> One of the reasons why these patches are not included in the kernel is that the free space calculations do not work properly.
It seems these patches possibly fix that.
sandreas · 8h ago
Well, first of all: I'm not trying to bash BTRFS at all, it probably is just not meant for me. However, I'm trying to gain information it is really considered stable (like rock solid) or it might have been a hardware Problem on my system.
I used cryptsetup with BTRFS because I encrypt all of my stuff. One day, the system froze and after reboot the partition was unrecoverably gone (the whole story[1]). Not a real problem because I had a recent backup, but somehow I lost trust in BTRFS that day. Anyone experienced something like that?
Since then I switched to ZFS (on the same hardware) and never had problems - while it was a real pain to setup until I finished my script [2], which still is kind of a collection of dirty hacks :-)
Yes, my story with btrfs is quite similar- used it for a couple years, suddenly threw some undocumented error and refused to mount, asked about it on the dev irc channel and was told apparently it was a known issue with no solution, have fun rebuilding from backups. No suggestion that anyone was interested in documenting this issue, let alone fixing it.
These same people are the only ones in the world suggesting btrfs is "basically" stable. I'll never touch this project again with a ten foot pole, afaic it's run by children. I'll trust adults with my data.
sandreas · 7h ago
Ok, thank you. At least I'm not alone with this. However, I'm not too much into it and would not go as far to say it's not a recommendable project, but boy was I mad it just died without any way to recover ANYTHING :-)
pa7ch · 7h ago
I worked on a linux distro some years ago that had to pull btrfs long after people had started saying thats its truly solid because customers had so many issues. Its probably improved since but its hard to know. Im surprised fedora workstation defaults to it now. I'm hoping bcachefs finds its way in the next few years as being the rock solid fs it aims to be.
BlimpSpike · 4h ago
I hadn't heard of bcachefs, but I looked it up and apparently Linus just removed it from the kernel source tree last month for non-technical reasons.
He hasn't, yet, but it's anyone's guess what's he's going to do.
Regardless, development isn't going to stop, we may just have to switch to shipping as a DKMS module. And considering the issues we've had with getting bugfixes out that might have been the better way all along.
sandreas · 7h ago
Yeah what really made me wonder is that I thought I had incomplete and wrong manpages in the recovery sections... examples did not work as described, but I can't remember what it was, I was too mad and ditched it completely :-)
fpoling · 7h ago
Have you used 4K sectors with cryptsetup? Many distributions still defaults to 512 bytes if SSD reports 512 bytes as its logical size and with 512 sectors there are heavier load on the system.
I was reluctant to use BTRFS on my Linux laptop but for the last 3 years I have been using it with 4K cryptsetup with no issues.
sandreas · 7h ago
I used the default archinstall... did not check the sector size, but good to hear it works for you. Maybe I'll check again with my next setup.
homebrewer · 3h ago
FWIW, basically all of arch linux' infrastructure has been running on top of btrfs for several years, and last time I asked them, they didn't have any more problems with it than with any other filesystem.
I've used it as my desktops main filesystem for many years and not had any problems. I have regular snapshots with snapper. I run the latest kernel, so ZFS is not an option.
That said, I avoid it like the plague on servers, to get acceptable performance (or avoid fragmentation) with VMs or databases you need to disable COW which disables many of it's features, so it's better just to roll with XFS (and get pseudo-snapshots anyway).
homebrewer · 3h ago
In the unlikely case you're running SQLite, it's possible to get okay performance on btrfs too:
My btrfs filesystem has been slowly eating my data for a while; large files will find their first 128k replaced with all nulls. Rewriting it will sometimes fix it temporarily, but it'll revert back to all nulls after some time. That said, this might be my fault for using raid6 for data and trying to replace a failing disk a while ago.
homebrewer · 3h ago
raid 5/6 is completely broken and there's not much interest in fixing it — nobody who's willing to pay for its development (which includes Facebook, SUSE, Oracle, and WD) uses raid 5/6; you shouldn't have been running it in the first place. I understand it's basically blaming the victim, but doing at least some research on the filesystem before starting to use it is a good idea in any case.
edit: just checked, it says the same thing in man pages — not for production use, testing/development only.
riku_iki · 7h ago
> One day, the system froze and after reboot the partition was unrecoverably gone (the whole story[1]).
it looks like you didn't use raid, so any FS could fail in case of disk corruption.
sandreas · 7h ago
Thank you for your opinion. Well... it did not just fail. Cryptsetup mounted everything fine, but the BTRFS tools did not find a valid filesystem on it.
While it could have been a bit flip that destroyed the whole encryption layer, BTRFS debugging revealed that there was some traces of BTRFS headers after mounting cryptsetup and some of the data on the decrypted partition was there...
This probably means the encryption layer was fine. The BTRFS part just could not be repaired or restored. The only explanation I have for this that something resulted in a dirty write, which destroyed the whole partition table, the backup partition table and since I used subvolumes and could not restore anything, most of the data.
Well, maybe it was my fault but since I'm using the exact same system with the same hardware right now (same NVMe SSD), I really doubt that.
riku_iki · 6h ago
> Well, maybe it was my fault but since I'm using the exact same system with the same hardware right now (same NVMe SSD), I really doubt that.
anecdotes could be exchanged in both directions: I run heavy data processing with max possible throughput on top of btrfs raid for 10 years already, and never had any data loss. I am absolutely certain if you expect data integrity while relying on single disk: it is your fault.
ahofmann · 5h ago
What the hell are you talking about? Any filesystem on any OS I've seen the last 3 decades had some kind of recovery path after any crash. Some of them lose more data, some of them less. But being unable to mount, is a bug that makes a filesystem untrustworthy and useless.
And how would RAID help in that situation?
riku_iki · 5h ago
> But being unable to mount, is a bug that makes a filesystem untrustworthy and useless.
we are in disagreement on this. If partition table entry corrupted, you can't mount without some low level surgery.
> And how would RAID help in that situation?
depending on raid, your data will be duplicated on another disk, and will survive in case of one/few disks corruption.
bjoli · 15h ago
I wonder if I can use a smaller SSD for this and make it avoid HDD wakeups due to some process reading metadata. That alone would make me love this feature.
the8472 · 14h ago
I think you'd rather want a cache device (or some more complicated storage tiering) for that so that both metadata and frequently accessed files get moved to that dynamically based on access patterns.
Afaik btrfs doesn't support that. LVM, bcache, device mapper, bcachefs and zfs support that (though zfs would require separate caches for reading and synchronous write). And idk which of these let you control the writeback interval.
Most likely yes, but the also envisioned periodically repacking oft multiple small data extents into one big that gets written to the HDD would wake up the HDD. And if you'd make the SSD "metadata only", browser cache and logging will keep the HDD spinning.
This feature is for performance, not the case you described.
ajross · 13h ago
Just buy more RAM and you get that for free. Really I guess that's my sense of patches like this in general: while sure, filesystem research has a long and storied history and it's a very hard problem in general that attracts some of the smartest people in the field to do genius-tier work...
Does it really matter in the modern world where a vanilla two-socket rack unit has a terabyte of DRAM? Everything at scale happens in RAM these days. Everything. Replicating across datacenters gets you all the reliability you need, with none of the fussing about storage latency and block device I/O strategy.
bayindirh · 11h ago
Actually, it doesn't work like that.
Sun's ZFS7420 had a terabyte of RAM per controller, and these work in tandem, and after a certain pressure, the thing can't keep up even though it also uses specialized SSDs to reduce HDD array access during requests, and these were blazingly fast boxes for their time.
When you drive a couple thousand physical nodes with a some-petabytes sized volumes, no amount of RAM can save you. This is why Lustre divides metadata servers and volumes from file ones. You can keep very small files in metadata area (a-la Apple's 0-sized, data-in-resource-fork implementation), but for bigger data, you need to have good filesystems. There are no workarounds from this.
If you want to go faster, take a look at Weka and GPUDirect. Again, when you are pumping tons of data to your GPUs to keep them training/inferring, no amount of RAM can keep that data (or sustain the throughput) during that chaotic access for you.
When we talked about performance, we used to say GB/sec. Now a single SSD provides that IOPS and throughput provided by storage clusters. Instead, we talk about TB/sec in some cases. You can casually connect terabit Ethernet (or Infiniband if you prefer that) to a server with a couple of cables.
ajross · 10h ago
> When you drive a couple thousand physical nodes with a some-petabytes sized volumes
You aren't doing that with ZFS or btrfs, though. Datacenter-scale storage solutions (c.f. Lustre, which you mention) have long since abandoned traditional filesystem techniques like the one in the linked article. And they rely almost exclusively on RAM behavior for their performance characteristics, not the underlying storage (which usually ends up being something analogous to a pickled transaction log, it's not the format you're expected to manage per-operation)
bayindirh · 10h ago
> You aren't doing that with ZFS or btrfs, though.
ZFS can, and is actually designed to, handle that kind of workloads, though. At full configuration, ZFS7420 is a 84U configuration. Every disk box has its own set of "log" SSDs and 10 additional HDDs. Plus it was one of the rare systems which supported Infiniband access natively, and was able to saturate all of its Infiniband links under immense load.
Lustre's performance is not RAM bound when driving that kind of loads, this is why MDT arrays are smaller and generally full-flash while OSTs can be selected from a mix of technologies. As I said, when driving that number of clients from a relatively small number of servers, it's not possible to keep all the metadata and query it from the RAM. Yes, Lustre recommends high RAM and core count for servers driving OSTs, but it's for file content throughput when many clients are requesting files, and we're discussing file metadata access primarily.
ajross · 10h ago
Again I think we're talking past each other. I'm saying "traditional filesystem-based storage management is not performance-limited at scale where everything is in RAM, so I don't see value to optimizations like that". You seem to be taking as a prior that at scale everything doesn't fit in RAM, so traditional filesystem-based storage management is still needed.
But... everything does fit in RAM at scale. I mean, Cloudflare basically runs a billion dollar business who's product is essentially "We store the internet in RAM in every city". The whole tech world is aflutter right now over a technology base that amounts to "We put the whole of human experience into GPU RAM so we can train our new overlords". It's RAM. Everything is RAM.
I'm not saying there is "no" home for excessively tuned genius-tier filesystem-over-persistent-storage code. I'm just saying that it's not a very big home, that the market has mostly passed the technology over, and that frankly patches like the linked article seem like a waste of effort to me vs. going to Amazon and buying more RAM.
piperswe · 7h ago
Cloudflare's cache is a tiered cache with RAM and SSDs, not just RAM.
> Our storage layer, which serves millions of cache hits per second globally, is powered by high IOPS NVMe SSDs.
j16sdiz · 9h ago
These patches came from oracle.
Pretty sure they have a client somewhere needs this.
bayindirh · 7h ago
No, it doesn't. You think in a very static manner. Yes, you can fit websites in RAM, but you can't fit the databases powering them. Yes, you can fit some part of the videos or images you're working on or serving on RAM, but you can't store whole catalogs in RAM.
Moreover, you again give examples from the end product. Finished sites, compacted JS files, compressed videos, compiled models...
There's much more than that. The model is in RAM, but you need to rake tons of data over that GPU. Sometimes terabytes of data. You have raw images to process, raw video to color-grade, unfiltered scientific data to sift through. These files are huge.
A well processed JPG from my camera is around 5MB, but RAW version I process is 25MB per frame, and it's a 24MP image, puny for today's standards. Your run of the mill 2K video takes a couple of GBs after final render at movie length. RAWs take 10s of terabytes, at minimum. Unfiltered scientific data again comes in terabytes to petabytes range depending on your project and instruments you work on, and multiple such groups pull their own big datasets to process real-time.
In my world, nothing fits in RAM except the runtime data, and that's your application plus some intermediate data structures. The rest is read from small to gigantic files and written in files of unknown sizes, by multiple groups, simultaneously. These systems experience the real meaning of "saturation", and they would really swear at us at some cases.
Sorry, but you can't solve this problem by buying more RAM, because these workloads can't be carried to clouds. They need to be local, transparent and fast. IOW, you need disk systems which feel like RAM. Again, look what Weka (https://www.weka.io/) does. It's one of the most visible companies which make systems behave like a huge RAM, but with multiple machines and tons of cutting edge SSDs, because what they process doesn't fit in RAM.
Lastly, oh, there's a law which I forget its name every time, which tells you if you cache 10 most used files, you can serve up to 90% of your requests from that cache, if your request pattern is static. In cases I cite, there's no "popular" file. Everybody wants their own popular files which makes access "truly random".
guenthert · 12h ago
Some time ago (back when we were using spinning rust) I was wondering whether one could bypass the latency of disk access when replicating to multiple hosts. I mean, how likely is it, that two hosts crash at the same time? Well, it turns out that there are some causes which take out multiple hosts simultaneously (a way too common occurrence seems to be diesel generators which fail to start after power failure). I think the good fellas at Amazon, Meta and Google even have stories to tell about a whole data center failing. So you need replication across data centers, but then network latency bites ya. Current NVMe storage devices are then faster (and for some access patterns nearly as fast as RAM).
And that's just at the largest scale. I'm pretty sure banks still insist that the data is written to (multiple) disks (aka "stable storage") before completing a transaction.
homebrewer · 13h ago
> Does it really matter in the modern world
Considering that multiple ZFS developers get paid to make ZFS work well on petabyte-sized disk arrays with SSD caching, and one of them often reports on progress in this area in his podcasts (2.5admins.com and bsdnow if you're interested) .. then yes?
- https://github.com/kakra/linux/pull/36
- https://wiki.tnonline.net/w/Btrfs/Allocator_Hints
What do you think?
It seems these patches possibly fix that.
I used cryptsetup with BTRFS because I encrypt all of my stuff. One day, the system froze and after reboot the partition was unrecoverably gone (the whole story[1]). Not a real problem because I had a recent backup, but somehow I lost trust in BTRFS that day. Anyone experienced something like that?
Since then I switched to ZFS (on the same hardware) and never had problems - while it was a real pain to setup until I finished my script [2], which still is kind of a collection of dirty hacks :-)
1: https://forum.cgsecurity.org/phpBB3/viewtopic.php?t=13013
2: https://github.com/sandreas/zarch
These same people are the only ones in the world suggesting btrfs is "basically" stable. I'll never touch this project again with a ten foot pole, afaic it's run by children. I'll trust adults with my data.
https://en.wikipedia.org/wiki/Bcachefs#History
Regardless, development isn't going to stop, we may just have to switch to shipping as a DKMS module. And considering the issues we've had with getting bugfixes out that might have been the better way all along.
I was reluctant to use BTRFS on my Linux laptop but for the last 3 years I have been using it with 4K cryptsetup with no issues.
https://gitlab.archlinux.org/archlinux/infrastructure
That said, I avoid it like the plague on servers, to get acceptable performance (or avoid fragmentation) with VMs or databases you need to disable COW which disables many of it's features, so it's better just to roll with XFS (and get pseudo-snapshots anyway).
https://wiki.tnonline.net/w/Blog/SQLite_Performance_on_Btrfs
https://btrfs.readthedocs.io/en/latest/Status.html
edit: just checked, it says the same thing in man pages — not for production use, testing/development only.
it looks like you didn't use raid, so any FS could fail in case of disk corruption.
While it could have been a bit flip that destroyed the whole encryption layer, BTRFS debugging revealed that there was some traces of BTRFS headers after mounting cryptsetup and some of the data on the decrypted partition was there...
This probably means the encryption layer was fine. The BTRFS part just could not be repaired or restored. The only explanation I have for this that something resulted in a dirty write, which destroyed the whole partition table, the backup partition table and since I used subvolumes and could not restore anything, most of the data.
Well, maybe it was my fault but since I'm using the exact same system with the same hardware right now (same NVMe SSD), I really doubt that.
anecdotes could be exchanged in both directions: I run heavy data processing with max possible throughput on top of btrfs raid for 10 years already, and never had any data loss. I am absolutely certain if you expect data integrity while relying on single disk: it is your fault.
And how would RAID help in that situation?
we are in disagreement on this. If partition table entry corrupted, you can't mount without some low level surgery.
> And how would RAID help in that situation?
depending on raid, your data will be duplicated on another disk, and will survive in case of one/few disks corruption.
This feature is for performance, not the case you described.
Does it really matter in the modern world where a vanilla two-socket rack unit has a terabyte of DRAM? Everything at scale happens in RAM these days. Everything. Replicating across datacenters gets you all the reliability you need, with none of the fussing about storage latency and block device I/O strategy.
Sun's ZFS7420 had a terabyte of RAM per controller, and these work in tandem, and after a certain pressure, the thing can't keep up even though it also uses specialized SSDs to reduce HDD array access during requests, and these were blazingly fast boxes for their time.
When you drive a couple thousand physical nodes with a some-petabytes sized volumes, no amount of RAM can save you. This is why Lustre divides metadata servers and volumes from file ones. You can keep very small files in metadata area (a-la Apple's 0-sized, data-in-resource-fork implementation), but for bigger data, you need to have good filesystems. There are no workarounds from this.
If you want to go faster, take a look at Weka and GPUDirect. Again, when you are pumping tons of data to your GPUs to keep them training/inferring, no amount of RAM can keep that data (or sustain the throughput) during that chaotic access for you.
When we talked about performance, we used to say GB/sec. Now a single SSD provides that IOPS and throughput provided by storage clusters. Instead, we talk about TB/sec in some cases. You can casually connect terabit Ethernet (or Infiniband if you prefer that) to a server with a couple of cables.
You aren't doing that with ZFS or btrfs, though. Datacenter-scale storage solutions (c.f. Lustre, which you mention) have long since abandoned traditional filesystem techniques like the one in the linked article. And they rely almost exclusively on RAM behavior for their performance characteristics, not the underlying storage (which usually ends up being something analogous to a pickled transaction log, it's not the format you're expected to manage per-operation)
ZFS can, and is actually designed to, handle that kind of workloads, though. At full configuration, ZFS7420 is a 84U configuration. Every disk box has its own set of "log" SSDs and 10 additional HDDs. Plus it was one of the rare systems which supported Infiniband access natively, and was able to saturate all of its Infiniband links under immense load.
Lustre's performance is not RAM bound when driving that kind of loads, this is why MDT arrays are smaller and generally full-flash while OSTs can be selected from a mix of technologies. As I said, when driving that number of clients from a relatively small number of servers, it's not possible to keep all the metadata and query it from the RAM. Yes, Lustre recommends high RAM and core count for servers driving OSTs, but it's for file content throughput when many clients are requesting files, and we're discussing file metadata access primarily.
But... everything does fit in RAM at scale. I mean, Cloudflare basically runs a billion dollar business who's product is essentially "We store the internet in RAM in every city". The whole tech world is aflutter right now over a technology base that amounts to "We put the whole of human experience into GPU RAM so we can train our new overlords". It's RAM. Everything is RAM.
I'm not saying there is "no" home for excessively tuned genius-tier filesystem-over-persistent-storage code. I'm just saying that it's not a very big home, that the market has mostly passed the technology over, and that frankly patches like the linked article seem like a waste of effort to me vs. going to Amazon and buying more RAM.
source: https://blog.cloudflare.com/why-we-started-putting-unpopular...
> Our storage layer, which serves millions of cache hits per second globally, is powered by high IOPS NVMe SSDs.
Moreover, you again give examples from the end product. Finished sites, compacted JS files, compressed videos, compiled models...
There's much more than that. The model is in RAM, but you need to rake tons of data over that GPU. Sometimes terabytes of data. You have raw images to process, raw video to color-grade, unfiltered scientific data to sift through. These files are huge.
A well processed JPG from my camera is around 5MB, but RAW version I process is 25MB per frame, and it's a 24MP image, puny for today's standards. Your run of the mill 2K video takes a couple of GBs after final render at movie length. RAWs take 10s of terabytes, at minimum. Unfiltered scientific data again comes in terabytes to petabytes range depending on your project and instruments you work on, and multiple such groups pull their own big datasets to process real-time.
In my world, nothing fits in RAM except the runtime data, and that's your application plus some intermediate data structures. The rest is read from small to gigantic files and written in files of unknown sizes, by multiple groups, simultaneously. These systems experience the real meaning of "saturation", and they would really swear at us at some cases.
Sorry, but you can't solve this problem by buying more RAM, because these workloads can't be carried to clouds. They need to be local, transparent and fast. IOW, you need disk systems which feel like RAM. Again, look what Weka (https://www.weka.io/) does. It's one of the most visible companies which make systems behave like a huge RAM, but with multiple machines and tons of cutting edge SSDs, because what they process doesn't fit in RAM.
Lastly, oh, there's a law which I forget its name every time, which tells you if you cache 10 most used files, you can serve up to 90% of your requests from that cache, if your request pattern is static. In cases I cite, there's no "popular" file. Everybody wants their own popular files which makes access "truly random".
And that's just at the largest scale. I'm pretty sure banks still insist that the data is written to (multiple) disks (aka "stable storage") before completing a transaction.
Considering that multiple ZFS developers get paid to make ZFS work well on petabyte-sized disk arrays with SSD caching, and one of them often reports on progress in this area in his podcasts (2.5admins.com and bsdnow if you're interested) .. then yes?