Even with O_DIRECT and aligned blocks, I still don't understand how the storage engine can return a "successful commit" to the client without a sync at some point, because a sync (IIRC) is the only way to guarantee an ATA/NVMe FUA command is sent, and the device write cache/buffer is committed.
To step back a bit, the device still has a filesystem on it, and the structures described here are files within the filesystem? Just you're able to write directly into them, bypassing the filesystem layer, because you've constrained yourself to writes that don't require updating other parts of the filesystem structure?
Author here. This is not a general argument against fsync; the design depends on SSD-only deployment, preallocated files, O_DIRECT, single-key atomicity, and device write guarantees.
Your approach looks interesting but I was curious when you talk about path-based splitting for ART, do you literally mean always on "/"? I know S3 directory buckets always use /, but the classical S3 model had no natural separator character and I was wondering if supporting those styles of prefix or custom delimiter queries suffered any impediment in your approach.
Bookmarked your whole blog for later consumption, interesting stuff!
I wonder why this is not more common. LVM is easy to set up, and it's already common to allocate volumes for things like disk images for VMs, so why not databases?
Because the speed increase is - on modern, properly tuned filesystems - surprisingly small, due to how RDBMS's manage their pool; by working on large container files, they avoid most of the filesystem overhead.
Famously not, as the man page says.
It is also said later in the article:
> POSIX strictly requires a parent-directory fsync to make a newly created file’s existence durable.
So I'm not sure why the dirent sync is claimed earlier.
Bookmarked your whole blog for later consumption, interesting stuff!
[1] https://news.ycombinator.com/item?id=42805425