There have been 2 submissions (out of 33 so far) missing some of the required questions. One has only the first question answered. One has most of the questions answered but values stop before the end of the survey. Both are missing the end metadata value. Both have 0 for AttachmentsExpected when all the others have 6. Status for all submissions in the CSV export is blank.
I'm curious about what could be happening.
Internet connection when uploading is very poor. Not sure about the ODK Collect version but I think all the data collectors installed it on their phones via the Play Store in the last 2 weeks. Server is ODK Central at the following version.
versions:
836112b5bf39de826e1bf141d21db7f7b8d54ecf
4732f7112a286165241aaf7f971f2c2e38d6bb8a client (v1.0.0)
e9ffd2c0c3aa1a9475852e1397b8259e2b03165a server (v1.0.3)
Would it be possible for you to talk to the data collectors who made those submissions? Some questions that would be very helpful to get answers to:
What Android versions do they use? (older Android versions will get older versions of Collect)
Did they experience any crash or any other unexpected behavior while filling out the form?
Did they have to try sending multiple times?
Did they actually attach media for those submissions?
The ideal would be if you could get access to the submission on their device, for example if you have a supervisor coming by with Briefcase. I know that may be quite a stretch.
I've seen a couple of reports of this kind before but have never had enough information to get anywhere with troubleshooting.
Another thing that would be useful to know is how big the images that those data collectors capture are. Knowing the device types might allow us to estimate that. Would you be able to set max-pixels to have some control over image size?
It's feeling like a client/network issue, especially since the prior similar report I'm thinking of was on Aggregate but it'd be good to more confidently rule out server issues and it would be helpful to know exactly what is in your database. Depending on what we learn from your field team, I'll see about giving you some database queries to run.
The QR code configuration set Image size to Very small (640px) for Collect. Is it different when set in Collect versus in the XLSForm? I'll let you know if we can figure out which data collector it was and what other details they might be able to provide.
That's great. I should have thought you might be using that option! If data collectors have old devices and had to fall back to older Collect versions, image scaling might not be supported, but their cameras would probably also not capture very large images. So that doesn't seem like a promising line of exploration anymore.
Thank you. I hope we can get to the bottom of it quickly.
I just thought of another check to do on your end that would be easier than directly looking at the database. Could you please try pulling submissions with Briefcase and then doing an export? Does that have the exact same contents as your Central export?
Server is an AWS t2.small with 2GiB memory.
Using Briefcase to pull the data from Central and then export from Briefcase results in the same data issues.
If there's something direct access to the server would help y'all investigate, I can provide it.
It might be interesting to see what the raw XML instance was for the offending submission; eg null XML elements, or media attachment values that reference attachment id's that never made it up in the POST data, etc. Is it just media (eg photo) required questions that are missing?
I think that would be useful. Let me get in touch when I have a chance to make a plan for what to do and I know one of us has a block of time available.
Yes, that's what I'd look at in the Central DB. I expect from the Briefcase export that the XML is well-formed but has missing nodes. From Dan's "one has only the first question answered" it looks like it doesn't have anything to do with question type.
Does Central persist the (raw)uploaded instance XML? If that is well-formed XML, but contains XML elements associated with required question bindings that are null/empty, that might suggest something is seriously askew on the Collect frontend (Collect shouldn't be able to perform an OpenRosa submission POST operation with an incomplete xml_submission_file payload...). But if these missing questions all happen to be media types - and the instance XML is complete - that might point to a problem POSTing the associated base64 attachments.
Thanks, @danbjoseph, for a good forensics session today. It is looking like a client issue but we don't yet have a hypothesis as to how it may have happened. We were able to identify that the two submissions were made by the same device and know the device make and model so hopefully Dan's team will be able to have a conversation with the data collector and see what that person's Collect installation shows in Finalized Forms and Sent Forms.
