feat: Support Windows for Warm Pools

[backjo]

Adds to #277

[codecov]

Codecov Report

Merging #290 (6cf774b) into master (dcabfb3) will increase coverage by 0.06%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #290      +/-   ##
==========================================
+ Coverage   70.02%   70.09%   +0.06%     
==========================================
  Files          19       19              
  Lines        3096     3103       +7     
==========================================
+ Hits         2168     2175       +7     
  Misses        788      788              
  Partials      140      140              

Impacted Files	Coverage Δ
controllers/provisioners/eks/helpers.go	91.46% <100.00%> (+0.08%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update dcabfb3...6cf774b. Read the comment docs.

[backjo]

Got around to testing this on the EKS recommended Windows AMI - it works!

Boot time for warm instances is around 2-3 minutes, compared to 7 for fresh instances. Might even be shorted without the custom PreBootstrap scripts that we have.

[eytan-avisror]

Got around to testing this on the EKS recommended Windows AMI - it works!

Boot time for warm instances is around 2-3 minutes, compared to 7 for fresh instances. Might even be shorted without the custom PreBootstrap scripts that we have.

Wow thats awesome, AL2 is kinda light weight so we only saw 1m instead of 2m approx, but still a 40-50% improvement.
Nice 💪

[backjo]

Yeah - I imagine BottleRocket will be even less of an improvement than AL2, since it's already starting up in 90 seconds or so.

[eytan-avisror]

Yeah - I imagine BottleRocket will be even less of an improvement than AL2, since it's already starting up in 90 seconds or so.

Well, AWS provisioning the instance takes around 45-60 seconds (which you can potentially save by warming), booting up bottlerocket machine in 30 seconds, and making it ready in another 30 will be pretty damn impressive.

[backjo]

Yeah - I imagine BottleRocket will be even less of an improvement than AL2, since it's already starting up in 90 seconds or so.

Well, AWS provisioning the instance takes around 45-60 seconds (which you can potentially save by warming), booting up bottlerocket machine in 30 seconds, and making it ready in another 30 will be pretty damn impressive.

For real - I just clocked a BottleRocket node (without warm pool) from

1. Manually updating the desired capacity on the ASG
2. The Kubelet reporting to the API server
3. The Node being ready on the API server

1-2 took 41 seconds
2-3 took 30 seconds

For a total of 71 seconds from ASG capacity change to being ready in K8S 😱 .

Given the design of BottleRocket with it's read-only root file system and the already bleeding-fast startup times, it probably doesn't make sense to even attempt to support warm pools for it.

[eytan-avisror]

Yeah - I imagine BottleRocket will be even less of an improvement than AL2, since it's already starting up in 90 seconds or so.

Well, AWS provisioning the instance takes around 45-60 seconds (which you can potentially save by warming), booting up bottlerocket machine in 30 seconds, and making it ready in another 30 will be pretty damn impressive.

For real - I just clocked a BottleRocket node (without warm pool) from

1. Manually updating the desired capacity on the ASG

2. The Kubelet reporting to the API server

3. The Node being ready on the API server

1-2 took 41 seconds
2-3 took 30 seconds

For a total of 71 seconds from ASG capacity change to being ready in K8S 😱 .

Given the design of BottleRocket with it's read-only root file system and the already bleeding-fast startup times, it probably doesn't make sense to even attempt to support warm pools for it.

Might reduce that by a good 20-30 seconds with warm pools :) don't forget you are completely removing the time it takes Amazon to simply create and propagate the instance object itself.
Still worth exploring given the cost of running this is almost free.

BTW, did you push the change for <persist>?

[backjo]

Just pushed the persist change

[eytan-avisror]

🏆