Given a file being returned as part of a http request. What is the correct method of creating an ETag for that file?
http://www.w3.org/Protocols/rfc2616/rfc2616-sec14.html#sec14.19
I have seen it done several ways
- Using the last write time of the file converted to ticks. Which is what NancyFX does https://github.com/NancyFx/Nancy/blob/master/src/Nancy/Responses/GenericFileResponse.cs
- Using a hash of the file http://hartzer.wordpress.com/2009/10/04/client-side-caching/
Answer
The answer is that it depends.
There are two kinds of Etags, weak and strong ones. The weak ones let you do certain conditional operations, but most of them will require strong etags.
The only restriction on a strong etag is that if the representation changes for whatever reason, so does the etag. If it's a file you can generate a hash, but that forces you to keep the entity in-memory while you generate that. Alternatively, you could simply stream the content and add the Etag as a trailing http header, but nearly no server currently supports that (very useful and unloved) piece of functionality.
Ticks have a low resolution, and it is possible that two contiguous writes on the same file will have the same number of ticks, even though the content is different. At that stage, the etag is now invalid if it was strong and you're in muddy water. Same issue as with Last-Modified really. The way most HTTP servers deal with that is to give a file an etag based on multiple of its properties, aka timestamp, size and probably file object ID (inode in apache, probably adding the full path to the object store entry on NT, and IIS adds a counter to that value too so two config changes on a server would generate different etags, in case something changes).
If you use some sort of database, the id + version ought to be a strong etag (provided, again, that the content you ahve is not an aggregate of multiple things that each may change independently without the root changing versions too).
So how to calculate it really depends on your scenario, and persisting a hash of the file on write (before starting serving) will probably help you the most, especially as it's a very useful feature to have for many other reasons.