🐛(playbook) avoid downtime during switch or rollback

[madmatah]

Purpose

The switch_routes ansible task leave the current route in an invalid state for a few seconds when switch or rollback playbooks are executed. It can introduce an unnecessary downtime during a deployment.

For example, I have the hello app deployed with the following versions :

• next : d-200113-09h05m11s
• current : d-200113-08h58m09s
• previous : d-200113-08h55m23s

If I execute a switch, I can find this in my logs :

$ bin/switch -e "apps_filter=hello" | grep --line-buffered -i "patching routes with prefix" | ts '[%Y-%m-%d %H:%M:%.S]'

[2020-01-13 10:08:25.433209] TASK [Patching routes with prefix previous for the hello application with deployment_stamp d-200113-08h58m09s] ***
[2020-01-13 10:08:26.760792] TASK [Patching routes with prefix current for the hello application with deployment_stamp] ***
[2020-01-13 10:08:29.672389] TASK [Patching routes with prefix current for the hello application with deployment_stamp d-200113-09h05m11s] ***
[2020-01-13 10:08:30.687389] TASK [Patching routes with prefix next for the hello application with deployment_stamp] ***

Between 10:08:26 and 10:08:29, the current route is pointing to an undefined deployment stamp.

Proposal

Do not execute the second step in the above example.

[jmaupetit]

Good catch!