Merge and delete
In the previous tutorial we “over-engineered” our service based solution a bit (on purpose of course). We don’t really need two backends services, so in this tutorial we will introduce our newly engineered solution and transition to it using Vamp blueprints and canary releasing methods. In this tutorial we will:
- Get some background and theory on merging services
- Prepare our blueprint
- Merge the new topology with the running deployment
- Decommission parts of the deployment
- Answer the all important question when would I use this?
Requirements: Docker machine should have access to at least 3GB memory
Some background and theory
What we are going to do is create a new blueprint that is completely valid by itself and merge it with our already running deployment. This might sound strange at first, but it makes sense. Why? Merging will enable us to slowly move from the previous solution to the next solution. Once moved over, we can remove any parts we no longer need, i.e. the former “over-engineered” topology.
In the diagram above, this is visualized as follows:
- We initiate a running deployment by deploying Blueprint A
- We introduce a new service to the deployment by merging Blueprint B with the running deployment.
- At this point, both blueprints are active allowing for a smooth transition, for example, by canary release.
- Once we are fully on running on Blueprint B, we can remove/decommission Blueprint A
Prepare our blueprint
The below blueprint describes our more reasonable service topology. This blueprint is completely valid by itself - you could just deploy it somewhere separately and not merge it with the over-engineered sava:1.2 topology.
Notice the following:
- There is only one backend cluster, with one service
- No gateway is specified (we will use the existing gateway in the running deployment)
name: sava:1.3 clusters: sava: # cluster 1 services: breed: name: sava-frontend:1.3.0 deployable: magneticio/sava-frontend:1.3.0 ports: webport: 8080/http environment_variables: BACKEND: http://$backend.host:$backend.ports.webport/api/message dependencies: backend: sava-backend:1.3.0 scale: cpu: 0.2 memory: 64MB instances: 1 health_checks: initial_delay: 10s port: webport timeout: 5s interval: 10s failures: 10 backend: # cluster 2 services: breed: name: sava-backend:1.3.0 deployable: magneticio/sava-backend:1.3.0 ports: webport: 8080/http scale: cpu: 0.2 memory: 64MB instances: 1
Merge the topologies
We can now merge the blueprint describing our new topology with the running sava-new deployment. This is done in the same way we added a new service in the run a canary release tutorial. Merging the blueprint will deploy the new sava:1.3 services alongside sava:1.2, without routing any traffic to them. You can complete the merge using either the Vamp UI or API.
Merge using the Vamp UI
- Go to the Blueprints page and click Add (top right)
- Paste in the above blueprint and click Save. Vamp will store the blueprint and make it available for deployment
- Open the action menu on the sava:1.3 blueprint and select Merge to
- You’ll be prompted to select the deployment you wish to merge the blueprint with - select sava-new
- Click Merge to deploy the services sava:1.3 and sava-backend:1.3 to the running sava-new deployment.
Merge using the API
You can complete the same merge action with the Vamp API - remember to set the
Content-Type: application/x-yaml for your requests:
- To create the blueprint,
POSTthe above blueprint YAML to
- To merge the blueprint,
PUTthe below YAML to
So what happened here? Vamp worked out which parts of the blueprint were already in the deployment and which parts should be added. This is done based on naming. The sava cluster already existed, so Vamp simply added a service to it with 0% route weight. A cluster named “backend” didn’t exist, so Vamp created it.
Notice we now have:
- Three backend clusters: two old ones (backend1 and backend2) and one from the new merge.
- Two services in the sava cluster: the old sava-frontend:1.2 and the new sava-frontend:1.3. If you open the sava cluster webport you will see that the sava-frontend:1.2.0 route has a weight of 100% and the new new sava-frontend:1.3.0 route has a weight of 0%. Whenever Vamp merges a new service to an existing cluster, the default weight of 0% is applied. This means no traffic will be routed to sava:1.3 yet.
Now both blueprints are deployed, moving from the old to the new topology is just a question of “turning the weight dial”. You could do this in one go, or slowly adjust it.
Decommission parts of the deployment
Once we are fully running on sava:1.3, we can decommission the old, over-engineered sava:1.2. We do this by updating the deployment again, this time to remove the sava:1.2 blueprint, effectively deleting all deployed sava:1.2 services.
Note that we won’t be able to remove any service unless its weight is first set to 0%.
Delete using the Vamp UI
- Go to the Blueprints page
- Open the action menu on the sava:1.2 blueprint and select Remove from
- You’ll be prompted to select the deployment you wish to remove the blueprint from - select sava-new
- Click Remove to delete the services sava:1.2, sava-backend1:1.2 and sava-backend2:1.2 from the running sava-new deployment.
Delete using the API
You can complete the same delete action with the Vamp API - remember to set the
Content-Type: application/x-yaml for your requests:
- To delete the sava:1.2 blueprint from the deployment,
DELETEthe below YAML to
Vamp will update the running deployment, removing all elements described in the sava:1.2 blueprint. This means that our running deployment sava-new now contains no sava:1.2 services.
When would I use this?
Sounds cool, but when would I use this in practice? Well, basically anytime you release something new!
For example a bugfix release for a mobile API that “didn’t change anything significantly”? You could test this separately and describe it in its own blueprint. After testing, you would merge that exact same blueprint with your already running production version (the one without the bugfix) and slowly move over to new version.
New major release of your customer facing app? You probably also have some new dependencies that come with that release. You create some containers and write up a blueprint that describes this new situation, run it in acceptance and test and what have you. Later, you merge it into your production setup, effectively putting it next to it and then slowly move from the old situation to the new situation, including dependencies.