Tag Archives: s3

I’ve recently started using Mailgun, which I like quite a bit, but I stumbled on an issue dealing with attachments, because the files I needed to attach are stored in S3. Using RestClient to send the emails, the expectation is that attachments are files. As it turns out, using the aws-sdk gem, S3 objects don’t quite behave like files, so it doesn’t work to simply toss the S3Object instance into the call.

The standard setup for sending an email is like the following:

 data =
 data[:from] = "My Self <>"
 data[:subject] = "Subject"
 data[:to] = "#{recipients.join(', ')}"
 data[:text] = text_version_of_your_email
 data[:html] = html_version_of_your_email
 data[:attachment] ="files", "attachment.txt")) "https://api:#{API_KEY}"\

As this shows, the attachment is expected to be a File that can be read in. So the challenges was to make an S3 object readable. One option, of course, is to do this in two steps: read in the S3 object and use Tempfile to write a file which can then be used as the attachment. This seemed pretty unfortunate. For one thing, I’m running stuff on Heroku, and try to avoid using the file system even for temp files. But primarily, it’s really wasteful to have to write, and then read, a transitory file. The better option, of course, was to see if there was a way to trick the client into reading from S3.

Thanks to some very nice help from Mailgun support (thanks Sasha!), the idea of writing a wrapper seemed feasible, and in fact it wasn’t too bad aside from a couple of tricky issues. A side-effect advantage was that it solved another problem: the naming of the attachment. By default, the name of the attachment is the name of the file, which is pretty ugly if you use a temp file. Not user-friendly.

I’ve put the wrapper file in a Github Gist at It’s pretty short, and there were only a couple of gotchas, which I describe below. The key for this wrapper is to provide the methods that RestClient needs: #read, #path, #original_filename, and #content_type. It’s pretty obvious what #read and #path are for. The attachment naming problem is solved by #original_filename: whatever it returns will be the name of the attachment in the email. It should be clear what #content_type does, but see below for why it’s important.

Using the wrapper is described in the header comment, but it’s mainly a change to give RestClient the wrapper instead of a File object:

data[:attachment] =, file_key)

The first gotcha was that RestClient calls #read repeatedly for 8124 bytes, and doesn’t pass a block. This forced me to write a crude sort of buffering — the wrapper reads the whole S3 object in, then hands out chunks of it when asked. This isn’t a problem for me because the files I’m dealing with aren’t very large, but it’s something I warn about in the comments. If you have large files, this may be a problem for you, so beware.

The second gotcha that threw me off for a little while is that the value returned by #content_type is important. I haven’t researched exactly why this is, but I found that if I tried to send a Word document but #content_type returns ‘text/plain’, the attachment comes through corrupted. It was easy enough for me to check the filename suffix and set the content type accordingly, but I can imagine cases where this might not work, so this is something else to beware of.

Anyway, this solved the issue for me, and hopefully it’ll be useful for others. There are ways to make this a bit more elegant, but it’s a short piece of code that works. Enjoy.