Support for AWS "spot" instances
Bug #945862 reported by
Jorge Castro
This bug affects 8 people
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
Canonical Juju |
Triaged
|
Low
|
Unassigned |
Bug Description
Since juju is asynchronous I can deploy what I want quickly after a bootstrap. I wouldn't mind waiting a bit extra for the instances to fire up if I'm really cheap, especially if it's something I can tear up and down whenever I want. We should perhaps support AWS spot instances. Something like this in the environments.yaml maybe:
default-
default-
spot-price-max: .060
Or something.
Changed in juju: | |
status: | New → Confirmed |
importance: | Undecided → Wishlist |
no longer affects: | juju-core |
no longer affects: | juju |
tags: | added: adoption |
Changed in juju-core: | |
status: | New → Triaged |
importance: | Undecided → Low |
Changed in juju-core: | |
status: | Triaged → Won't Fix |
affects: | juju-core → juju |
Changed in juju: | |
importance: | Low → Wishlist |
status: | Won't Fix → Triaged |
Changed in juju: | |
status: | Confirmed → Triaged |
To post a comment you must log in.
Spot instance support for AWS provisioner is something we are looking very much for.
The idea is: especially with recent adoption of Juju in Apache Software Foundation (Apache BigTop, Hadoop, Spark, Zeppelin, etc) more and more people will start looking into Juju as an open source option for their hobby\part-time projects and data analytics gigs and those are people who love opensource in part because it's free, so cost is very important factor here and this value proposition will be dear to their hearts.
And then the same people will bring it thought the doors of their organizations to daytime jobs later on.
Although it's been a while since the issue was raised, it looks like a great time now to gauge the priority a bit and implement it.
It would be awesome to control the price using constraints (same as other resources of CPU and memory) and be able to user `add-unit` to scale the cluster.
It would also make sense to be able mix on-demand and spot instances to be able to have enough on-demand instances to keep the service running (i.e HDFS namenode, yarn resource manager, etc) but allow workers to come and go to keep it really cheap.