jvanasco at 2xlp.com
Thu Apr 12 00:31:20 MSD 2007
On Apr 11, 2007, at 3:45 PM, David Phillips wrote:
> On 4/11/07, Jonathan Vanasco <nginx at 2xlp.com> wrote:
>> my setup will have me querying postgres for every url to find the
>> correct url on amazon's s3 which i proxy for.
> Have you considered using deterministic S3 object keys (for example,
> using a hash)? This allows you to avoid the database entirely. You
> could add hash support to the rewrite module.
> nginx's proxy module does not block. If your S3 objects are not
> publicly readable then you could modify the proxy module to support S3
> request signing.
Thats a pretty good idea...
The issue that I would have with that setup though is that I'd only
be replacing the db call with an s3 call.
my options were:
30+ s3 data stores of a single file, all publicly readable
1 s3 data store of a single file , use a proxy to transparantly
serve that file to clients
because of their pricing and my usage patterns, i went with the
second option -- which requires that the client can never know the
address of the object on s3.
More information about the nginx