Rolling Updates
When a Release is promoted, new processes are gracefully rolled out into production.
If there are errors starting new processes, new processes are not verified as healthy, or the rollout doesn't complete in 10 minutes, an automatic rollback is performed.
The number of new processes started at a time is configurable so large Apps can be fully rolled out in 10 minutes.
Rollouts and rollbacks do not cause any service downtime.
Rollout
A rollout coordinates starting new processes in a way that maintains Service uptime and capacity. The basic flow is:
- Start 1 new process
- Verify new process is healthy
- Stop 1 old process
- Repeat for all processes in the formation
Deployment Configuration
Rolling updates respect the deployment configuration to control the minimum number of healthy processes and maximum number of overall processes running at any one time during the update. These values are percentages of the desired count. The defaults are a minimum of 50% and a maximum of 200%.
You can configure these values in your convox.yml:
services:
web:
deployment:
minimum: 25
maximum: 125
In this example, if the Service has a count of 4, a minimum of 25% means at least 1 process must remain healthy during the update, and a maximum of 125% means at most 5 processes can exist simultaneously.
For Agent or Singleton Services, the defaults are 0% minimum and 100% maximum.
Automatic Rollback
If there are errors starting new processes, new processes are not verified as healthy, or the rollout doesn't complete in 10 minutes, an automatic rollback is performed. This results in a rollback or failed state for the App:
$ convox apps info
Name httpd
Status rollback
$ convox apps info
Name httpd
Status failed
Health Checks
For a rolling update to succeed, certain criteria must be met. For more information, see Health Checks.
See Also
