Docker Swarm: Resource Reservation, Placement Constraints, and Placement Preferences

```console $ docker service create \ --name my_web \ --replicas 3 \ nginx ``` To start a global service on each available node, pass `--mode global` to `docker service create`. Every time a new node becomes available, the scheduler places a task for the global service on the new node. For example to start a service that runs alpine on every node in the swarm: ```console $ docker service create \ --name myservice \ --mode global \ alpine top ``` Service constraints let you set criteria for a node to meet before the scheduler deploys a service to the node. You can apply constraints to the service based upon node attributes and metadata or engine metadata. For more information on constraints, refer to the `docker service create` [CLI reference](/reference/cli/docker/service/create.md). #### Reserve memory or CPUs for a service To reserve a given amount of memory or number of CPUs for a service, use the `--reserve-memory` or `--reserve-cpu` flags. If no available nodes can satisfy the requirement (for instance, if you request 4 CPUs and no node in the swarm has 4 CPUs), the service remains in a pending state until an appropriate node is available to run its tasks. ##### Out Of Memory Exceptions (OOME) If your service attempts to use more memory than the swarm node has available, you may experience an Out Of Memory Exception (OOME) and a container, or the Docker daemon, might be killed by the kernel OOM killer. To prevent this from happening, ensure that your application runs on hosts with adequate memory and see [Understand the risks of running out of memory](/manuals/engine/containers/resource_constraints.md#understand-the-risks-of-running-out-of-memory). Swarm services allow you to use resource constraints, placement preferences, and labels to ensure that your service is deployed to the appropriate swarm nodes. #### Placement constraints Use placement constraints to control the nodes a service can be assigned to. In the following example, the service only runs on nodes with the [label](manage-nodes.md#add-or-remove-label-metadata) `region` set to `east`. If no appropriately-labelled nodes are available, tasks will wait in `Pending` until they become available. The `--constraint` flag uses an equality operator (`==` or `!=`). For replicated services, it is possible that all services run on the same node, or each node only runs one replica, or that some nodes don't run any replicas. For global services, the service runs on every node that meets the placement constraint and any [resource requirements](#reserve-memory-or-cpus-for-a-service). ```console $ docker service create \ --name my-nginx \ --replicas 5 \ --constraint node.labels.region==east \ nginx ``` You can also use the `constraint` service-level key in a `compose.yaml` file. If you specify multiple placement constraints, the service only deploys onto nodes where they are all met. The following example limits the service to run on all nodes where `region` is set to `east` and `type` is not set to `devel`: ```console $ docker service create \ --name my-nginx \ --mode global \ --constraint node.labels.region==east \ --constraint node.labels.type!=devel \ nginx ``` You can also use placement constraints in conjunction with placement preferences and CPU/memory constraints. Be careful not to use settings that are not possible to fulfill. For more information on constraints, refer to the `docker service create` [CLI reference](/reference/cli/docker/service/create.md). #### Placement preferences While [placement constraints](#placement-constraints) limit the nodes a service can run on, _placement preferences_ try to place tasks on appropriate nodes in an algorithmic way (currently, only spread evenly). For instance, if you assign each node a `rack` label, you can set a placement preference to spread the service evenly across nodes with the `rack` label, by value. This way, if you lose a rack, the service is still running on nodes on other racks.

This section covers resource reservation for Docker Swarm services using `--reserve-memory` and `--reserve-cpu` flags. It also discusses Out Of Memory Exceptions (OOME) and how to prevent them. The section then details placement constraints using the `--constraint` flag to control which nodes a service can be assigned to based on node labels, along with examples. Finally, it explains placement preferences, which attempt to distribute tasks evenly across nodes based on labels.