improve docker_* test speed

[felixfontein]

SUMMARY

Trying to speed up docker_* tests. The longest are (taken from a CI run on RHEL7):

• docker_container: ~8 minutes
• docker_node: ~7 minutes
• docker_swarm_service: ~5 minutes
• docker_network: ~3.5 minutes

First try: using resolve_image: no for docker_swarm_service, which should stop registry lookups for the image (with a new enough docker daemon).

ISSUE TYPE

• Bugfix Pull Request

COMPONENT NAME

docker_swarm_service

[felixfontein]

docker_swarm_service is now down to 2 1/2 minutes, docker_node to 30 seconds. I'll try to see whether I can improve docker_network and docker_container tomorrow.

[felixfontein]

The problem with docker_network is that creating a network with IPAM config takes 15 seconds. Happens both on my local machine and on the RHEL7 CI VM. Here's a profile_tasks result for the docker_network tests:

Wednesday 06 March 2019  22:23:05 +0100 (0:00:00.019)       0:03:01.707 ******* 
=============================================================================== 
Change subnet of network with deprecated custom IPAM config ------------------------------------------------------------------------------ 16.04s
Change subnet of network with IPv6 IPAM config ------------------------------------------------------------------------------------------- 15.99s
Remove IPv6 from network with custom IPv4 and IPv6 IPAM config --------------------------------------------------------------------------- 15.99s
Change subnet, gateway, iprange and auxiliary addresses of network with custom IPAM config ----------------------------------------------- 15.94s
Create network with IPv6 and custom IPv4 IPAM config ------------------------------------------------------------------------------------- 15.88s
Create network with IPv6 IPAM config ----------------------------------------------------------------------------------------------------- 15.86s
Create network with deprecated custom IPAM config ---------------------------------------------------------------------------------------- 15.81s
Create network with custom IPAM config --------------------------------------------------------------------------------------------------- 15.80s
Create containers ------------------------------------------------------------------------------------------------------------------------- 4.14s
Make sure all networks are removed -------------------------------------------------------------------------------------------------------- 3.96s
Delete containers ------------------------------------------------------------------------------------------------------------------------- 3.11s
Cleanup ----------------------------------------------------------------------------------------------------------------------------------- 1.40s
Make sure all containers are removed ------------------------------------------------------------------------------------------------------ 1.39s
Gathering Facts --------------------------------------------------------------------------------------------------------------------------- 1.37s
swarm ------------------------------------------------------------------------------------------------------------------------------------- 1.36s
swarm ------------------------------------------------------------------------------------------------------------------------------------- 1.29s
Cleanup ----------------------------------------------------------------------------------------------------------------------------------- 1.06s
Create network ---------------------------------------------------------------------------------------------------------------------------- 1.02s
Connect network to container 3 ------------------------------------------------------------------------------------------------------------ 0.94s
Connect network to containers 1 and 2 ----------------------------------------------------------------------------------------------------- 0.92s

As you can see, by far the most time was eaten by creating networks with IPAM config. I'm not sure we can really improve this (except by removing tests).

[ansibot]

cc @DBendit @WojciechowskiPiotr @akshay196 @danihodovic @dariko @hannseman @jwitko @kassiansun @tbouvet
click here for bot help

[felixfontein]

@WojciechowskiPiotr @dariko @hannseman @jwitko FYI: we merged this pretty quickly because a) tests were still passing, and b) tests ran a lot faster, so hopefully there won't be a timeout on the nightly coverage test this time. Please still take a look at these changes if they are OK from your point of view! If not, we can always change some of them back :)

@WojciechowskiPiotr I think the change to module_utils/docker/swarm.py should be harmless: it simply leves the last sleep away in case the node is not down; this sleep was executed before returning, and no further check was done after it, so I don't think skipping it should have any impact, except that the module feels much faster now in most situations (since it usually takes 5 seconds less time).

@dariko @hannseman @jwitko I've added resolve_image: no to almost all docker_swarm_service tests. On my machine, with docker >= 3.2, this improves runtime a lot. With docker >= 3.2 and resolve_image: yes, essentially the module looks up the image at the registry's side to determine whether it needs to be updated. Which obviously takes a bit of time, and which isn't needed during the integration tests (and which actually is counterproductive: in case the image changes during the tests, the service might be updated/restarted during some idemptence check where the tests want to make sure it's not changed). The only place where I didn't add it are the tests for resolve_image itself. I hope I didn't break anything with this.

[dariko]

@felixfontein Looks good, thank you!