Ok, that title is a mouthful, but so was the bug. It was one of the more painful debugging sessions I’ve had in a while; probably a solid 10 hours.
I’m using SimpleRecord to access Amazon’s SimpleDB from my Rails app. SimpleRecord sits on top of the Appoxy’s AWS gem which is a fork of RightScale’s AWS gem
I’m testing out importing all my web logs into SimpleDB and was getting this intermittent error: Aws::AwsError: SignatureDoesNotMatch
At first I narrowed it down to occasional UTF-8 errors and then read this discussion thread and got a little more background. I did a bunch of trial-n-error with encoding, double-encoding, not encoding, etc. to determine a pattern of what, when and where it was failing.
Here’s the general matrix for when it succeeded/failed:
Success: PutAttributes w/ no UTF-8 characters with an HTTP GET Success: PutAttributes w/ UTF-8 characters with an HTTP GET Success: PutAttributes w/ no UTF-8 characters with an HTTP POST Failure: PutAttributes w/ UTF-8 characters with an HTTP POST
So I even suspected Ruby’s HTTP library that it was some how messing with the body of the POST, so I turned on Apache’s DumpIO module:
DumpIOInput On DumpIOLogLevel notice
and had the AWS library do the GET’s and POST to my localhost so I could watch what was happening. Everything looked fine there.
By this time, my eyes were seeing double. So I started down the line of investigation: “Other languages must have run into this. Let me look at their code…”
Voila the first article I read was this PHP example. It is using the content type of:
that the AWS gem is using and the Amazon SimpleDB API specifies. That made so much sense that you could say it really tied the room together.
Solution / Patch
So the 1 line solution is in right_awsbase.rb line # 398
request['Content-Type'] = 'application/x-www-form-urlencoded'
request['Content-Type'] = 'application/x-www-form-urlencoded; charset=utf-8'