Comment 16 for bug 1755897

The bulk of the regression in deployment time appears to come from juju itself. From observation it seems that each execution of juju hook tools take significantly longer time with recent versions of juju compared to older versions.

Data:
Model contains 3-node keystone cluster, mysql and two instances of glance. (bundle: https://pastebin.ubuntu.com/p/jrpcMs7Qq8/)
Deployment runtime measured on non-leader unit from execution of first hook after `install` hook.
Deployment deemed completed when execution of final `cluster-relation-changed` hook completes.

Baseline: juju 2.3.5, stable/18.02 charms, juju model-config logging-config=<root>=WARNING;unit=DEBUG
1. run
  * start `2018-04-10 11:01:21`
  * end `2018-04-10 11:21:38`
  * time 20m17s (1217s)
2. run
  * start `2018-04-10 11:56:07`
  * end `2018-04-10 12:22:04`
  * time 26m57s (1617s)
 3. run
  * start `2018-04-10 12:35:18`
  * end `2018-04-10 12:58:47`
  * time 23m29s (1409s)

 juju 2.2.6, stable/18.02 charms, juju model-config logging-config=<root>=WARNING;unit=DEBUG
 1. run
   * start `2018-04-11 06:01:21`
   * end `2018-04-11 06:14:05`
   * time 13m16s (796s)
 2. run
   * start `2018-04-11 06:29:06`
   * end `2018-04-11 06:40:30`
   * time 11m24s (684s)
 3. run
   * start `2018-04-11 06:56:06`
   * end `2018-04-11 07:07:12`
   * time 11m6s (666s)