Inside the get method of a handler, I would like to do the following:

1. query some data store, and get the data
2. create a bunch of .csv files using the data
3. create a .zip file containing all those .csv files
4. serve this .zip file to the client

In addition, I have no need to store .csv or .zip files, so would prefer
that by the end of the request there is no file left on the server.

I wonder how should I optimally achieve steps 2, 3, 4 here in Tornado?
Thanks!

While it may not be the most efficient implementation, I think a good start
is to create a StringIO for each file and pass it as the file object to
csv.writer():

http://docs.python.org/library/csv.html#csv.writer

Once you've written all the CSVs to StringIOs, create a ZipFile with
another StringIO as its output file:

output = String()
zip = zipfile.ZipFile(output, 'w')

and write each CSV to the zip archive with writestr(my_stringio.getvalue(),
...):

http://docs.python.org/library/zipfile.html#zipfile.ZipFile.writestr

And finalize the ZipFile with close(). Finally:

self.set_header('Content-Type, 'application/zip')
self.write(output.getvalue())
self.finish()

Obviously, this requires having several copies of your data in memory at
once. If this isn't a problem then don't sweat it. If it is, you could
experiment with streaming results into the zipfile, which would save some
overhead, and more interestingly you could try streaming results *out* by
passing self.connection.stream into the ZipFile constructor as its output
file. If that works then you wouldn't have to keep much data in memory at
all.

Thanks! That's more or less what I've ended up with.

The only difference is that I don't create a StringIO object for each .csv
file, but use a single StringIO object for the zipfile object (passing it
into the constructor), and then I just do:

zf.writestr('file1.csv', content1)
zf.writestr('file2.csv', content2)
zf.writestr('file3.csv', content3)
...

The self.connection.stream approach sounds interesting, but what would you
pass into self.write() method when you flush the data to the browser client?

The idea is you don't call self.write, but instead let ZipFile do the writing. On second thought this isn't going to work because you need to set a content-length header before you can start writing, so I'm not sure the best way to do what I'm thinking of.