Registering the Service Broker
Smoke Tests and Acceptance Tests
Deregistering the Service Broker
Additional Configuration Options
A BOSH release of a MySQL database-as-a-service for Cloud Foundry using MariaDB Galera Cluster and a v2 Service Broker.
| Component | Description |
|---|---|
| CF MySQL Broker | Advertises the MySQL service and plans. Creates and deletes MySQL databases and credentials (bindings) at the request of Cloud Foundry's Cloud Controller. |
| MySQL Server | MariaDB 10.0.17; database instances are hosted on the servers. |
| Proxy | Switchboard; proxies to MySQL, severing connections on MySQL node failure. |
Traffic to the MySQL cluster is routed through one or more proxy nodes. The current proxy implementation is Switchboard. This proxy acts as an intermediary between the client and the MySQL server, providing failover between MySQL nodes. The number of nodes is configured by the proxy job instance count in the deployment manifest.
NOTE: If the number of proxy nodes is set to zero, apps will be bound to the IP address of the first MySQL node in the cluster. If that IP address should change for any reason (e.g. loss of a VM) or a proxy was subsequently added, one would need to re-bind all apps to the IP address of the new node.
For more details see the proxy documentation.
A user-facing service dashboard is provided by the service broker that displays storage utilization information for each service instance.
The dashboard is accessible by users via Single Sign-On (SSO) once authenticated with Cloud Foundry.
The dashboard URL can be found by running cf service MY_SERVICE_INSTANCE.
Service authors interested in implementing a service dashboard accessible via SSO can follow documentation for Dashboard SSO.
-
SSO is initiated when a user navigates to the URL found in the
dashboard_urlfield. This value is returned to cloud controller by the broker in response to a provision request, and is exposed in the cloud controller API for the service instance. A users client must expose this field as a link, or it can be obtained via curl (cf curl /v2/service_instances/:guid) and copied into a browser. -
SSO requires the following OAuth client to be configured in cf-release. This client is responsible for creating the OAuth client for the MySQL dashboard. Without this client configured in cf-release, the MySQL dashboard will not be accessible but the service will be otherwise functional. Registering the broker will display a warning to this effect.
properties: uaa: clients: cc-service-dashboards: secret: cc-broker-secret scope: cloud_controller.write,openid,cloud_controller.read,cloud_controller_service_permissions.read authorities: clients.read,clients.write,clients.admin authorized-grant-types: client_credentials -
SSO was implemented in v169 of cf-release; if you are on an older version of cf-release you'll encounter an error when you register the service broker. If upgrading cf-release is not an option, try removing the following lines from the cf-mysql-release manifest and redeploy.
dashboard_client: id: p-mysql secret: yoursecret
The following links show how this release implements Dashboard SSO integration.
- Update the broker catalog with the dashboard client properties
- Implement oauth workflow with the omniauth-uaa-oauth2 gem
- Use the cf-uaa-lib gem to get a valid access token and request permissions on the instance
- Before showing the user the dashboard, the broker checks to see if the user is logged-in and has permissions to view the usage details of the instance.
The dashboard URL defaults to using the https scheme. This means any requests using http will automatically be redirected to https instead.
To override this, you can change jobs.cf-mysql-broker_z1.ssl_enabled to false.
Keep in mind that changing the ssl_enabled setting for an existing broker will not update previously advertised dashboard URLs.
Visiting the old URL may fail if you are using the SSO integration,
because the OAuth2 client registered with UAA will expect users to both come from and return to a URI using the scheme
implied by the ssl_enabled setting.
Note:
If using https, the broker must be reached through an SSL termination proxy.
Connecting to the broker directly on https will result in a port 443: Connection refused error.
By default, the broker will not trust a self-signed SSL certificate when communicating with cf-release.
To trust self-signed SSL certificates, you can change jobs.cf-mysql-broker_z1.skip_ssl_validation to true.
Final releases are designed for public use, and are tagged with a version number of the form "v". The releases and corresponding release notes can be found on github.
See our contributing docs for instructions on how to make a pull request.
This BOSH release doubles as a $GOPATH. It will automatically be set up for
you if you have direnv installed.
# fetch release repo
mkdir -p ~/workspace
cd ~/workspace
git clone https://github.com/cloudfoundry/cf-mysql-release.git
cd cf-mysql-release/
# switch to develop branch (not master!)
git checkout develop
# automate $GOPATH and $PATH setup
direnv allow
# initialize and sync submodules
./update
If you do not wish to use direnv, you can simply source the .envrc file in the root
of the release repo. You may manually need to update your $GOPATH and $PATH variables
as you switch in and out of the directory.
For release notes and known issues, see the release wiki.
- A deployment of BOSH
- A deployment of Cloud Foundry, final release 193 or greater
- Instructions for installing BOSH and Cloud Foundry can be found at http://docs.cloudfoundry.org/.
- Upload Stemcell
- Upload Release
- Create Infrastructure
- Deployment Components
- Create Manifest and Deploy
After installation, the MySQL service will be visible in the Services Marketplace; using the CLI, run cf marketplace.
The latest final release expects the Ubuntu Trusty (14.04) go_agent stemcell version 2859 by default. Older stemcells are not recommended. Stemcells can be downloaded from http://bosh.io/stemcells; choose the appropriate stemcell for your infrastructure (vsphere esxi or aws hvm).
You can use a pre-built final release or build a dev release from any of the branches described in Getting the Code.
Final releases are stable releases created periodically for completed features. They also contain pre-compiled packages, which makes deployment much faster. To deploy the latest final release, simply check out the master branch. This will contain the latest final release and accompanying materials to generate a manifest. If you would like to deploy an earlier final release, use git checkout <tag> to obtain both the release and corresponding manifest generation materials. It's important that the manifest generation materials are consistent with the release.
If you'd like to deploy the latest code, build a release yourself from the develop branch.
Run the upload command, referencing the latest config file in the releases directory.
$ cd ~/workspace/cf-mysql-release
$ git checkout master
$ ./update
$ bosh upload release releases/cf-mysql-<N>.yml
If deploying an older final release than the latest, check out the tag for the desired version; this is necessary for generating a manifest that matches the code you're deploying.
$ cd ~/workspace/cf-mysql-release
$ git checkout v<N>
$ ./update
$ bosh upload release releases/cf-mysql-<N>.yml
- Checkout one of the branches described in Getting the Code. Build a BOSH development release.
$ cd ~/workspace/cf-mysql-release
$ git checkout release-candidate
$ ./update
$ bosh create release
When prompted to name the release, call it cf-mysql.
- Upload the release to your bosh environment:
$ bosh upload release
Note: No infrastructure changes are required to deploy to bosh-lite
Prior to deployment, the operator should define three subnets via their infrastructure provider. The MySQL release is designed to be deployed across three subnets to ensure availability in the event of a subnet failure. During installation, a fourth subnet is required for compilation vms. The sample_aws_stub.yml demonstrates how these subnets can be configured on AWS across multiple availability zones.
In order to route requests to both proxies, the operator should create a load balancer. Manifest changes required to configure a load balancer can be found in the proxy documentation. Once a load balancer is configured, the brokers will hand out the address of the load balancer rather than the IP of the first proxy. Currently, load balancing requests across both proxies can increase the possibility of deadlocks. See the routing documentation for more information. To avoid this problem, configure the load balancer to route requests to the second proxy only in the event of a failure.
There are three mysql jobs (mysql_z1, mysql_z2, mysql_z3) which should be deployed with one instance each. Each of these instances will reside in separate subnets as described in the previous section. The number of mysql nodes should always be odd, with a minimum count of three, to avoid split-brain. When the failed node comes back online, it will automatically rejoin the cluster and sync data from one of the healthy nodes. Note: Due to our bootstrapping procedure, if you are bringing up a cluster for the first time, there must be a database node in the first subnet.
There are two proxy jobs (proxy_z1, proxy_z2), which should be deployed with one instance each to different subnets. The second proxy is intended to be used in a failover capacity. In the event the first proxy fails, the second proxy will still be able to route requests to the mysql nodes.
There are also two broker jobs (cf-mysql-broker_z1, cf-mysql-broker_z2) which should be deployed with one instance each to different subnets. The brokers each register a route with the router, which load balances requests across the brokers.
-
Generate the manifest using a bosh-lite specific script and a stub provided for you,
bosh-lite/cf-mysql-stub-spiff.yml.$ ./bosh-lite/make_manifestThe resulting file,
bosh-lite/manifests/cf-mysql-manifest.ymlis your deployment manifest. To modify the deployment configuration, you can edit the stub and regenerate the manifest or edit the manifest directly. -
The
make_manifestscript will set the deployment tobosh-lite/manifests/cf-mysql-manifest.ymlfor you, so to deploy you only need to run:
$ bosh deploy
-
Create a stub file called
cf-mysql-vsphere-stub.ymlby copying and modifying the sample_vsphere_stub.yml intemplates/sample_stubs. Thesample_plans_stub.ymlcan also be copied if values need changing. -
Generate the manifest:
$ ./generate_deployment_manifest \
vsphere \
plans_stub.yml \
cf-mysql-vsphere-stub.yml > cf-mysql-vsphere.yml
The resulting file, cf-mysql-vsphere.yml is your deployment manifest. To modify the deployment configuration, you can edit the stub and regenerate the manifest or edit the manifest directly.
- To deploy:
$ bosh deployment cf-mysql-vsphere.yml && bosh deploy
-
Create a stub file called
cf-mysql-aws-stub.ymlby copying and modifying the sample_aws_stub.yml intemplates/sample_stubs. Thesample_plans_stub.ymlcan also be copied if values need changing. -
Generate the manifest:
$ ./generate_deployment_manifest \
aws \
plans_stub.yml \
cf-mysql-aws-stub.yml > cf-mysql-aws.yml
The resulting file, cf-mysql-aws.yml is your deployment manifest. To modify the deployment configuration, you can edit the stub and regenerate the manifest or edit the manifest directly.
- To deploy:
$ bosh deployment cf-mysql-aws.yml && bosh deploy
Manifest properties are described in the spec file for each job; see jobs.
You can find your director_uuid by running bosh status.
The MariaDB cluster nodes are configured by default with 100GB of persistent disk. This can be configured in your stub or manifest using disk_pools.mysql-persistent-disk.disk_size, however your deployment will fail if this is less than 3GB; we recommend allocating 10GB at a minimum.
BOSH errands were introduced in version 2366 of the BOSH CLI, BOSH Director, and stemcells.
$ bosh run errand broker-registrar
Note: the broker-registrar errand will fail if the broker has already been registered, and the broker name does not match the manifest property jobs.broker-registrar.properties.broker.name. Use the cf rename-service-broker CLI command to change the broker name to match the manifest property then this errand will succeed.
-
First register the broker using the
cfCLI. You must be logged in as an admin.$ cf create-service-broker p-mysql BROKER_USERNAME BROKER_PASSWORD URLBROKER_USERNAMEandBROKER_PASSWORDare the credentials Cloud Foundry will use to authenticate when making API calls to the service broker. Use the values for manifest propertiesjobs.cf-mysql-broker_z1.properties.auth_usernameandjobs.cf-mysql-broker_z1.properties.auth_password.URLspecifies where the Cloud Controller will access the MySQL broker. Use the value of the manifest propertyjobs.cf-mysql-broker_z1.properties.external_host. By default, this value is set top-mysql.<properties.domain>(in spiff:"p-mysql." .properties.domain).For more information, see Managing Service Brokers.
Note: adding additional security groups for cf-mysql is not required on bosh-lites running cf-release v212 or later.
Since cf-release v175, applications by default cannot to connect to IP addresses on the private network. This prevents applications from connecting to the MySQL service. To enable access to the service, create a new security group for the IP configured in your manifest for the property jobs.cf-mysql-broker_z1.mysql_node.host.
- Add the rule to a file in the following json format; multiple rules are supported.
[
{
"destination": "10.10.163.1-10.10.163.255",
"protocol": "all"
},
{
"destination": "10.10.164.1-10.10.164.255",
"protocol": "all"
},
{
"destination": "10.10.165.1-10.10.165.255",
"protocol": "all"
}
]
-
Create a security group from the rule file.
$ cf create-security-group p-mysql rule.json
-
Enable the rule for all apps
$ cf bind-running-security-group p-mysql
Security group changes are only applied to new application containers; existing apps must be restarted.
The smoke tests are a subset of the acceptance tests, useful for verifying a deployment. The acceptance tests are for developers to validate changes to the MySQL Release. These tests can be run manually or from a BOSH errand. For details on running these tests manually, see Acceptance Tests.
The MySQL Release contains an "acceptance-tests" job which is deployed as a BOSH errand. The errand can then be run to verify the deployment. A deployment manifest generated with the provided spiff templates will include this job. The errand can be configured to run either the smoke tests (default) or the acceptance tests.
To run the MySQL Release Smoke tests you will need:
- a running CF instance
- credentials for a CF Admin user
- a deployed MySQL Release with the broker registered and the plan made public
Run the smoke tests via bosh errand as follows:
$ bosh run errand acceptance-tests
Modifying values under jobs.acceptance-tests.properties may be required. Configuration options can be found in the job spec.
The following commands are destructive and are intended to be run in conjuction with deleting your BOSH deployment.
BOSH errands were introduced in version 2366 of the BOSH CLI, BOSH Director, and stemcells.
This errand runs the two commands listed in the manual section below from a BOSH-deployed VM.
$ bosh run errand broker-deregistrar
Run the following:
$ cf purge-service-offering p-mysql
$ cf delete-service-broker p-mysql
The service is configured to have a small footprint out of the box. These resources are sufficient for development, but may be insufficient for production workloads. If the service appears to be performing poorly, redeploying with increased resources may improve performance. See deployment resources for further details.
Updating the service instances is supported; see Service plans and instances for details.
Normally databases are created via the cf create-service command, and
a MySQL user is created and given access to that database when an app is bound to that service instance.
However, it is sometimes useful to have databases and users already available when the service is deployed,
without having to run cf create-service or bind an app.
To specify any preseeded databases, add the following to the deployment manifest:
jobs:
- name: mysql_z1
properties:
seeded_databases:
- name: db1
username: user1
password: pw1
- name: db2
username: user2
password: pw2
Note: If all you need is a database deployment, it is possible to deploy this release with zero broker instances and completely remove any dependencies on Cloud Foundry. See the proxy and acceptance-tests spec files for standalone configuration options.
On larger databases, the default database startup timeout may be too low. This would result in the job reporting as failing, while MySQL continues to bootstrap in the background (see Known Issues > Long SST Transfers). To increase the duration that the startup script waits for MySQL to start, add the following to your deployment stub:
jobs:
- name: mysql_z1
properties:
database_startup_timeout: 360Note: This is independent of the overall BOSH timeout which is also configurable in the manifest. The BOSH timeout should always be higher than the database startup timeout:
update:
canary_watch_time: 30000-600000
update_watch_time: 30000-600000