Bare Metal Service Upgrade Guide

This document outlines various steps and notes for operators to consider when upgrading their ironic-driven clouds from previous versions of OpenStack.

The Bare Metal (ironic) service is tightly coupled with the ironic driver that is shipped with the Compute (nova) service. Some special considerations must be taken into account when upgrading your cloud.

Both offline and rolling upgrades are supported.

Plan your upgrade

  • Rolling upgrades are available starting with the Pike release; that is, when upgrading from Ocata. This means that it is possible to do an upgrade with minimal to no downtime of the Bare Metal API.

  • Upgrades are only supported between two consecutive named releases. This means that you cannot upgrade Ocata directly into Queens; you need to upgrade into Pike first.

  • The release notes should always be read carefully when upgrading the Bare Metal service. Specific upgrade steps and considerations are documented there.

  • The Bare Metal service should always be upgraded before the Compute service.

    Note

    The ironic virt driver in nova always uses a specific version of the ironic REST API. This API version may be one that was introduced in the same development cycle, so upgrading nova first may result in nova being unable to use the Bare Metal API.

  • Make a backup of your database. Ironic does not support downgrading of the database. Hence, in case of upgrade failure, restoring the database from a backup is the only choice.

  • Before starting your upgrade, it is best to ensure that all nodes have reached, or are in, a stable provision_state. Nodes in states with long running processes such as deploying or cleaning, may fail, and may require manual intervention to return them to the available hardware pool. This is most likely in cases where a timeout has occurred or a service was terminated abruptly. For a visual diagram detailing states and possible state transitions, please see Bare Metal State Machine.

Offline upgrades

In an offline (or cold) upgrade, the Bare Metal service is not available during the upgrade, because all the services have to be taken down.

When upgrading the Bare Metal service, the following steps should always be taken in this order:

  1. upgrade the ironic-python-agent image

  2. update ironic code, without restarting services

  3. run database schema migrations via ironic-dbsync upgrade

  4. restart ironic-conductor and ironic-api services

Once the above is done, do the following:

  • update any applicable configuration options to stop using any deprecated features or options, and perform any required work to transition to alternatives. All the deprecated features and options will be supported for one release cycle, so should be removed before your next upgrade is performed.

  • upgrade python-ironicclient along with any other services connecting to the Bare Metal service as a client, such as nova-compute

  • run the ironic-dbsync online_data_migrations command to make sure that data migrations are applied. The command lets you limit the impact of the data migrations with the --max-count option, which limits the number of migrations executed in one run. You should complete all of the migrations as soon as possible after the upgrade.

    Warning

    You will not be able to start an upgrade to the release after this one, until this has been completed for the current release. For example, as part of upgrading from Ocata to Pike, you need to complete Pike’s data migrations. If this not done, you will not be able to upgrade to Queens – it will not be possible to execute Queens’ database schema updates.

Rolling upgrades

To Reduce downtime, the services can be upgraded in a rolling fashion, meaning to upgrade one or a few services at a time to minimize impact.

Rolling upgrades are available starting with the Pike release. This feature makes it possible to upgrade between releases, such as Ocata to Pike, with minimal to no downtime of the Bare Metal API.

Requirements

To facilitate an upgrade in a rolling fashion, you need to have a highly-available deployment consisting of at least two ironic-api and two ironic-conductor services. Use of a load balancer to balance requests across the ironic-api services is recommended, as it allows for a minimal impact to end users.

Concepts

There are four aspects of the rolling upgrade process to keep in mind:

  • API and RPC version pinning, and versioned object backports

  • online data migrations

  • graceful service shutdown

  • API load balancer draining

API & RPC version pinning and versioned object backports

Through careful RPC versioning, newer services are able to talk to older services (and vice-versa). The [DEFAULT]/pin_release_version configuration option is used for this. It should be set (pinned) to the release version that the older services are using. The newer services will backport RPC calls and objects to their appropriate versions from the pinned release. If the IncompatibleObjectVersion exception occurs, it is most likely due to an incorrect or unspecified [DEFAULT]/pin_release_version configuration value. For example, when [DEFAULT]/pin_release_version is not set to the older release version, no conversion will happen during the upgrade.

For the ironic-api service, the API version is pinned via the same [DEFAULT]/pin_release_version configuration option as above. When pinned, the new ironic-api services will not service any API requests with Bare Metal API versions that are higher than what the old ironic-api services support. HTTP status code 406 is returned for such requests. This prevents new features (available in new API versions) from being used until after the upgrade has been completed.

Online data migrations

To make database schema migrations less painful to execute, we have implemented process changes to facilitate upgrades.

  • All data migrations are banned from schema migration scripts.

  • Schema migration scripts only update the database schema.

  • Data migrations must be done at the end of the rolling upgrade process, after the schema migration and after the services have been upgraded to the latest release.

All data migrations are performed using the ironic-dbsync online_data_migrations command. It can be run as a background process so that it does not interrupt running services; however it must be run to completion for a cold upgrade if the intent is to make use of new features immediately.

(You would also execute the same command with services turned off if you are doing a cold upgrade).

This data migration must be completed. If not, you will not be able to upgrade to future releases. For example, if you had upgraded from Ocata to Pike but did not do the data migrations, you will not be able to upgrade from Pike to Queens. (More precisely, you will not be able to apply Queens’ schema migrations.)

Graceful conductor service shutdown

The ironic-conductor service is a Python process listening for messages on a message queue. When the operator sends the SIGTERM signal to the process, the service stops consuming messages from the queue, so that no additional work is picked up. It completes any outstanding work and then terminates. During this process, messages can be left on the queue and will be processed after the Python process starts back up. This gives us a way to shutdown a service using older code, and start up a service using newer code with minimal impact.

Note

This was tested with RabbitMQ messaging backend and may vary with other backends.

Nodes that are being acted upon by an ironic-conductor process, which are not in a stable state, will be put into a failed state when [DEFAULT]graceful_shutdown_timeout is reached. Node failures that occur during an upgrade are likely due to timeouts, resulting from delays involving messages being processed and acted upon by a conductor during long running, multi-step processes such as deployment or cleaning.

Drain conductor service shutdown

A drain shutdown is similar to graceful shutdown, differing in the following ways:

  • Triggered by sending signal SIGUSR2 to the process instead of SIGTERM

  • The timeout for process termination is determined by [DEFAULT]drain_shutdown_timeout instead of [DEFAULT]graceful_shutdown_timeout

[DEFAULT]drain_shutdown_timeout is set long enough so that any node in a not stable state will have time to reach a stable state (complete or failed) before the ironic-conductor process terminates.

API load balancer draining

If you are using a load balancer for the ironic-api services, we recommend that you redirect requests to the new API services and drain off of the ironic-api services that have not yet been upgraded.

Rolling upgrade process

Before maintenance window

  • Upgrade the ironic-python-agent image

  • Using the new release (ironic code), execute the required database schema updates by running the database upgrade command: ironic-dbsync upgrade. These schema change operations should have minimal or no effect on performance, and should not cause any operations to fail (but please check the release notes). You can:

    • install the new release on an existing system

    • install the new release in a new virtualenv or a container

    At this point, new columns and tables may exist in the database. These database schema changes are done in a way that both the old and new (N and N+1) releases can perform operations against the same schema.

Note

Ironic bases its API, RPC and object storage format versions on the [DEFAULT]/pin_release_version configuration option. It is advisable to automate the deployment of changes in configuration files to make the process less error prone and repeatable.

During maintenance window

  1. All ironic-conductor services should be upgraded first. Ensure that at least one ironic-conductor service is running at all times. For every ironic-conductor, either one by one or a few at a time:

    • shut down the service. Messages from the ironic-api services to the conductors are load-balanced by the message queue and a hash-ring, so the only thing you need to worry about is to shut the service down gracefully (using SIGTERM signal) to make sure it will finish all the requests being processed before shutting down.

    • upgrade the installed version of ironic and dependencies

    • set the [DEFAULT]/pin_release_version configuration option value to the version you are upgrading from (that is, the old version). Based on this setting, the new ironic-conductor services will downgrade any RPC communication and data objects to conform to the old service. For example, if you are upgrading from Ocata to Pike, set this value to ocata.

    • start the service

  2. The next service to upgrade is ironic-api. Ensure that at least one ironic-api service is running at all times. You may want to start another temporary instance of the older ironic-api to handle the load while you are upgrading the original ironic-api services. For every ironic-api service, either one by one or a few at a time:

    • in HA deployment you are typically running them behind a load balancer (for example HAProxy), so you need to take the service instance out of the balancer

    • shut it down

    • upgrade the installed version of ironic and dependencies

    • set the [DEFAULT]/pin_release_version configuration option value to the version you are upgrading from (that is, the old version). Based on this setting, the new ironic-api services will downgrade any RPC communication and data objects to conform to the old service. In addition, the new services will return HTTP status code 406 for any requests with newer API versions that the old services did not support. This prevents new features (available in new API versions) from being used until after the upgrade has been completed. For example, if you are upgrading from Ocata to Pike, set this value to ocata.

    • restart the service

    • add it back into the load balancer

    After upgrading all the ironic-api services, the Bare Metal service is running in the new version but with downgraded RPC communication and database object storage formats. New features (in new API versions) are not supported, because they could fail when objects are in the downgraded object formats and some internal RPC API functions may still not be available.

  3. For all the ironic-conductor services, one at a time:

    • remove the [DEFAULT]/pin_release_version configuration option setting

    • restart the ironic-conductor service

  4. For all the ironic-api services, one at a time:

    • remove the [DEFAULT]/pin_release_version configuration option setting

    • restart the ironic-api service

After maintenance window

Now that all the services are upgraded, the system is able to use the latest version of the RPC protocol and able to access all the features of the new release.

  • Update any applicable configuration options to stop using any deprecated features or options, and perform any required work to transition to alternatives. All the deprecated features and options will be supported for one release cycle, so should be removed before your next upgrade is performed.

  • Upgrade python-ironicclient along with other services connecting to the Bare Metal service as a client, such as nova-compute.

    Warning

    A nova-compute instance tries to attach VIFs to all active instances on start up. Make sure that for all active nodes there is at least one running ironic-conductor process to manage them. Otherwise the instances will be moved to the ERROR state on the nova-compute start up.

  • Run the ironic-dbsync online_data_migrations command to make sure that data migrations are applied. The command lets you limit the impact of the data migrations with the --max-count option, which limits the number of migrations executed in one run. You should complete all of the migrations as soon as possible after the upgrade.

    Warning

    Note that you will not be able to start an upgrade to the next release after this one, until this has been completed for the current release. For example, as part of upgrading from Ocata to Pike, you need to complete Pike’s data migrations. If this not done, you will not be able to upgrade to Queens – it will not be possible to execute Queens’ database schema updates.