Pros of putting images in a Database.
-
Transactions. When you save the blob, you can commit it just like any other piece of DB data. That means you can commit the blob along with any of the associate meta-data and be assured that the two are in sync. If you run out of disk space? No commit. File didn’t upload completely? No commit. Silly application error? No commit. If keeping the images and their associated meta data consistent with each other is important to your application, then the transactions that a DB can provide can be a boon.
-
One system to manage. Need to back up the meta data and blobs? Back up the database. Need to replicate them? Replicate the database. Need to recover from a partial system failure? Reload the DB and roll the logs forward. All of the advantages that DBs bring to data in general (volume mapping, storage control, backups, replication, recovery, etc.) apply to your blobs. More consistency, easier management.
-
Security. Databases have very fine grained security features that can be leveraged. Schemas, user roles, even things like “read only views” to give secure access to a subset of data. All of those features work with tables holding blobs as well.
-
Centralized management. Related to #2, but basically the DBAs (as if they don’t have enough power) get to manage one thing: the database. Modern databases (especially the larger ones) work very well with large installations across several machines. Single source of management simplifies procedures, simplifies knowledge transfer.
-
Most modern databases handle blobs just fine. With first class support of blobs in your data tier, you can easily stream blobs from the DB to the client. While there are operations that you can do that will “suck in” the entire blob all at once, if you don’t need that facility, then don’t use it. Study the SQL interface for your DB and leverage its features. No reason to treat them like “big strings” that are treated monolithically and turn your blobs in to big, memory gobbling, cache smashing bombs.
-
Just like you can set up dedicated file servers for images, you can set up dedicated blob servers in your database. Give them dedicated disk volumes, dedicated schemas, dedicated caches, etc. All of your data in the DB isn’t the same, or behaves the same, no reason to configure it all the same. Good databases have the fine level of control.
The primary nit regarding serving up an blob from a DB is ensuring that your HTTP layer actually leverages all of the HTTP protocol to perform the service.
Many naive implementations simply grab the blob, and dump them wholesale down the socket. But HTTP has several important features well suited to streaming images, etc. Notably caching headers, ETags, and chunked transfer to allow clients to request “pieces” of the blob.
Ensure that your HTTP service is properly honoring all of those requests, and your DB can be a very good Web citizen. By caching the files in a filesystem for serving by the HTTP server, you gain some of those advantages “for free” (since a good server will do that anyway for “static” resources), but make sure if you do that, that you honor things like modification dates etc. for images.
For example, someone requests spaceshuttle.jpg, an image created on Jan 1, 2009. That ends up cached on the file system on the request date, say, Feb 1, 2009. Later, the image is purged from the cache (FIFO policy, or whatever), and someone, later, on Mar 1, 2009 requests it again. Well, now it has a Mar 1, 2009 “create date”, even though the entire time its create date was really Jan 1. So, you can see, especially if your cache turns around a lot, clients that may be using If-Modified headers may be getting more data than they actually need, since the server THINKS the resource has changed, when in fact it has not.
If you keep the cache creation date in sync with the actual creation date, this can be less of a problem.
But the point is that it’s something to think through about the entire problem in order be a “good web citizen”, and save you and your clients potentially some bandwidth etc.
I’ve just gone through all this for a Java project serving videos from a DB, and it all works a treat.