I’ve been playing with eventstore and kubernetes, and got it running on kubernetes on gcloud a while back (http://blog.2mas.xyz/setting-up-event-store-with-kubernetes-on-google-cloud/). I haven’t used that in production, and am wondering if you think there might be an implications of doing so? I understand that I most likely get a performance hit, but that isn’t that important at the moment. I mainly concerned about possible data corruption, if that is a case I need to consider hosting it on kubernetes?
I don't know enough about the underlying storage in GCE to comment. I
would have to imagine though that they offer a form of persistent
damn you respond fast! In the post http://blog.2mas.xyz/setting-up-event-store-with-kubernetes-on-google-cloud/ I do use the persistent storage with X nodes, but I’m by no means an expert in evaluating if the storage mechanism at GCE.
To be fair the only to evaluate it is to do power pull testing which
you are not capable of doing.
It would be so nice if there where an event store as a service some where, would most likely pay for that
We are actually working on doing hosted clusters in about a week.
It will be AWS to start.
How about doing what the elasticSearch folks and GitHub have, providing a way of running ES in our own vpc's?
This would be really good!
It already exists. We even give TF (terraform) scripts as part of the
The TF scripts:
Setup Bastion node
Build out an ES cluster (3/5/etc nodes)
Setup an ASG for nodes
Setup ELB for admin interface
This is meant as the starting point (we don't know what you want from here!)
Would you consider making this a turnkey solution, I'm not at the stage where I need full commercial support, but need the scripts to sort the setup, happy to discuss privately!
We are looking next week at even offering the management of it.
FWIW, we are dong exactly that - running a 3 node EventStore cluster under Kubernetes on Google Cloud (GKE). We’re using SSD backed persistent disk.
Our transaction rates are still very modest, but it’s all been stable. An alternative for Google cloud would be to use local SSD and rely on backups/EventStore clustering.
Interesting, may I ask how you do deployment to kubernetes? Is it something similar to the post I linked to earlier.
Earlier you wrote that you need to be able to do power pull testing, how are you going to do that on AWS?
You can’t do it there either. You can however simulate various failures in most clouds by force detaching (without respect to mount status) whatever volumes you have. If you’re in the cloud you rarely have the on-premise situation of “power goes out” - you are more likely to suffer network partitions (across AZs in AWS) or just pain storage outages (in Azure). The tradeoffs and necessary configuration are different from the defaults which are intended for real hardware.
Usual things apply - if you care about your data, run ZFS.