Driver `ip.netns`

Minimal configlet:

[ip#1]
type = netns
mode = bridge
name = 1.2.3.4
netmask = 24
netns = container#0

Minimal setup command:

om test/svc/foo set \
	--kw="type=netns" \
	--kw="mode=bridge" \
	--kw="name=1.2.3.4" \
	--kw="netmask=24" \
	--kw="netns=container#0"

Supported keywords:

addr
alias
blocking_post_provision
blocking_post_start
blocking_post_stop
blocking_post_unprovision
blocking_pre_provision
blocking_pre_start
blocking_pre_stop
blocking_pre_unprovision
check_carrier
comment
del_net_route
dev
disable
dns_name_suffix
encap
expose
gateway
macaddr
mode
monitor
name
netmask
netns
network
nsdev
optional
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
post_provision
post_start
post_stop
post_unprovision
pre_provision
pre_start
pre_stop
pre_unprovision
provision
provision_requires
restart
restart_delay
shared
standby
start_requires
stat_timeout
stop_requires
subset
tags
type
unprovision
unprovision_requires
vlan_mode
vlan_tag
wait_dns

Keyword `addr`

required:    false
scopable:    true
recorded:    written when what it names is made, and reset when the object is cloned
stop first:  a change is refused while the resource is up
rbac:        An address of the user's choosing requires the root grant: om writes the address it draws from the network.

Example:

addr=fd01:2345:6789:2902::5:e8

Description:

The address drawn from the lan network the network keyword names, which om writes here when the resource first starts, and removes when it is unprovisioned.

On a lan network the resource keeps its address from a start to the next, until it is unprovisioned, as an ip.host resource does: the hosts of the segment know the service by its address. Every node holds the configuration, so every node uses this address for the resource, whichever node drew it and whatever became of that node. Set it to keep an address drawn before, or unset it to have the next start draw another one, and remove the one it replaces from the namespace.

On the other networks the address is drawn at each start and released at each stop, and the clients reach the service by its DNS record, which follows the address: this keyword is not used.

A configuration copied to create another object leaves it behind, so the copy draws an address of its own rather than bring this one up twice.

A resource not shared of a flex object draws an address per instance, as its instances run at once. The address of an instance is not written here: the node keeps it in its reservation store, from a start to the next, until the instance is unprovisioned. Set addr@ to choose the address of an instance; an addr without a node is not taken as one.

Keyword `alias`

required:    false
scopable:    true
default:     true
convert:     bool
rbac:        Requires the root grant.

Description:

Use network interface stacking.

Modern ip stack support multiple ip addresses per interface, so alias should be set to false when possible.

Keyword `blocking_post_provision`

required:    false
scopable:    true
rbac:        Triggers require the root grant.

Description:

A command or script to execute after the resource provision action.

Errors interrupt the action.

This trigger is only executed on leaders.

Keyword `blocking_post_start`

required:    false
scopable:    true
rbac:        Triggers require the root grant.

Description:

A command or script to execute after the resource start action.

Errors interrupt the action.

Keyword `blocking_post_stop`

required:    false
scopable:    true
rbac:        Triggers require the root grant.

Description:

A command or script to execute after the resource stop action.

Errors interrupt the action.

Keyword `blocking_post_unprovision`

required:    false
scopable:    true
rbac:        Triggers require the root grant.

Description:

A command or script to execute after the resource unprovision action.

Errors interrupt the action.

This trigger is only executed on leaders.

Keyword `blocking_pre_provision`

required:    false
scopable:    true
rbac:        Triggers require the root grant.

Description:

A command or script to execute before the resource provision action.

Errors interrupt the action.

Keyword `blocking_pre_start`

required:    false
scopable:    true
rbac:        Triggers require the root grant.

Description:

A command or script to execute before the resource start action.

Errors interrupt the action.

Keyword `blocking_pre_stop`

required:    false
scopable:    true
rbac:        Triggers require the root grant.

Description:

A command or script to execute before the resource stop action.

Errors interrupt the action.

Keyword `blocking_pre_unprovision`

required:    false
scopable:    true
rbac:        Triggers require the root grant.

Description:

A command or script to execute before the resource unprovision action.

Errors interrupt the action.

Keyword `check_carrier`

required:    false
scopable:    true
default:     true
convert:     bool
rbac:        Requires the root grant.

Description:

Activate the link carrier check.

Set to false if dev is a backend bridge or switch.

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 `del_net_route`

required:    false
scopable:    true
default:     false
convert:     bool
rbac:        Requires the root grant.

Description:

Some docker ip configurations require dropping the network route autoconfigured when installing the ip address.

In this case set del_net_route=true and network=<cidr>.

Keyword `dev`

required:    false
scopable:    true
aliases:     ipdev
stop first:  a change is refused while the resource is up
rbac:        Requires the root grant.

Example:

dev=br-prd

Description:

The interface name to setup the ip address on.

This interface can be different from one node to the other, in which case the dev@<nodename> scoping syntax can be used.

If the value is expressed as <intf>:<n>, the stacked interface index is forced to <n> instead of the lowest free index.

If the value is expressed as <name>@<intf>, a macvtap interface named <name> is created and attached to <intf>.

Keyword `disable`

required:    false
scopable:    true
convert:     bool

Description:

A disabled resource will be ignored on start, stop, provision and unprovision actions.

A disabled resource status is n/a.

If set in the DEFAULT section of an object, the object is disabled and ignores start, stop, shutdown, provision and unprovision actions.

These actions immediately return success.

om <path> disable sets DEFAULT.disable=true.

om <path> enable sets DEFAULT.disable=false.

Note: The enable and disable actions preserve the individual resource disable state.

Keyword `dns_name_suffix`

required:    false
scopable:    true

Example:

dns_name_suffix=-backup

Description:

Add the value as a suffix to the DNS record name. The record created is thus formatted as:

<name>-<dns_name_suffix>.<app>.<managed zone> in the collector managed zone
<name>-<dns_name_suffix>.<namespace>.<kind>.<clustername> in the cluster dns zone.

Keyword `encap`

required:    false
scopable:    false
convert:     bool

Description:

Set to true to ignore this resource in the nodes context and consider it in the encapnodes context. The resource is thus handled by agents deployed in the service containers.

Keyword `expose`

required:    false
scopable:    true
convert:     list

Example:

expose=443/tcp:8443 53/udp

Description:

A whitespace-separated list of <port>/<protocol>[:<host port>] describing socket services that mandate a SRV exposition.

Keyword `gateway`

required:    false
scopable:    true
provision:   used only by the provision of the resource
rbac:        Requires the root grant.

Description:

The gateway the default route of the network namespace goes through.

When not set, it is the gateway of the network the address is drawn from: the address of the bridge of a bridge or routed_bridge network, the gateway keyword of a lan network. With no gateway at all, the default route goes through the interface of the resource.

Several ip resources may share a namespace, and a namespace has one default route. The resource setting it is the one whose gateway is set here, rather than taken from its network, rather than none; the lowest resource id among equals, whatever the order the resources start in. The others leave the route alone, and a gateway set here that is not used is reported in the status.

The default route is set only when the namespace has none: one another resource of the container set, as an ip.cni one, is kept.

Keyword `macaddr`

required:    false
scopable:    true
rbac:        Requires the root grant.

Example:

macaddr=ce:32:cc:ca:41:33

Description:

The hardware address to set on the interface in the network namespace.

Keyword `mode`

required:    false
scopable:    true
stop first:  a change is refused while the resource is up
candidates:  bridge, dedicated, macvlan, ipvlan-l2, ipvlan-l3, ipvlan-l3s, ovs
rbac:        Requires the root grant.

Default:

The first of the modes of the network the address is drawn from, else bridge.

Example:

mode=bridge

Description:

The ip link mode.

bridge

Proven, simple and versatile default mode.
dedicated

Used when a container or namespace gets exclusive ownership of a physical NIC. Ideal for performance-sensitive workloads that need direct hardware access without virtualization overhead.
macvlan

Used when each container needs its own MAC address and appears as a distinct device on the physical network (L2 presence). Common in environments where containers must be directly reachable on the LAN, like VMs would be. This mode is not appropiate on VMWare with Vswitch default policies.
ipvlan-l2

Similar to macvlan but all sub-interfaces share the parent’s MAC address. Used when the upstream switch limits the number of MAC addresses per port (e.g., cloud environments or strict switch policies).
ipvlan-l3

Used when you want the host to act as a router between namespaces and the external network. No L2 broadcast/multicast is propagated; each namespace is isolated at Layer 2. Suited for high-density, routed environments.
ipvlan-l3s

A variant of ipvlan-l3 that keeps traffic in the kernel’s netfilter/iptables path (s = symmetric routing). Used when you still need NAT, firewalling, or policy routing rules to apply to ipvlan traffic, unlike pure l3 mode.
ovs (Open vSwitch)

Used in SDN (Software-Defined Networking) and multi-host environments. Chosen when you need advanced L2/L3 features like VLANs, tunneling (VXLAN, GRE), flow-based forwarding, or centralized network control.

Communication scope:

Mode	Host → Container	Container → Host	Container → Container
bridge	ok	ok	ok
dedicated
macvlan			ok
ipvlan-l2			ok
ipvlan-l3			ok
ipvlan-l3s	ok	ok	ok
ovs	ok	ok	ok

Keyword `monitor`

required:    false
scopable:    true
convert:     bool

Description:

A resource with monitor=true will trigger the monitor_action (crash or reboot the node, freezestop or switch the service) if:

The resource is down.
The instance has local_expect=started in its daemon monitor data, which means the daemon considers this instance is and should remain started.
All restart tentatives failed.

Keyword `name`

required:    false
scopable:    true
aliases:     ipname
stop first:  a change is refused while the resource is up
rbac:        Requires the root grant.

Example:

name=1.2.3.4

Description:

The DNS name or IP address of the ip resource.

Scoping

Can be different from one node to the other, in which case the name@<nodename> scoping syntax can be used.

This is most useful to specify a different ip when the service starts in DRP mode, where subnets are likely to be different than those of the production datacenter.

DNS name caching

If the value is a DNS name, a lookup is done for every resource status evaluation.

If the lookup succeeds, the addr is store in a cache file.
If the lookup fails, the address is fetched from the cache file.
- If the cache is not expired (refreshed in the past 19 minutes) the address is trusted:
  - The resource status is degraded to from up to warn.
  - The resource restarts begin if restarts > 0. The restart is a no-op but the resource status is re-evaluated. If the lookup succeeds, the resource status to up.
  - The monitor action is triggered if all restarts failed and monitor_action is set. For example, a freezestop can still succeed using the cached ip address.
- If the cache is expired (older than 19 minutes):
  - The resource status is warn, weither the actual untrusted ip address is plumbed or not.
  - The stop and start actions no longer work. An admin action is necessary to repair the configuration or dns lookups.

Keyword `netmask`

required:    false
scopable:    true
rbac:        Requires the root grant.

Example:

netmask=24

Description:

The netmask to configure with the address resolved from name.

If an ip is already plumbed on the root interface, the netmask default is the netmask of this existing ip.

netmask is mandatory for interfaces dedicated to the object. Dummy interfaces are likely to be in this case.

The format is:

dotted or octal for IPv4, ex: 255.255.252.0 or 22.
octal only for IPv6, ex: 64.

Keyword `netns`

required:    true
scopable:    true
aliases:     container_rid
stop first:  a change is refused while the resource is up
rbac:        The host network namespace requires the root grant.

Example:

netns=container#0

Description:

The resource id of the container to plumb the ip into.

Keyword `network`

required:    false
scopable:    true
stop first:  a change is refused while the resource is up
rbac:        A network named by a path requires the root grant.

Example:

network=default

Description:

The name of the om network the address is drawn from.

The default network exists on every node of a fresh installation, so an object needs nothing more than network=default to receive an address.

Naming a network is enough: the dev, netmask and gateway of the resource are the network’s to know, and are taken from it when the configuration does not set them. Leaving name empty has the address allocated from the network too.

On a bridge or routed_bridge network, the address is drawn at each start and released at each stop, and the clients reach the service by its DNS record, which follows the address.

On a lan network, the address is kept from a start to the next until the resource is unprovisioned, and recorded in addr. The dev is the interface of the node on the segment, and the netmask the prefix length of the segment. The gateway is the one of the network, the router of the segment, and none when the network names none. The macvlan and ipvlan-l2 modes put the address on the segment. The bridge mode needs dev to be a linux bridge on the segment, and the ovs mode needs dev set to an openvswitch bridge on the segment. The ipvlan-l3 and ipvlan-l3s modes are refused, as the hosts of the segment would not reach the address, and the dedicated mode needs a dev of its own, not the interface of the node on the segment.

This keyword used to hold the address of the network, in dotted notation, to set the destination of the route removed by del_net_route. That destination is derived from the address and the netmask now. A value in that form is reported as obsolete and ignored.

Keyword `nsdev`

required:    false
scopable:    true

Example:

nsdev=front

Description:

The first eth<n> available in the network namespace.

Keyword `optional`

required:    false
scopable:    true
convert:     bool

Description:

Action errors on optional resources are logged but do not interrupt the action sequence.

The status of optional resources is not included in the instance availability status but is considered in the overall status.

The status of task and sync resources is always included in the overall status, regardless of whether they are marked as optional.

Resources tagged as noaction are considered optional by default.

Dump filesystems are a typical use case for optional=true.

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.

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.

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.

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.

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.

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.

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.

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.

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.