314773504 rios=192 wios=353 8:0 rbytes=90430464 wbytes=299008000 rios=8950 wios=1252 io.weight A read-write flat-keyed file which exists on non-root cgroups. The default is "default 100". The first line is the default weight applied to devices without specific override. The rest are overrides keyed by $MAJ:$MIN device numbers and not ordered. The weights are in the range [1, 10000] and specifies the relative amount IO time the cgroup can use in relation to its siblings. The default weight can be updated by writing either "default $WEIGHT" or simply "$WEIGHT". Overrides can be set by writing "$MAJ:$MIN $WEIGHT" and unset by writing "$MAJ:$MIN default". An example read output follows. default 100 8:16 200 8:0 50 io.max A read-write nested-keyed file which exists on non-root cgroups. BPS and IOPS based IO limit. Lines are keyed by $MAJ:$MIN device numbers and not ordered. The following nested keys are defined. rbps Max read bytes per second wbps Max write bytes per second riops Max read IO operations per second wiops Max write IO operations per second When writing, any number of nested key-value pairs can be specified in any order. "max" can be specified as the value to remove a specific limit. If the same key is specified multiple times, the outcome is undefined. BPS and IOPS are measured in each IO direction and IOs are delayed if limit is reached. Temporary bursts are allowed. Setting read limit at 2M BPS and write at 120 IOPS for 8:16. echo "8:16 rbps=2097152 wiops=120" > io.max Reading returns the following. 8:16 rbps=2097152 wbps=max riops=max wiops=120 Write IOPS limit can be removed by writing the following. echo "8:16 wiops=max" > io.max Reading now returns the following. 8:16 rbps=2097152 wbps=max riops=max wiops=max 5-3-2. Writeback Page cache is dirtied through buffered writes and shared mmaps and written asynchronously to the backing filesystem by the writeback mechanism. Writeback sits between the memory and IO domains and regulates the proportion of dirty memory by balancing dirtying and write IOs. The io controller, in conjunction with the memory controller, implements control of page cache writeback IOs. The memory controller defines the memory domain that dirty memory ratio is calculated and maintained for and the io controller defines the io domain which writes out dirty pages for the memory domain. Both system-wide and per-cgroup dirty memory states are examined and the more restrictive of the two is enforced. cgroup writeback requires explicit support from the underlying filesystem. Currently, cgroup writeback is implemented on ext2, ext4 and btrfs. On other filesystems, all writeback IOs are attributed to the root cgroup. There are inherent differences in memory and writeback management which affects how cgroup ownership is tracked. Memory is tracked per page while writeback per inode. For the purpose of writeback, an inode is assigned to a cgroup and all IO requests to write dirty pages from the inode are attributed to that cgroup. As cgroup ownership for memory is tracked per page, there can be pages which are associated with different cgroups than the one the inode is associated with. These are called foreign pages. The writeback constantly keeps track of foreign pages and, if a particular foreign cgroup becomes the majority over a certain period of time, switches the ownership of the inode to that cgroup. While this model is enough for most use cases where a given inode is mostly dirtied by a single cgroup even when the main writing cgroup changes over time, use cases where multiple cgroups write to a single inode simultaneously are not supported well. In such circumstances, a significant portion of IOs are likely to be attributed incorrectly. As memory controller assigns page ownership on the first use and doesn't update it until the page is released, even if writeback strictly follows page ownership, multiple cgroups dirtying overlapping areas wouldn't work as expected. It's recommended to avoid such usage patterns. The sysctl knobs which affect writeback behavior are applied to cgroup writeback as follows. vm.dirty_background_ratio vm.dirty_ratio These ratios apply the same to cgroup writeback with the amount of available memory capped by limits imposed by the memory controller and system-wide clean memory. vm.dirty_background_bytes vm.dirty_bytes For cgroup writeback, this is calculated into ratio against total available memory and applied the same way as vm.dirty[_background]_ratio. P. Information on Kernel Programming This section contains kernel programming information in the areas where interacting with cgroup is necessary. cgroup core and controllers are not covered. P-1. Filesystem Support for Writeback A filesystem can support cgroup writeback by updating address_space_operations->writepage[s]() to annotate bio's using the following two functions. wbc_init_bio(@wbc, @bio) Should be called for each bio carrying writeback data and associates the bio with the inode's owner cgroup. Can be called anytime between bio allocation and submission. wbc_account_io(@wbc, @page, @bytes) Should be called for each data segment being written out. While this function doesn't care exactly when it's called during the writeback session, it's the easiest and most natural to call it as data segments are added to a bio. With writeback bio's annotated, cgroup support can be enabled per super_block by setting SB_I_CGROUPWB in ->s_iflags. This allows for selective disabling of cgroup writeback support which is helpful when certain filesystem features, e.g. journaled data mode, are incompatible. wbc_init_bio() binds the specified bio to its cgroup. Depending on the configuration, the bio may be executed at a lower priority and if the writeback session is holding shared resources, e.g. a journal entry, may lead to priority inversion. There is no one easy solution for the problem. Filesystems can try to work around specific problem cases by skipping wbc_init_bio() or using bio_associate_blkcg() directly. D. Deprecated v1 Core Features - Multiple hierarchies including named ones are not supported. - All mount options and remounting are not supported. - The "tasks" file is removed and "cgroup.procs" is not sorted. - "cgroup.clone_children" is removed. - /proc/cgroups is meaningless for v2. Use "cgroup.controllers" file at the root instead. R. Issues with v1 and Rationales for v2 R-1. Multiple Hierarchies cgroup v1 allowed an arbitrary number of hierarchies and each hierarchy could host any number of controllers. While this seemed to provide a high level of flexibility, it wasn't useful in practice. For example, as there is only one instance of each controller, utility type controllers such as freezer which can be useful in all hierarchies could only be used in one. The issue is exacerbated by the fact that controllers couldn't be moved to another hierarchy once hierarchies were populated. Another issue was that all controllers bound to a hierarchy were forced to have exactly the same view of the hierarchy. It wasn't possible to vary the granularity depending on the specific controller. In practice, these issues heavily limited which controllers could be put on the same hierarchy and most configurations resorted to putting each controller on its own hierarchy. Only closely related ones, such as the cpu and cpuacct controllers, made sense to be put on the same hierarchy. This often meant that userland ended up managing multiple similar hierarchies repeating the same steps on each hierarchy whenever a hierarchy management operation was necessary. Furthermore, support for multiple hierarchies came at a steep cost. It greatly complicated cgroup core implementation but more importantly the support for multiple hierarchies restricted how cgroup could be used in general and what controllers was able to do. There was no limit on how many hierarchies there might be, which meant that a thread's cgroup membership couldn't be described in finite length. The key might contain any number of entries and was unlimited in length, which made it highly awkward to manipulate and led to addition of controllers which existed only to identify membership, which in turn exacerbated the original problem of proliferating number of hierarchies. Also, as a controller couldn't have any expectation regarding the topologies of hierarchies other controllers might be on, each controller had to assume that all other controllers were attached to completely orthogonal hierarchies. This made it impossible, or at least very cumbersome, for controllers to cooperate with each other. In most use cases, putting controllers on hierarchies which are completely orthogonal to each other isn't necessary. What usually is called for is the ability to have differing levels of granularity depending on the specific controller. In other words, hierarchy may be collapsed from leaf towards root when viewed from specific controllers. For example, a given configuration might not care about how memory is distributed beyond a certain level while still wanting to control how CPU cycles are distributed. R-2. Thread Granularity cgroup v1 allowed threads of a process to belong to different cgroups. This didn't make sense for some controllers and those controllers ended up implementing different ways to ignore such situations but much more importantly it blurred the line between API exposed to individual applications and system management interface. Generally, in-process knowledge is available only to the process itself; thus, unlike service-level organization of processes, categorizing threads of a process requires active participation from the application which owns the target process. cgroup v1 had an ambiguously defined delegation model which got abused in combination with thread granularity. cgroups were delegated to individual applications so that they can create and manage their own sub-hierarchies and control resource distributions along them. This effectively raised cgroup to the status of a syscall-like API exposed to lay programs. First of all, cgroup has a fundamentally inadequate interface to be exposed this way. For a process to access its own knobs, it has to extract the path on the target hierarchy from /proc/self/cgroup, construct the path by appending the name of the knob to the path, open and then read and/or write to it. This is not only extremely clunky and unusual but also inherently racy. There is no conventional way to define transaction across the required steps and nothing can guarantee that the process would actually be operating on its own sub-hierarchy. cgroup controllers implemented a number of knobs which would never be accepted as public APIs because they were just adding control knobs to system-management pseudo filesystem. cgroup ended up with interface knobs which were not properly abstracted or refined and directly revealed kernel internal details. These knobs got exposed to individual applications through the ill-defined delegation mechanism effectively abusing cgroup as a shortcut to implementing public APIs without going through the required scrutiny. This was painful for both userland and kernel. Userland ended up with misbehaving and poorly abstracted interfaces and kernel exposing and locked into constructs inadvertently. R-3. Competition Between Inner Nodes and Threads cgroup v1 allowed threads to be in any cgroups which created an interesting problem where threads belonging to a parent cgroup and its children cgroups competed for resources. This was nasty as two different types of entities competed and there was no obvious way to settle it. Different controllers did different things. The cpu controller considered threads and cgroups as equivalents and mapped nice levels to cgroup weights. This worked for some cases but fell flat when children wanted to be allocated specific ratios of CPU cycles and the number of internal threads fluctuated - the ratios constantly changed as the number of competing entities fluctuated. There also were other issues. The mapping from nice level to weight wasn't obvious or universal, and there were various other knobs which simply weren't available for threads. The io controller implicitly created a hidden leaf node for each cgroup to host the threads. The hidden leaf had its own copies of all the knobs with "leaf_" prefixed. While this allowed equivalent control over internal threads, it was with serious drawbacks. It always added an extra layer of nesting which wouldn't be necessary otherwise, made the interface messy and significantly complicated the implementation. The memory controller didn't have a way to control what happened between internal tasks and child cgroups and the behavior was not clearly defined. There were attempts to add ad-hoc behaviors and knobs to tailor the behavior to specific workloads which would have led to problems extremely difficult to resolve in the long term. Multiple controllers struggled with internal tasks and came up with different ways to deal with it; unfortunately, all the approaches were severely flawed and, furthermore, the widely different behaviors made cgroup as a whole highly inconsistent. This clearly is a problem which needs to be addressed from cgroup core in a uniform way. R-4. Other Interface Issues cgroup v1 grew without oversight and developed a large number of idiosyncrasies and inconsistencies. One issue on the cgroup core side was how an empty cgroup was notified - a userland helper binary was forked and executed for each event. The event delivery wasn't recursive or delegatable. The limitations of the mechanism also led to in-kernel event delivery filtering mechanism further complicating the interface. Controller interfaces were problematic too. An extreme example is controllers completely ignoring hierarchical organization and treating all cgroups as if they were all located directly under the root cgroup. Some controllers exposed a large amount of inconsistent implementation details to userland. There also was no consistency across controllers. When a new cgroup was created, some controllers defaulted to not imposing extra restrictions while others disallowed any resource usage until explicitly configured. Configuration knobs for the same type of control used widely differing naming schemes and formats. Statistics and information knobs were named arbitrarily and used different formats and units even in the same controller. cgroup v2 establishes common conventions where appropriate and updates controllers so that they expose minimal and consistent interfaces. R-5. Controller Issues and Remedies R-5-1. Memory The original lower boundary, the soft limit, is defined as a limit that is per default unset. As a result, the set of cgroups that global reclaim prefers is opt-in, rather than opt-out. The costs for optimizing these mostly negative lookups are so high that the implementation, despite its enormous size, does not even provide the basic desirable behavior. First off, the soft limit has no hierarchical meaning. All configured groups are organized in a global rbtree and treated like equal peers, regardless where they are located in the hierarchy. This makes subtree delegation impossible. Second, the soft limit reclaim pass is so aggressive that it not just introduces high allocation latencies into the system, but also impacts system performance due to overreclaim, to the point where the feature becomes self-defeating. The memory.low boundary on the other hand is a top-down allocated reserve. A cgroup enjoys reclaim protection when it and all its ancestors are below their low boundaries, which makes delegation of subtrees possible. Secondly, new cgroups have no reserve per default and in the common case most cgroups are eligible for the preferred reclaim pass. This allows the new low boundary to be efficiently implemented with just a minor addition to the generic reclaim code, without the need for out-of-band data structures and reclaim passes. Because the generic reclaim code considers all cgroups except for the ones running low in the preferred first reclaim pass, overreclaim of individual groups is eliminated as well, resulting in much better overall workload performance. The original high boundary, the hard limit, is defined as a strict limit that can not budge, even if the OOM killer has to be called. But this generally goes against the goal of making the most out of the available memory. The memory consumption of workloads varies during runtime, and that requires users to overcommit. But doing that with a strict upper limit requires either a fairly accurate prediction of the working set size or adding slack to the limit. Since working set size estimation is hard and error prone, and getting it wrong results in OOM kills, most users tend to err on the side of a looser limit and end up wasting precious resources. The memory.high boundary on the other hand can be set much more conservatively. When hit, it throttles allocations by forcing them into direct reclaim to work off the excess, but it never invokes the OOM killer. As a result, a high boundary that is chosen too aggressively will not terminate the processes, but instead it will lead to gradual performance degradation. The user can monitor this and make corrections until the minimal memory footprint that still gives acceptable performance is found. In extreme cases, with many concurrent allocations and a complete breakdown of reclaim progress within the group, the high boundary can be exceeded. But even then it's mostly better to satisfy the allocation from the slack available in other groups or the rest of the system than killing the group. Otherwise, memory.max is there to limit this type of spillover and ultimately contain buggy or even malicious applications. The combined memory+swap accounting and limiting is replaced by real control over swap space. The main argument for a combined memory+swap facility in the original cgroup design was that global or parental pressure would always be able to swap all anonymous memory of a child group, regardless of the child's own (possibly untrusted) configuration. However, untrusted groups can sabotage swapping by other means - such as referencing its anonymous memory in a tight loop - and an admin can not assume full swappability when overcommitting untrusted jobs. For trusted jobs, on the other hand, a combined counter is not an intuitive userspace interface, and it flies in the face of the idea that cgroup controllers should account and limit specific physical resources. Swap space is a resource like all others in the system, and that's why unified hierarchy allows distributing it separately.