# Driver `task.acme` **Minimal configlet:** [task#1] type = acme **Minimal setup command:** om test/vol/foo set --kw="type=acme" **Supported keywords:** - comment - pg_blkio_weight - pg_cpu_burst - pg_cpu_quota - pg_cpu_shares - pg_cpus - pg_mem_high - pg_mem_limit - pg_mem_oom_control - pg_mem_swappiness - pg_mems - pg_pids_max - pg_vmem_limit - stat_timeout - type ## Keyword `comment` required: false scopable: false **Description:** A free form text describing the role of the object, of the node, or of the section it is set in. The keyword is accepted in any section, so the DEFAULT section can document a configuration as a whole, and a resource, pool, heartbeat, array or network section can document itself. The agent does not interpret the value. ## Keyword `pg_blkio_weight` required: false scopable: true **Example:** pg_blkio_weight=50 **Description:** Block IO relative weight. Value: between `10` and `1000`. The kernel default is `100`. Setting this keyword to `default` puts the capping back where a node that never capped anything leaves it. Removing the keyword does not: what was written stays written, whether om wrote it or something else did. ## Keyword `pg_cpu_burst` required: false scopable: true **Example:** pg_cpu_burst=20% **Description:** The cpu time the group banks under its quota, and spends above it. It is the time the processes of the group may bank while under their `pg_cpu_quota`, and spend above it in a short spike, in the notation of `pg_cpu_quota`: `20%` lets the group run a fifth of a cpu above its quota during a period, out of what it left unused before. It serves a latency-sensitive service capped tightly: a request arriving after an idle spell is served at the speed of the burst, rather than throttled for the rest of the period. Over a longer time the group still uses no more than its quota. It needs `pg_cpu_quota`, and cannot exceed it: the kernel refuses a burst larger than the quota. Only the unified cgroup hierarchy has it. Setting this keyword to `default` puts the capping back where a node that never capped anything leaves it. Removing the keyword does not: what was written stays written, whether om wrote it or something else did. ## Keyword `pg_cpu_quota` required: false scopable: true **Example:** pg_cpu_quota=50%@all **Description:** The percentage of cpu time the processes of the group may use. Accepted expressions: * `50%` Half of one cpu. * `50%@all` Half of every cpu the node has. * `10%@2` A tenth of two cpus. Unlike `pg_cpu_shares`, which only arbitrates when the node is cpu-bound, a quota caps the group whether or not the node is busy. Setting this keyword to `default` puts the capping back where a node that never capped anything leaves it. Removing the keyword does not: what was written stays written, whether om wrote it or something else did. ## Keyword `pg_cpu_shares` required: false scopable: true convert: size **Example:** pg_cpu_shares=512 **Description:** The share of cpu the group gets when the node is cpu-bound. The share is relative to the other groups. The actual percentile depends on the shares allowed to the other services. Unset, the kernel default is used, which usually is 1024 shares. Setting this keyword to `default` puts the capping back where a node that never capped anything leaves it. Removing the keyword does not: what was written stays written, whether om wrote it or something else did. ## Keyword `pg_cpus` required: false scopable: true depends: create_pg=true **Example:** pg_cpus=0-2 **Description:** The cpus the processes of the group may run on. Cpus are specified as list or range : `0,1,2` or `0-2`. Setting this keyword to `default` puts the capping back where a node that never capped anything leaves it. Removing the keyword does not: what was written stays written, whether om wrote it or something else did. ## Keyword `pg_mem_high` required: false scopable: true convert: size **Example:** pg_mem_high=384m **Description:** The memory past which the kernel throttles the group, in bytes. Past it, the kernel slows the processes of the group down, reclaiming their memory and throttling their allocations. Unlike `pg_mem_limit`, reaching it does not wake the Out-Of-Memory killer: a group above it keeps running, slower, while the kernel reclaims. Set below `pg_mem_limit`, it gives a process growing too much the time to be noticed, or to shrink, before it is killed. It is a soft limit, so it is not what a memory claim counts: the claim counts `pg_mem_limit`. Only the unified cgroup hierarchy has it. Setting this keyword to `default` puts the capping back where a node that never capped anything leaves it. Removing the keyword does not: what was written stays written, whether om wrote it or something else did. ## Keyword `pg_mem_limit` required: false scopable: true convert: size **Example:** pg_mem_limit=512m **Description:** The memory the processes of the group may use, in bytes. The Out-Of-Memory killer is triggered in case of tresspassing. Setting this keyword to `default` puts the capping back where a node that never capped anything leaves it. Removing the keyword does not: what was written stays written, whether om wrote it or something else did. ## Keyword `pg_mem_oom_control` required: false scopable: true **Example:** pg_mem_oom_control=1 **Description:** Whether the Out of Memory killer runs for the group: 0 enables it. It is a flag (0 or 1) enabling or disabling the Out of Memory killer for the processes of the group. * If enabled (0), tasks that attempt to consume more memory than they are allowed are immediately killed by the OOM killer. * If disabled (1), tasks are allowed to continue to try allocating memory, stressing the system. The OOM killer is enabled by default in every cgroup using the memory controller. This keyword caps nothing on a node holding the unified cgroup hierarchy, which has no `memory.oom_control`: that file is of the v1 hierarchy. Setting it there is warned about, and ignored. ## Keyword `pg_mem_swappiness` required: false scopable: true **Example:** pg_mem_swappiness=40 **Description:** Set a swappiness percentile value for the process group. This keyword caps nothing on a node holding the unified cgroup hierarchy, which has no `memory.swappiness`: that file is of the v1 hierarchy. Setting it there is warned about, and ignored. ## Keyword `pg_mems` required: false scopable: true **Example:** pg_mems=0-2 **Description:** The memory nodes the processes of the group may allocate from. Memory nodes are specified as list or range : `0,1,2` or `0-2`. Setting this keyword to `default` puts the capping back where a node that never capped anything leaves it. Removing the keyword does not: what was written stays written, whether om wrote it or something else did. ## Keyword `pg_pids_max` required: false scopable: true **Example:** pg_pids_max=512 **Description:** The most processes and threads the group may run at once. A fork beyond it fails in the process that asked for it, and nothing else of the node is touched: it is what keeps a runaway fork loop from exhausting the process table of the node. Setting this keyword to `default` puts the capping back where a node that never capped anything leaves it. Removing the keyword does not: what was written stays written, whether om wrote it or something else did. ## Keyword `pg_vmem_limit` required: false scopable: true convert: size **Example:** pg_vmem_limit=1g **Description:** The memory plus swap the processes of the group may use, in bytes. The Out-Of-Memory killer is triggered in case of tresspassing. The specified value must be greater than `pg_mem_limit`. Setting this keyword to `default` puts the capping back where a node that never capped anything leaves it. Removing the keyword does not: what was written stays written, whether om wrote it or something else did. ## Keyword `stat_timeout` required: false scopable: true convert: duration **Description:** The fs resources status evaluation includes a stat syscall test. This keyword defines the maximum wait time for those stat calls to respond. When expired, the resource status is degraded is to warn, which can trigger a monitor action (reboot or crash the node) if the resource is monitored. ## Keyword `type` required: false scopable: false rbac: Requires the root grant, except for the values oci, docker, podman, acme. **Description:** The resource driver name.