Vagrant Orchestrate supports several deployment strategies that allow you to orchestrate the behavior pushes to remote servers. Here we'll cover how to use the various strategies as well as describing situations when each might be useful.
Deploy to the target servers one at a time. This can be useful if you have a small number servers, or if you need to keep the majority of your servers online in order to support your application's load.
$ vagrant orchestrate push --strategy serial
config.orchestrate.strategy = :serial
Deploy to all of the target servers at the same time. This is useful if you want to minimize the total amount of time that an deployment takes. Depending on how you've written your provisioners, this could cause downtime for the application that is being deployed.
$ vagrant orchestrate push --strategy parallel
config.orchestrate.strategy = :parallel
Deploy to a single server, pause to allow for testing, and then deploy the remainder of the servers in parallel. This is a great opportunity to test one node of your cluster before blasting your changes out to them all. This can be particularly useful when combined with post provision trigger to run a smoke test.
$ vagrant orchestrate push --strategy canary
config.orchestrate.strategy = :canary
The prompt can be surpressed with the --force (-f) flag.
Deploys to half of the cluster in parallel, then the other half, with a pause in between. This won't manage any of your load balancing or networking configuration for you, but if your application has a healthcheck that your load balancer respects, it should be easy to turn it off at the start of your provisioning and back on at the end. If your application can serve the load on half of its nodes then this will be the best blend of getting the deployment done quickly and maintaining a running application. If the total number of target servers is odd then the smaller number will be deployed to first.
$ vagrant orchestrate push --strategy half_half
config.orchestrate.strategy = :half_half
Combines the two immediately above - deploying to a single server, pausing, then to half of the remaining cluster in parallel, pausing, and then the other half, also in parallel. This is good if you have a large number of servers and want to do a smoke test of a single server before committing to pushing to half of your farm.
$ vagrant orchestrate push --strategy canary_half_half
config.orchestrate.strategy = :canary_half_half
Strategies can be passed on the command line with the --strategy parameter
$ vagrant orchestrate push --strategy parallel
Alternatively, you can specify the deployment strategy in your Vagrantfile
config.orchestrate.strategy = :parallel
Command line parameters take precedence over configuration values set in the Vagrantfile.
In order to automate the deployment process, you'll need to suppress prompts. You can achieve that in two ways:
From the command line, add the --force or -f parameters
$ vagrant orchestrate push --strategy canary -f
Within your Vagrantfile, set the force_push setting to true
config.orchestrate.force_push = true
If you have ideas for other strategies that you think would be broadly useful, open an issue and we'll discuss.