5
netikras
14d

I've never been a big fan of the "Cloud hype".

Take today for example. What decent persistent storage options do I have for my EKS cluster?

- EBS -- does not support ReadWriteMany, meaning all the pods mounting that volume will have to be physically running on the same server. No HA, no HP. Bummer

- EFS -- expensive. On top of that, its performance is utter shit. Sure, I could buy more IOPS, but then again.. even more expensive.

S3 -- half-assed filesystem. Does not support O_APPEND, so basically any file modifications will have to be in a
`createFile(file+"_new", readAll(file) + new_data); removeFile(file); renameFile(file + "_new", file);`
way.
ON TOP of that, the s3 CSI has even more limitations, limiting my ability to cross-mount volumes across different applications (permission issues)

I'm running out of options. And this does not help my distrust in cloud infras...

Comments
  • 3
    yeah, switching EFS from generalPurpose to maxIO was the change I needed. So yeah, getting ready for the $$$ rainfall :)

    damn it... I never thought decent storage was such a lobbyist topic in 2024
  • 4
    I see the cloud as the same bullshit which happened with real estate. You buy all the land and rent out the houses. Keep the prices high so the people can’t afford owning the resources themselves.
  • 0
    instead of being programmers this sounds like being experts on gouging opportunities of other companies

    miserable

    you're just a business consultant now
  • 0
    @jestdotty what / who are you referring to?
  • 0
    EBS sucks because its not replicated across AZs. I used EFS, but for performance and price having some sort of ebs Replicator service running in the cluster would be dope.

    All that being said cloud is still great. Good luck getting data copying across seperate AZs on prem. Just build 3 data centers... Lol
  • 1
    @lungdart sure. If I stayed 2 nights in a certain city I’d take a hotel.

    Problem is, I’m seeing it being sold to companies which have a very predictable server load, and for that cloud seriously is bullshit. Why would I trust companies like Amazon, google or ESPECIALLY Microcuck with keeping my infrastructure operational, let alone secure. Microcuck can’t even do that with their own products…
  • 0
    @Chewbanacas cloud makes sense for all but the bigger companies and the smallest companies.

    If you're company can run on a on Pentium 2 teetering on the edge of a shelf, go for it.

    But if you need uptime SLAs of anything with a 9 in it, you need redundant hardware, internet, power, and cooling. Then you have to hire operational and network engineers, don't forget observability!

    I'd rather an ec2 instance and an ALB any day.
  • 0
    The biggest problem here is that what you are trying to do is either not suited for object storage or your software is not designed for the cloud.

    You can always use cloud building blocks to build your own Ceph server as this seems to be more like what you want. It's a bit like what you would need to do in your own rack but with cloud scalability with EKS (as that seems to add value to you).

    I found GKE with filestore to be much simpler for this by the way. Had a similar issue with a scalable EC2 setup (and traditional CMS needing a real file system). Moving that to GKE was nicer.
  • 0
    @Chewbanacas but in this cas, owning your own servers is in fact cost efficiënt, that is, if you invest time in tuning them and be happy with what you got.
  • 0
    @hjk101 Interesting...

    I faced this issue when I deployed Loki+promtail to my EKS cluster and misconfigured Loki's s3 IAM roles, so it accumulated quite some data locally.

    I wouldn't rush into claiming that Loki is not suitable for cloud (afaik it's under the CNCF wing) and/or that I should use something else but a filesystem to store Loki's... files.
Add Comment