Posix Compatibility Comparison: GCP Filestore, Amazon EFS, and Azure Files
juicefs.com
juicefs.com
Imagine going to a hospital run like this, where the surgeons just do whatever they're told ("I got told to amputate, so I did. What do you mean check which leg to cut off?"), have zero interest in basic hygiene let alone antiseptic practices, and then angry patients can submit a piece of paper into a suggestion box with requests like: "Please cut the correct leg off! Also, it would be nice not to get a massive infection after surgery."
To which there will be a curt reply saying: "Customers can scale to additional legs, and they can purchase their own antibiotics if they so choose. Closed, will-not-fix."
It's a rhetorical question by the way. I already know the answer.
Yes it was.
>The context was that people have to tell support about bugs
The word "bug" does not appear in the post.
>but support can’t talk to developers, so you end up in a feature request forum.
Yes, you want features so you get directed to a feature request forum.
And let's not lose sight of the fact that this whole subthread arose because someone thinks that a cloud filesystem that doesn't preserve permissions bits is as evil as doctors killing people due to medical malpractice.
A small but frustrating example I experienced just today: In Azure ML, you can create data assets (basically a sort of centeal syslink to your actual data in a storage container) so that you can version your data and access it easily in your pipelines. Sounds great! But they recently pushed a new version that makes it impossible to create a data asset with multiple folders (!). You used to be able to do it, but now your data asset can either be a single file, a single folder (with no subfolders!), or a tabular asset.
It's completely insane to me because you now need to create a data asset for every folder you have. on a product made for data-heavy ML workloads. If you have data across a few dozens/hundreds of folders, you can't aggregate them even if they are to be used on the same pipeline, unless you do it manually. Also, now that you have dozens of unnecessary data assets, you can't even set your pipeline to take an arbitrary amount of them as input. You have 20 data assets/folders? Have fun adding 20 inputs in your component's YAML.
I can get around it, and maybe I'm dumb and all of our data should have been in a single folder anyway. But it used to work, with no downsides!
This stuff happens every day and can be especially aggravating in "heavier" services like AKS.
"Transcending POSIX: The End of an Era?"
https://www.usenix.org/publications/loginonline/transcending...
We will update the article to make it clear, thanks!
Now, ignoring the weird grammar of the article, I'm not sure what it's comparing considering it's looking at three different products each specifically claiming to support a different feature set. It's doubly confusing for Azure since they do have a CIFS/SMB offering as well as two different implementations of NFS and the author doesn't reveal which one he picked (of course samba won't map 1:1 to all features a *NIX system expects).
To me it looks like clickbait. The company behind this also pushed another marketing piece on the frontpage today https://news.ycombinator.com/item?id=34706227