Flocker
jogster opened this issue · comments
Hi, love your work -- didn't realise it was possible to save/restore the state of etcd and start up without a discovery token.
I have a similar requirement to get a HA CoreOS cluster working in AWS, I had been apprehensive about the cluster getting damaged during a catastrophic reboot.
Having understood what you are doing, have you considered Flocker?
Since you are running the etcd_aws in a docker you will get the load/save into an EBS volume. Don't know if it will make much difference to your design but it removes the need for a named S3 bucket....
Thanks for the feedback.
For backups, the core issue is that the built-in backup provided by etcd includes both the data stored and the cluster state. If you have a complete failure of the cluster -- I think you don't want the cluster state restored.
In other words, the problem with built-in backup isn't how it is stored, it is what is stored.
After that, the question is where to store the backups, and S3 seems like a logical place, but that isn't essential to the approach.
Hope that helps and thanks for your comment!
Yes it does help me. I currently been playing with local coreos clusters (created 10 - 5 masters/5 slaves) but had noticed instability a few times, especially when trying to replace a master. I am using etcd as my bootstrap kv, then running consul against that and using zookeeper to power my mesos cluster.
Makes me think that I probably need to save consul and zookeeper as they all help to bootstrap the cluster.
I do not have a huge experience with go-lang (at the moment) but was wondering if there was a possibility of reading/writing the data for etcd_aws from a mounted share?
This is so that I can test it locally on my mac with vagrant (it exposes a mount into the machine) and get a good understanding before trying to grapple with it on AWS.