ZFS, which is known for having crazy high limits, will fit somewhat over 280 trillion entries in a single directory. Curiously, though, you can have over 18 million trillion file systems in a single storage pool.
I thought this changed from ext3 to ext4? I know I've run into this issue before and moving to ext4 was the easiest way to resolve it. Ext4 is much quicker at doing operations on very large directories too.
During my young naïve programming days (which coming to think of it haven't really ended) I thought to myself "Why bother using some third party database when I could just store user information in text files, the OS handles that sort of thing really well, right?". I looked into it as much as I could be bothered and found no huge red flags.
After a few years of this server merrily creating a new file for each user I found that my disk was full, as this poster did. After a lot of hunting around, I found out that I had hit my inode limit, a concept I was completely unaware of.
I know for sure I won't be making THAT mistake again.
"After a lot of hunting around, I found out that I had hit my inode limit, a concept I was completely unaware of."
What was the number you hit?
I have a flat file system for something I'm prototyping. The files are in a database with a random SHA256 filename associated with them. I was thinking about doing what Git does with creating folders for parts of the string to alleviate the number of files in one single directory. I'm sure that makes file retrieval a prettier process.
Do all the files for have to be in the same directory for the iNode limit to be reached or just there in general?
NOTE: I ask because the prototype isn't ever meant to go into production, but my customer keeps getting confused and I'm concerned this very important system may inherit (or even use) my design.
I've apparently used 1241165 out of 19005440, or about 6.53% on my development machine. It's not inconceivable to have 19 million files, though, if you had one per user. That's not great.
How is that number calculated? It's less than 2^25 but greater than 2^24, so I imagine it's even awkward to store. I guess that has something to do with the file table size for HDD - you can't have more nodes than you have addressable space for those nodes.
I knew Git did it for speed, but I was wondering whether it was also about the number of files that can exist in a single directory.
It's a property of the filesystem that is set when it was created. In the case of the mkfs.ext{2,3,4}, you can set the number of inodes with the -N flag, but by default, it is the size of the filesystem divided by inode_ratio (probably 4096).
Each user has some small arbitrary information associated with them, with potentially tens of millions of users (and no doubt more in the future). There's not really a structure due to the nature of the project, each user needs personalised information and structure associated to them.
I don't think that amount of information is really suitable for a database? It may be larger in some cases too, so there's not guarantees I can even make about size.
Document stores shine in that case. Depending on the exact case something like postgres's HSTORE, couchdb, couchbase, mongodb or similar would work. They're all capable of storing arbitrary json docs under a given key and efficiently retrieve it. Partial indexing is possible as well.
This seems to be one of those bits of old UNIX lore that's being lost - it used to be fairly well-known in sysadmin circles that if you were going to be storing a lot of tiny files, you should create your filesystem with a higher than usual number of inodes (the classic case was an NNTP spool mount).
> if you were going to be storing a lot of tiny files, you should create your filesystem with a higher than usual number of inodes (the classic case was an NNTP spool mount).
I wonder how many people understand what the term "news server" of the partition wizard in the Debian installer means.
I literally just solved an issue this weekend that was inode exhaustion: there were years of snapshots and traces laying around and the DBAs were confused as to why the database couldn't create a directory to start and was exiting with "No space left on device" when there was 5% storage free.
That lead to the second discussion of why you don't fix it by piping filenames one by one to rm, but either use find's inbuilt unlink/rm if there is one or xargs to rm if not.
How is that number calculated? It's less than 2^25 but greater than 2^24
It depends on the filesytem, but it is typically a function of filesystem size (and in ext, the number of block groups and inodes per block group size). To take an example:
sudo tune2fs -l <device> | egrep -i 'group'
Blocks per group: 32768
Fragments per group: 32768
Inodes per group: 8192
So, we have got 1 inode per 4 blocks. This particular filesystem has 79472640 blocks (of 4KiB) and 19873792 inodes. This corresponds to the 4:1 ratio: 79472640 / 19873792 =~ 4.
There is probably a upper limit, but practically you don't want more inodes than blocks. Since a non-empty file at least takes up one block, assuming that your files are all at least 1 byte, you couldn't use more inodes than blocks anyway.
(The exception are empty files, because on most filesystems an empty file does not use any data blocks, but possibly an inode to store file size, permissions, etc.)
The whole FS has a limited number of inodes, additionally some FS have limited number of files or directories inside a directory. If you don't control the FS chosen by your customer, I'd design conservatively (find the dumbest set of limits, stay under them).
Completely reasonable thing to do, for a little while. Ycombinator itself did this initially. The key is knowing you'll need to do something better soon.
Completely reasonable thing to do, for a little while. Ycombinator itself did this initially. The key is knowing you'll need to do something better soon.
inodes are by the way a great topic to talk about in a job interview. People working with unix based systems should know about them, or at least encountered them once in their career.
Another good one that even less people know about is the sticky bit.
I did the same thing to support about 10 million customer websites, but had the opposite experience - it worked pretty well, we used UFS and a 4 level directory structure (i.e. customer "apple" went into '/a/p/p/l/apple' and just wrote a simple apache module to map URL's to this directory structure.(using a hash of the filename would have given more even distribution, but this naive algorithm worked pretty effectively)
It was much easier than coding a database back-end, with no need to run a database server. And, when we needed to scale beyond one machine, it was trivial to use a central NFS server (read-only NFS worked really well for this) and multiple front-end web servers.
Though we did know how many files we'd be hosting, so we sized the number of bytes per inode appropriately.
It's worth checking that out, but I suspect the folder solution described on a sibling comment is a better solution. If a file system gets too full, it's trivial to move the files in a folder to another file system an mount it where the folder originally was. Also, spreading the files across multiple NFS mounts is also trivial.
Standard practice to store a large number of files but in such a way that they don't end up in a single directory, is to create a crypto hash of your object names and use the first n characters as the directory name. Properties of hash functions will ensure that the files are distributed evenly between all the directories.
git is an example of a program that uses this technique (among others) to store blobs. In case of git, the blob names are already SHA-1 hashes.
You don't actually need a cryptographic hash here [1]. I'd rather use a fast hash like Murmur instead.
[1] Exception: I don't know how prone Murmur is to hash collisions, so if collision-based attacks are part of your threat model, a randomly salted cryptographic hash might be appropriate.
Although the exception would probably be rare if it used a large enough number, do you think the "copying utility" (whatever it may be) should handle those extreme edge cases? Or is it just "disk full" that's used as a fall-back?
Probably lucky. File names are md5 of the content, i am not sure if it helped on collision on 32 bit md4.
The linked post mentions dir_index default and tune2fs -l lists the dir_index under Filesystem features only.
Not knowing this default behavior, i was considering enabling dir_index explicitly to aid chocking rsync.
EDIT:
I ran the reproduction code with 200K instead of 100K just to be sure. There was no problem creating file as well as copying all the files to a different folder. The disk was formatted about 3 months ago and mounted with default options "ext4 noatime,barrier=0 0 0" . On original as well as copied folder `ls | wc -l` still gives 200000.
"If you're working with huge files or a huge number of files, use XFS" went the slogan.
I was hit by this recently - with our 500ish (!!) entries in /node_modules and their dependencies of dependencies of dependencies, a standard 8GB VM formatted with ext4 just simply didn't have enough inodes for us to have two copies of our app (deployment swapover). Had to increase the size to 20GB, just to get the extra inodes. XFS would have handled it much better (from testing on a parallel machine). It was inode exhaustion, though the 'disk full' error message was the same - confusing at first, as df showed 40% free space...
You could have also tuned ext4 when you created it with mke2fs. You can specify with -N the max # of inodes desired. Granted, it's not something very many people consider when creating their filesystems; for one thing it's hard to predict how many you will need.
It would be great if it had switch that simply says "double the default number of inodes". Per stories in comments, seems that on large file systems it's easier to run out of inodes, the default must be too low.
A good filesystem to use as pseudo-db is ZFS. It's designed to have practically limitless number of files, and it's snapshotting/cloning come in really handy for things like backups etc. Also you get full checksumming.
I came to the same conclusion about sharding files from the get go. I had a similar experience where my host wouldn't accept to synchronize a directory with more than 5000 files in it. (Software producing the files is my own note-taking application where each page is stored as an individual file to speed up backup of only the pages that changes).
Wouldn't 5000 files of note-taking easily fit into RAM? Why bother dealing with thousands of small files? Also, rsync does quite a good job with intransferring multi-MB (or even larger) files efficiently while copying only what's actually changed.
I feel more secure by dealing with many individual files. For example in case of a conflict during sync, I can visually diff between the two pages very easily. This happens once in a while. Also for example if more than 5% of files have changed, the sync is stopped and asks for manual confirmation to prevent catastrophies.
Brings back memories of the old DOS days, when there was quite a low limit on the number of files that you could store in the root directory. I can't remember the exact numbers, but it lead to my early habits of creating a new sub directory for every little thing so nowadays (over 3 decades later) nearly everything I need is at least 5 levels deep! ;)
NB: My current working folders are now in Google Drive, but is essentially the same folder structure as I started on our Novell server about 25 years ago...
That sounds about right. It was a ridiculously low number, and even back in the days of 10MB hard drives, it was easy to bust that number with just batch and text files...
.. aided by CHKDSK, which would find orphaned allocation units and create files for them in the root directory. Could quite easily fill up your root directory on its own.
I've had some bad experience with btrfs. When I had to hard-reset a hung computer, some directories became unremovable and unrenameable. All else failed, and I attempted to --repair, it did so much damage that I had to reformat the disk. This was probably about 3-4 years ago. Have you had such issues on your preferred filesystems in the last couple of years?
The reason I stay with ext4 is that it has proven amazingly robust in the event of hard reboots.
In my experience XFS is rock solid. We use it almost everywhere at work. I've never had an issue with hard reboots with it. Only recently started playing with btrfs, so can't speak to it's reliability from personal experience.
I hard-reboot about once a week, caused by my carelessness and running out of 16GB of RAM. I always recommend against btrfs for system partitions, and if you don't make frequent backups, against user partitions too. Read through "man btrfsck", that should caution you sufficiently. I'll try out XFS on an external HDD I'll buy soon.
You really shouldn't need to be hard rebooting just because you ran out of ram...
did you try a sysrq f (invokes the out of memory killer) first?
If that doesn't work you should at least be able to force a sync and remount read-only before taking it down.
If you have access to a keyboard and the kernel is still responding to sysrq you should never just hard power off without at least trying to regain control of the system.
Some reading if you're not familiar with sysrq:
https://en.wikipedia.org/wiki/Magic_SysRq_key
Depending on your distro you might have to enable it, I believe ubuntu (for some dumb reason) ships with emergency sync as the only command available.
Enabling it should amount to adding: kernel.sysrq=1 to /etc/sysctl.conf on older machines or editing
/etc/sysctl.d/10-magic-sysrq.conf on newer ones.
also if you are running out of ram that often you might want to consider a swap partition. Or if you have one already try setting a more aggressive swapiness value.
I've had similar error in production. We had linux server with C++ program doing many things, and in the process it created temporary files for a barcode printer. The files were named uniquely, the server had lots of free space, we left them there to allow easier debugging if sth wrong happened (it happened a few times each year - we created the labels to print by glueing escape codes and sometimes there were labels that didn't worked for one reason or another), and only cleaned the old files when there was not much space left.
At some point there was a bug when sometimes users got "wrong permissions or no space left" errors (sorry don't remember the exact error and IIRC our app just tried to save a file and failed with unspecific error if it couldn't write the new temporary file, so we haven't had the exact system error).
What was weird is - there was still lots of space on that disk, permissions were OK, and I've tried creating new files there, both from our app and by hand and it worked.
It turned out after a few days of frantic debugging, that when there's too many files with similar names in one directory - creating a new file can fail because of hash conflict, depending on the filename only. So,
ls A1234.txt S1234.txt # no results
touch S12345.txt # fails
touch S1234.txt # works
:) It was one of the biggest WTF moments I had while programming.
This does not seem like the exactly correct diagnosis.
ext4 doesn't care if there's collisions in the dir_index hash - they're entirely expected. The dir_index hashes are just used to look up in a tree, the lowest level of which points you to the directory block that contains all the directory entries for names with hashes in a given range.
However, those hashes do get used for the directory position used by telldir() / seekdir() - and those are only 32 bits on 32 bit architectures. So I suspect that something involving those calls is generating the ENOSPC.
76 comments
[ 2.9 ms ] story [ 152 ms ] threadExplorer will not be too happy displaying such a folder though!
After a few years of this server merrily creating a new file for each user I found that my disk was full, as this poster did. After a lot of hunting around, I found out that I had hit my inode limit, a concept I was completely unaware of.
I know for sure I won't be making THAT mistake again.
What was the number you hit?
I have a flat file system for something I'm prototyping. The files are in a database with a random SHA256 filename associated with them. I was thinking about doing what Git does with creating folders for parts of the string to alleviate the number of files in one single directory. I'm sure that makes file retrieval a prettier process.
Do all the files for have to be in the same directory for the iNode limit to be reached or just there in general?
NOTE: I ask because the prototype isn't ever meant to go into production, but my customer keeps getting confused and I'm concerned this very important system may inherit (or even use) my design.
The thing that git does is for speed.
How is that number calculated? It's less than 2^25 but greater than 2^24, so I imagine it's even awkward to store. I guess that has something to do with the file table size for HDD - you can't have more nodes than you have addressable space for those nodes.
I knew Git did it for speed, but I was wondering whether it was also about the number of files that can exist in a single directory.
EDIT: Corrected percentage.
That's crazy. Thanks for that information.
That's crazy.
Why? Do you expect to store many files less than 16KiB?
I don't think that amount of information is really suitable for a database? It may be larger in some cases too, so there's not guarantees I can even make about size.
Hopefully not now, this is why we all appreciate people sharing these sorts of information :)
I wonder how many people understand what the term "news server" of the partition wizard in the Debian installer means.
That lead to the second discussion of why you don't fix it by piping filenames one by one to rm, but either use find's inbuilt unlink/rm if there is one or xargs to rm if not.
It depends on the filesytem, but it is typically a function of filesystem size (and in ext, the number of block groups and inodes per block group size). To take an example:
So, we have got 1 inode per 4 blocks. This particular filesystem has 79472640 blocks (of 4KiB) and 19873792 inodes. This corresponds to the 4:1 ratio: 79472640 / 19873792 =~ 4.(The exception are empty files, because on most filesystems an empty file does not use any data blocks, but possibly an inode to store file size, permissions, etc.)
https://developer.apple.com/legacy/library/documentation/Dar...
Maybe Linux and the BSDs could adopt that as well.
Another good one that even less people know about is the sticky bit.
People who have the vague notion of 'an inode stores metadata' will usually answer this incorrectly.
It was much easier than coding a database back-end, with no need to run a database server. And, when we needed to scale beyond one machine, it was trivial to use a central NFS server (read-only NFS worked really well for this) and multiple front-end web servers.
Though we did know how many files we'd be hosting, so we sized the number of bytes per inode appropriately.
Btrfs?
git is an example of a program that uses this technique (among others) to store blobs. In case of git, the blob names are already SHA-1 hashes.
[1] Exception: I don't know how prone Murmur is to hash collisions, so if collision-based attacks are part of your threat model, a randomly salted cryptographic hash might be appropriate.
"The lesson here is that even in 2016, filesystems still are finicky with large numbers of files in a single directory."
The linked post mentions dir_index default and tune2fs -l lists the dir_index under Filesystem features only.
Not knowing this default behavior, i was considering enabling dir_index explicitly to aid chocking rsync.
EDIT: I ran the reproduction code with 200K instead of 100K just to be sure. There was no problem creating file as well as copying all the files to a different folder. The disk was formatted about 3 months ago and mounted with default options "ext4 noatime,barrier=0 0 0" . On original as well as copied folder `ls | wc -l` still gives 200000.
I was hit by this recently - with our 500ish (!!) entries in /node_modules and their dependencies of dependencies of dependencies, a standard 8GB VM formatted with ext4 just simply didn't have enough inodes for us to have two copies of our app (deployment swapover). Had to increase the size to 20GB, just to get the extra inodes. XFS would have handled it much better (from testing on a parallel machine). It was inode exhaustion, though the 'disk full' error message was the same - confusing at first, as df showed 40% free space...
NB: My current working folders are now in Google Drive, but is essentially the same folder structure as I started on our Novell server about 25 years ago...
512?
https://en.wikipedia.org/wiki/File_Allocation_Table#Final_FA... (see the last paragraph.)
The reason I stay with ext4 is that it has proven amazingly robust in the event of hard reboots.
Some reading if you're not familiar with sysrq: https://en.wikipedia.org/wiki/Magic_SysRq_key Depending on your distro you might have to enable it, I believe ubuntu (for some dumb reason) ships with emergency sync as the only command available.
Enabling it should amount to adding: kernel.sysrq=1 to /etc/sysctl.conf on older machines or editing /etc/sysctl.d/10-magic-sysrq.conf on newer ones.
also if you are running out of ram that often you might want to consider a swap partition. Or if you have one already try setting a more aggressive swapiness value.
It's good RHEL/CentOS 7 defaults to XFS now which doesn't suffer (as much) from this problem.
https://gagor.pl/2013/11/inodes-on-xfs/
At some point there was a bug when sometimes users got "wrong permissions or no space left" errors (sorry don't remember the exact error and IIRC our app just tried to save a file and failed with unspecific error if it couldn't write the new temporary file, so we haven't had the exact system error).
What was weird is - there was still lots of space on that disk, permissions were OK, and I've tried creating new files there, both from our app and by hand and it worked.
It turned out after a few days of frantic debugging, that when there's too many files with similar names in one directory - creating a new file can fail because of hash conflict, depending on the filename only. So,
:) It was one of the biggest WTF moments I had while programming.ext4 doesn't care if there's collisions in the dir_index hash - they're entirely expected. The dir_index hashes are just used to look up in a tree, the lowest level of which points you to the directory block that contains all the directory entries for names with hashes in a given range.
However, those hashes do get used for the directory position used by telldir() / seekdir() - and those are only 32 bits on 32 bit architectures. So I suspect that something involving those calls is generating the ENOSPC.
Statements like this make me trust a developer significantly less. It's our job to care about "technical reasons."