Intermittent 500s

[msabramo]

Sometimes the same call succeeds and sometimes it returns a 500 error that looks like this:

________________________________________________________________ TestDocverter.test_basic_conversion_from_file _________________________________________________________________
test_docverter.py:51: in test_basic_conversion_from_file
    received = docverter.convert(file_name, 'rst')
docverter.py:28: in convert
    format, extra_args, encoding=encoding)
docverter.py:60: in _convert
    return processor(source, to, format, extra_args)
docverter.py:96: in _process_file
    % (resp, resp.content))
E   RuntimeError: Call to docverter failed - 
    resp = <Response [500]>; 
    resp.content = '{"error":"No such file or directory - tmpkElYDL"}'

[msabramo]

Another one:

_______________________________________________________________ TestDocverter.test_basic_conversion_from_string ________________________________________________________________
test_docverter.py:74: in test_basic_conversion_from_string
    received = docverter.convert('#some title', 'rst', format='md')
docverter.py:28: in convert
    format, extra_args, encoding=encoding)
docverter.py:60: in _convert
    return processor(source, to, format, extra_args)
docverter.py:96: in _process_file
    % (resp, resp.content))
E   RuntimeError: Call to docverter failed - 
    resp = <Response [500]>; 
    resp.content = b'{"error":"No such file or directory - manifest.orig.yml"}'

[msabramo]

Another; the error is slightly different in this one.

<Response [500]>;
resp.content = '{"error":"command error: pandoc: tmpqBtaRr: openFile: does not exist 
                  (No such file or directory)\\n"}'

[peterkeen]

Are you using the pubic instance or your own instance?

[msabramo]

Public. c.docverter.com

[msabramo]

Maybe I'll see if I can reproduce on a self-hosted instance.

[msabramo]

Hmmm. Maybe it's a bug in my Python code (I'm writing a Python client for Docverter), because I can't reproduce with HTTPie.

❯ while true; do http --form -f POST http://c.docverter.com/convert from=markdown to=rst 'input_files[]@/tmp/tmpqBtaRr' --print=h --pretty=none | grep '^HTTP'; sleep 1; done
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK
HTTP/1.1 200 OK

Though it seems to me like those temp file errors are happening on the server, because the filenames don't match the temp file filenames that my program is using.

[peterkeen]

Hm. The only time the server makes a new filename is when it's encrypting a PDF. The rest of the time it's working with the given filenames in a tempdir.

[msabramo]

Maybe those temp files are being created by the web server and then being passed into the Ruby code?

[peterkeen]

For both input_files[] and other_files[] we copy the given tempfile handle to the original given filename:

https://github.com/Docverter/docverter/blob/master/lib/docverter-server/app.rb#L20-L27

We have to do this because your input files are going to refer each other and to things in other_files[] by filename.

[msabramo]

Hmmm. IIRC from my PHP days, sometimes the web server or the platform will keep uploaded files in memory (especially if the files are small and mine are) and not always write them to a temp file on the disk.

[peterkeen]

It doesn't really matter. The :tempfile key in the upload is something that responds to read and the :filename key is the filename specified by the MIME header for that part of the request.

Also, in almost three years of development this is the first time someone has mentioned a bug like this, which makes me think the problem is not necessarily in Docverter. It may be a weird combination of edge cases, though.

[msabramo]

Yeah, this is weird. I have not seen it yet with HTTPie and it is sporadic when running my Python test suite. I have gotten it to happen a few times when running the tests locally, though it seems anecdotally to happen more when running tests on Travis CI. But it's not exclusive to Travis CI.

[msabramo]

Here's a dump of the HTTP request for a recent failure:

POST http://c.docverter.com/convert
Content-Length: 345
Content-Type: multipart/form-data; boundary=03377687359d44d78a5e88a807d085aa
--03377687359d44d78a5e88a807d085aa
Content-Disposition: form-data; name="from"
markdown
--03377687359d44d78a5e88a807d085aa
Content-Disposition: form-data; name="to"
rst
--03377687359d44d78a5e88a807d085aa
Content-Disposition: form-data; name="input_files[]"; filename="tmpCg8ugv"
#some title
--03377687359d44d78a5e88a807d085aa--

The response was a 500 with:

No such file or directory - /tmp/d20150202-3-fx2rd3/output.dab677eb282d3aa6ec5e.rst

Interesting - for this one - the error is with the output file (note the .rst extension); not the input file.

[peterkeen]

Thank you for the dump. I'm seeing occasional Heroku errors in the logs, but it doesn't look like they're related to this problem. I'll dig into it when I have time.