Is it a good way to store files on a server?

Question

MySQL

Task: a user without registration can upload and download their files. In the future, he can register and get additional functionality.
File storage idea.
There is a storage class that can write, read and store an array of files by key (md5). Structure:

/user_files/ef/eff7d5dba32b4da32d9a67a519434d3f.zip
/user_files/d5/d58e3582afa99040e27b92b13c8f2280.zip

There is a controller class that controls the storage class (with access control). Key generation occurs, for example, according to the following scheme: md5(user_id+file_name+...). The database stores user_id, original file name and key.
Download (given via php):

/user-files/download/d58e3582afa99040e27b92b13c8f2280/ -> Мой документ.zip

1. All files on 1 server.
2. High loads are not expected.
Are there any downsides to this scheme?

Answer 1

MD5 is not unique - theoretically they can be repeated.

Answer 2

And "Download (given via php)" can be replaced with download via X-Accel-Redirect if nginx is installed

Answer 3

Normal system.

Answer 4

The scheme is not bad, but it's better to use sha1_file() to detect the same files. Those. first generate a checksum, then search the database for a duplicate, if there is one, check if the files match (this may take time, the background processing should be thought out). If the files are the same, then there is no point in keeping copies. In this case, the scheme in the database becomes a little more complicated, but disk space is significantly saved due to deduplication.
Metadata for each user must be stored individually and the link should be unique for each user.

Answer 5

Quite a normal system
I do almost the same myself. Only files "faceless". Title and meta in the database.
It is given through php, the folders themselves are naturally closed.

• /uploads/05/03/2014/2a7db79344e0bfe58975de0b505310c6
• /uploads/03.05.2014/37d7af3859d055fc8685334b03b052dd

Answer 6

Quite nginx.org/en/docs/http/ngx_http_proxy_module.html#...

Answer 7

Cons are standard.

• For a million files, one level of nesting is enough, but for ten, problems will already begin when there are several dozen files in each directory. I remember about "no more lyama", but "640K of memory is enough for everyone ...", oh.
• Known to be non-unique user filenames. Do not forget to add microtime to the hashed string.
• Make sure that the hashed string does not have a common prefix (say, the full path to a file on disk) - then the first characters of the hash will be unevenly distributed, and fill the directories unevenly.