Running Trove in production

This document is not a definitive guide for deploying Trove in every production environment. There are many ways to deploy Trove depending on the specifics and limitations of your situation. We hope this document provides the cloud operator or distribution creator with a basic understanding of how the Trove components fit together practically. Through this, it should become more obvious how components of Trove can be divided or duplicated across physical hardware in a production cloud environment to aid in achieving scalability and resiliency for the database as a service software.

In the interest of keeping this guide somewhat high-level and avoiding obsolescence or operator/distribution-specific environment assumptions by specifying exact commands that should be run to accomplish the tasks below, we will instead just describe what needs to be done and leave it to the cloud operator or distribution creator to “do the right thing” to accomplish the task for their environment. If you need guidance on specific commands to run to accomplish the tasks described below, we recommend reading through the script in devstack subdirectory of this project. The devstack plugin exercises all the essential components of Trove in the right order, and this guide will mostly be an elaboration of this process.

Environment Assumptions

The scope of this guide is to provide a basic overview of setting up all the components of Trove in a production environment, assuming that the default in-tree drivers and components are going to be used.

For the purposes of this guide, we will therefore assume the following core components have already been set up for your production OpenStack environment:

  • RabbitMQ

  • MySQL

  • Keystone

  • Nova

  • Cinder

  • Neutron

  • Glance

  • Swift

Production Deployment Walkthrough

Create Trove Service User

By default Trove will use the ‘trove’ user with ‘admin’ role in ‘service’ tenant for both keystone authentication and interactions with all other services.

Service Tenant Deployment

In production, almost all the cloud resources(except the Swift objects for backup data and floating IP addresses for public instances) created for a Trove instance should be only visible to the Trove service user. As DBaaS users, they should only see a Trove instance after creating, and know nothing about the Nova VM, Cinder volume, Neutron management network and security groups under the hood. The only way to operate Trove instance is to interact with Trove API.

Service tenant deployment is the default configuration in Trove since Ussuri release.

Install Trove Controller Software

Trove controller services should be put somewhere that has access to the database, the oslo messaging system, and other OpenStack services. Trove uses the standard python setuptools, so installation of the software itself should be straightforward.

Running multiple instances of the individual Trove controller components on separate physical hosts is recommended in order to provide scalability and availability of the controller software.

Management Network

Trove makes use of a “Management Network” exclusively that the controller uses to talk to guest agent running inside Trove instance and vice versa. All the instances that Trove deploys will have interfaces on this network. Therefore, it’s important that the subnet deployed on this network be sufficiently large to allow for the maximum number of instances and controllers likely to be deployed throughout the lifespan of the cloud installation.

Usually, after a Trove instance is created, there are 2 nics attached to the instance VM, one for the database traffic on user-defined network, one for management purpose. Trove will check if the user’s subnet conflicts with the management network.

You can also create a management Neutron security group that will be applied to the management port. Basically, nothing needs to be allowed to access the management port, most of the network communication within the Trove instance is egress traffic(e.g. the guest agent initiates connection with RabbitMQ). However, It can be helpful to allow SSH access to the Trove instance from the controller for troubleshooting purposes (ie. TCP port 22), though this is not strictly necessary in production environments.

In order to SSH into the Trove instance(as mentioned above, it’s helpful but not necessary), the cloud administrators need to create and config a Nova keypair.

Finally, you need to add routing or interfaces to this network so that the Trove guest agent running inside the instance is able to connect with RabbitMQ.

RabbitMQ Considerations

Both trove-taskmanager and trove-conductor talk to guest agent inside Trove instance via the messaging system, ie. RabbitMQ. Once the guest agent is up and running, it’s listening on a message queue named guestagent.<guest ID> specifically set up for that particular instance, receiving requests from trove-taskmanager for operations like set up the database software, create databases and users, restart database service etc. At the mean while, trove-guestagent periodically sends status update information to trove-conductor through the messaging system.

With all that said, a proper RabbitMQ user name and password need to be configured in the trove-guestagent config file, which may bring security concern for the cloud deployers. If the guest instance is compromised, then guest credentials are compromised, which means the messaging system is compromised.

As part of the solution, Trove introduced a security enhancement in Ocata release, using encryption keys to protect the messages between the control plane and the guest instances, which guarantees that one compromised guest instance doesn’t affect other instances nor other cloud users.

Configuring Trove

The default Trove configuration file location is /etc/trove/trove.conf. You can generate a sample config file by running:

cd <trove dir>
pip install -e .
oslo-config-generator --namespace trove.config --namespace oslo.messaging --namespace oslo.log --namespace oslo.policy --output-file /etc/trove/trove.conf.sample

The typical config options (not a full list) are:


Should RPC messaging traffic be secured by encryption.


The key (OpenSSL aes_cbc) used to encrypt RPC messages sent to trove-taskmanager, used by trove-api.


The key (OpenSSL aes_cbc) used to encrypt RPC messages sent to guest instance from trove-taskmanager and the messages sent from guest instance to trove-conductor. This key is generated by trove-taskmanager automatically and is injected into the guest instance when creating.


The database encryption key to encrypt per-instance PRC encryption key before storing to Trove database.


The management network, currently only one management network is allowed.


List of the management security groups that are applied to the management port of the database instance.


Cinder volume type used to create volume that is attached to Trove instance.


Name of a Nova keypair to inject into a database instance to enable SSH access.


The default datastore id or name to use if one is not provided by the user. If the default value is None, the field becomes required in the instance create request.


The default maximum volume size (in GB) for an instance.


Default maximum number of instances per tenant.


Default maximum number of backups per tenant.


The messaging server connection URL, e.g. rabbit://stackrabbit:password@


The Trove exchange name for the messaging service, could be overridden by an exchange name specified in the transport_url option.


Maximum time (in seconds) to wait for a server reboot.


Maximum time (in seconds) to wait for Trove instance to become ACTIVE for creation.


Maximum time (in seconds) to wait for Trove instance to become ACTIVE for restore.


Maximum time (in seconds) to wait for Guest Agent ‘slow’ requests (such as restarting the instance server) to complete.


The UID(GID) of database service user.

keystone_authtoken group

Like most of other OpenStack services, Trove uses Keystone Authentication Middleware for authentication and authorization.

service_credentials group

Options in this section are pretty much like the options in keystone_authtoken, but you can config another service user for Trove to communicate with other OpenStack services like Nova, Neutron, Cinder, etc.

  • auth_url

  • region_name

  • project_name

  • username

  • password

  • project_domain_name

  • user_domain_name

database group

The SQLAlchemy connection string to use to connect to the database, e.g. mysql+pymysql://root:password@

The cloud administrator also needs to provide a policy file /etc/trove/policy.yaml if the default API access policies don’t satisfy the requirement. To generate a sample policy file with all the default policies, run tox -egenpolicy in the repo folder and the new file will be located in etc/trove/policy.yaml.sample.


JSON formatted policy file is deprecated since Trove 15.0.0 (Wallaby). This oslopolicy-convert-json-to-yaml tool will migrate your existing JSON-formatted policy file to YAML in a backward-compatible way.

Configure Trove Guest Agent

The config file of trove guest agent is copied from trove controller node (default file path /etc/trove/trove-guestagent.conf) when creating instance.

Some config options specifically for trove guest agent:

  • Custom container image registry.

    Trove guest agent pulls container images from docker hub by default, this can be changed by setting:

    container_registry =
    container_registry_username =
    container_registry_password =

    Then in the specific database config section, the customized container registry can be used, e.g.

    docker_image = your-registry/your-repo/mysql
    backup_docker_image = your-registry/your-repo/db-backup-mysql:1.1.0

Initialize Trove Database

This is controlled through sqlalchemy-migrate scripts under the trove/db/sqlalchemy/migrate_repo/versions directory in this repository. The script trove-manage (which should be installed together with Trove controller software) could be used to aid in the initialization of the Trove database. Note that this tool looks at the /etc/trove/trove.conf file for its database credentials, so initializing the database must happen after Trove is configured.

Launching the Trove Controller

We recommend using upstart / systemd scripts to ensure the components of the Trove controller are all started and kept running.

Preparing the Guest Images

Currently supported databases are: MySQL 5.7.X, MariaDB 10.4.X. PostgreSQL 12.4 is partially supported.

Now that the Trove system is installed, the next step is to build the images that we will use for the DBaaS to function properly. This is possibly the most important step as this will be the gold standard that Trove will use for a particular data store.


For the sake of simplicity and especially for testing, we can use the prebuilt images that are available from OpenStack itself. These images should strictly be used for testing and development use and should not be used in a production environment. The images are available for download and are located at

From Victoria release, Trove uses a single guest image for all the supported datastores. Database service is running as docker container inside the trove instance which simplifies the datastore management and maintenance.

For use with production systems, it is recommended to create and maintain your own images in order to conform to standards set by the company’s security team. In Trove community, we use Disk Image Builder(DIB) to create Trove images, all the elements are located in integration/scripts/files/elements folder in the repo.

Trove provides a script named trovestack to help build the image, refer to Build images using trovestack for more information. Make sure to use dev_mode=false for production environment.

After image is created successfully, the cloud administrator needs to upload the image to Glance and make it only accessible to service users. It’s recommended to use tags when creating Glance image.

Preparing the Datastore

After image is uploaded, the cloud administrator should create datastores, datastore versions and the configuration parameters for the particular version.

It’s recommended to config a default version for each datastore.

trove-manage can be only used on trove controller node.

Command examples:

$ # Creating datastore 'mysql' and datastore version 5.7.29.
$ openstack datastore version create 5.7.29 mysql mysql "" \
  --image-tags trove,mysql \
  --active --default \
  --version-number 5.7.29
$ # Register configuration parameters for the datastore version
$ trove-manage db_load_datastore_config_parameters mysql 5.7.29 ${trove_repo_dir}}/trove/templates/mysql/validation-rules.json

Quota Management

The amount of resources that could be created by each OpenStack project is controlled by quota. The default trove resource quota for each project is set in Trove config file as follows unless changed by the cloud administrator via Quota API.

max_instances_per_tenant = 10
max_backups_per_tenant = 50

In the meantime, trove service project itself also needs quota to create cloud resources corresponding to the trove instances, e.g.

openstack quota set \
  --instances 200 \
  --server-groups 200 \
  --volumes 200 \
  --secgroups 200 \
  --ports 400 \

Trove Deployment Verfication

If all of the above instructions have been followed, it should now be possible to deploy Trove instances using the OpenStack CLI, communicating with the Trove V1 API.

Refer to Create and access a database for detailed steps.