QUESTION: Can the CRIU library be packaged along with the garden-runc release? about garden-runc-release HOT 4 CLOSED

We have created an issue in Pivotal Tracker to manage this:

https://www.pivotaltracker.com/story/show/156751366

The labels on this github issue will be updated when the story is started.

from garden-runc-release.

Hey @abh1kg, garden doesn't currently support checkpoint and restore for containers right now, so there's no need for us to package up criu with the release. We do occasionally get asked about this though. Out of interest what are you looking to with checkpoint/restore?

from garden-runc-release.

@teddyking One of the reasons I wanted to use criu was to be able to migrate in-flight TCP connections to another up-and-coming Garden container as part of a blue-green deployment workflow. Such TCP connections include stuff like SSH connections/Websocket connections etc.
Out of curiosity, why doesn't Garden support checkpoint and restore though? Since it is runc-compliant and runc supports criu operations out of the box, I was assuming it to be also available for Garden containers. I can understand the entire workflow and support not being there directly from Cloud Foundry but wouldn't one be able to log on to a diego cell and run the checkpoint operations manually?

from garden-runc-release.

Hi @abh1kg, there's no reason you couldn't manually run criu (as you say it's all just runc under the covers!), but it's not something we support or that Diego/cloud foundry exposes to users. The reason for that is the CF model is to have stateless apps so we can automatically patch and manage the cluster easily. You can do a blue/green deployment by deploying the new version of your application in CC and then swapping the route over using cf map-route without needing checkpoint/restore. Also, when hosts are upgraded, Diego first ensures there's enough new instances of your apps on other cells before shutting down a cell to upgrade. Since all the instances should be equivalent and stateless there's generally no need in CF to worry about a serialising the memory of one particular container of an app, and it introduces a lot of complexity to the scheduling and container lifecycle if we have to deal with rescheduling running containers instead of treating them as immutable, stateless and replaceable.

Having said that if you want to experiment with criu on a garden cell there's no obvious reason it won't work (you'll have to install the binary yourself, perhaps in your own bosh release), and feel free to drop by the #garden channel on cf slack if we can help!

from garden-runc-release.