======================== Key Compute API Concepts ======================== The OpenStack Compute API is defined as a RESTful HTTP service. The API takes advantage of all aspects of the HTTP protocol (methods, URIs, media types, response codes, etc.) and providers are free to use existing features of the protocol such as caching, persistent connections, and content compression among others. Providers can return information identifying requests in HTTP response headers, for example, to facilitate communication between the provider and client applications. OpenStack Compute is a compute service that provides server capacity in the cloud. Compute Servers come in different flavors of memory, cores, disk space, and CPU, and can be provisioned in minutes. Interactions with Compute Servers can happen programmatically with the OpenStack Compute API. User Concepts ============= To use the OpenStack Compute API effectively, you should understand several key concepts: - **Server** A virtual machine (VM) instance, physical machine or a container in the compute system. Flavor and image are requisite elements when creating a server. A name for the server is also required. For more details, such as server actions and server metadata, please see: :doc:`server_concepts` - **Flavor** Virtual hardware configuration for the requested server. Each flavor has a unique combination of disk space, memory capacity and priority for CPU time. - **Flavor Extra Specs** TODO: Short description at here. The detail reference to :doc:`extra_specs_and_properties` - **Image** A collection of files used to create or rebuild a server. Operators provide a number of pre-built OS images by default. You may also create custom images from cloud servers you have launched. These custom images are useful for backup purposes or for producing “gold” server images if you plan to deploy a particular server configuration frequently. - **Image Properties** TODO: Short description at here. The detail reference to :doc:`extra_specs_and_properties` - **Key Pair** An ssh or x509 keypair that can be injected into a server at it's boot time. This allows you to connect to your server once it has been created without having to use a password. If you don't specify a key pair, Nova will create a root password for you, and return it in plain text in the server create response. - **Volume** A block storage device that Nova can use as permanent storage. When a server is created it has some disk storage available, but that is considered ephemeral, as it is destroyed when the server is destroyed. A volume can be attached to a server, then later detached and used by another server. Volumes are created and managed by the Cinder service, though the Nova API can proxy some of these calls. - **Quotas** An upper limit on the amount of resources any individual tenant may consume. Quotas can be used to limit the number of servers a tenant creates, or the amount of disk space consumed, so that no one tenant can overwhelm the system and prevent normal operation for others. Changing quotas is an administrator-level action. - **Rate Limiting** Please see :doc:`limits` - **Availability zone** A grouping of host machines that can be used to control where a new server is created. There is some confusion about this, as the name "availability zone" is used in other clouds, such as Amazon Web Services, to denote a physical separation of server locations that can be used to distribute cloud resources for fault tolerance in case one zone is unavailable for any reason. Such a separation is possible in Nova if an administrator carefully sets up availability zones for that, but it is not the default. Networking Concepts ------------------- In this section we focus on this related to networking. - **Port** TODO - **Floating IPs, Pools and DNS** TODO - **Security Groups** TODO - **Cloudpipe** TODO - **Extended Networks** TODO Administrator Concepts ====================== Some APIs are largely focused on administration of Nova, and generally focus on compute hosts rather than servers. - **Services** Services are provided by Nova components. Normally, the Nova component runs as a process on the controller/compute node to provide the service. These services may be end-user facing, such as the OpenStack Compute REST API service, but most just work with other Nova services. The status of each service is monitored by Nova, and if it is not responding normally, Nova will update its status so that requests are not sent to that service anymore. The service can also be controlled by an Administrator in order to run maintenance or upgrades, or in response to changing workloads. - **nova-osapi_compute** This service provides the OpenStack Compute REST API to end users and application clients. - **nova-metadata** This service provides the OpenStack Metadata API to servers. The metadata is used to configure the running servers. - **nova-scheduler** This service provides compute request scheduling by tracking available resources, and finding the host that can best fulfill the request. - **nova-conductor** This service provides database access for Nova and the other OpenStack services, and handles internal version compatibility when different services are running different versions of code. The conductor service also handles long-running requests. - **nova-compute** This service runs on every compute node, and communicates with a hypervisor for managing compute resources on that node. - **nova-network** This service handles networking of virtual servers. It is no longer under active development, and is being replaced by Neutron. - **nova-ec2(deprecated)** This service provides AWS EC2 API compatibility. - **nova-consoleauth** This service provides authorization for compute instances consoles. - **nova-cert** This service handles the management of X509 certificates. - **Services Actions** - **enable, disable, disable-log-reason** The service can be disabled to indicate the service is not available anymore. This is used by administrator to stop service for maintenance. For example, when Administrator wants to maintain a specific compute node, Administrator can disable nova-compute service on that compute node. Then nova won't dispatch any new compute request to that compute node anymore. Administrator also can add note for disable reason. - **forced-down** This action allows you set the state of service down immediately. Actually Nova only provides the health monitor of service status, there isn't any guarantee about health status of other parts of infrastructure, like the health status of data network, storage network and other components. The more complete health monitor of infrastructure is provided by external system normally. An external health monitor system can mark the service down for notifying the fault. `(This action is enabled in Microversion 2.11)` - **Hosts** Hosts are the *physical machines* that provide the resources for the virtual servers created in Nova. They run a ``hypervisor`` (see definition below) that handles the actual creation and management of the virtual servers. Hosts also run the ``Nova compute service``, which receives requests from Nova to interact with the virtual servers on that machine. When compute service receives a request, it calls the appropriate methods of the driver for that hypervisor in order to carry out the request. The driver acts as the translator from generic Nova requests to hypervisor-specific calls. Hosts report their current state back to Nova, where it is tracked by the scheduler service, so that the scheduler can place requests for new virtual servers on the hosts that can best fit them. - **Host Actions** A *host action* is one that affects the physical host machine, as opposed to actions that only affect the virtual servers running on that machine. There are three 'power' actions that are supported: *startup*, *shutdown*, and *reboot*. There are also two 'state' actions: enabling/disabling the host, and setting the host into or out of maintenance mode. Of course, carrying out these actions can affect running virtual servers on that host, so their state will need to be considered before carrying out the host action. For example, if you want to call the 'shutdown' action to turn off a host machine, you might want to migrate any virtual servers on that host before shutting down the host machine so that the virtual servers continue to be available without interruption. - **Hypervisors** A hypervisor, or virtual machine monitor (VMM), is a piece of computer software, firmware or hardware that creates and runs virtual machines. In nova, each Host (see `Hosts`) runs a hypervisor. Administrators are able to query the hypervisor for information, such as all the virtual servers currently running, as well as detailed info about the hypervisor, such as CPU, memory, or disk related configuration. Currently nova-compute also supports Ironic and LXC, but they don't have a hypervisor running. - **Aggregates** See `Aggregates developer information `_. - **Migrations** Migrations are the process where a virtual server is moved from one host to another. Please see :doc:`server_concepts` for details about moving servers. Administrators are able to query the records in database for information about migrations. For example, they can determine the source and destination hosts, type of migration, or changes in the server's flavor. - **Certificates** Nova service "nova-cert" handles the management of X509 certificates which are used to generate certificates for euca-bundle-image. Relationship with Volume API ============================ Here we discuss about Cinder's API and how Nova users volume UUIDs. TODO - add more details. Relationship with Image API =========================== Here we discuss about Glance's API and how Nova uses image UUIDs. We also discuss how Nova proxies setting image metadata. TODO - add more details. Interactions with Neutron and Nova-Network ========================================== We talk about how networking can be provided be either Nova or Neutron. Here we discuss about Neutron's API and how Nova users port UUIDs. We also discuss Nova automatically creating ports, proxying security groups, and proxying floating IPs. Also talk about the APIs we do not proxy. TODO - add more details.