Detecting and fixing production data on training instances

After onboarding CHWs, sometimes data ends up on the wrong CHT instance. There are some passive and active actions you can take to help deal with this situation.

Monitoring

Monitoring is a good way to see if CHWs are sending forms to the wrong CHT server. By catching such a problem early, it may be easy to fix manually which avoids more laborious fixes on the command line for admins.

Manual device checks

Supervisors can actively monitor CHWs as they register their first household. If each CHW is given a deadline to do this, the Supervisor can follow up promptly with CHWs who have not met the deadline.

Centralized database checks

From an administrative perspective, passive monitoring can be done by querying the Postgres instance that couch2pg is populating. If 2020-09-14 was the date to stop use the training instance, this SQL query would show CHWs using the wrong instance after 2020-09-14:

with forms as (
  select
    doc #>> '{fields,inputs,user,contact_id}' as chw_id,
    count(*) as n_forms
  from
    couchdb
  where to_timestamp((doc ->> 'reported_date' )::bigint / 1000) > '2020-09-14 00:00:00'
  group by 1
), chw_names as (
  select 
  doc ->> '_id' as chw_id, 
  doc ->> 'name' as chw_name
  from couchdb 
  where doc ->> 'role' = 'chw'
  and doc ->> 'name' not like 'ASC%'
)
select 
  n.chw_name,
  f.n_forms
from 
forms f 
left join 
chw_names n 
using(chw_id)
order by n_forms desc;

A limitation of this technique is that CHWs who have not synchronized their device will not show up. In this case, a supervisor proactively checking per above is a better failsafe.

CHT based supervisor tasks

Tasks for supervisors can also be used for the CHWs that report to them. This can be used to trigger a task when a CHW hasn’t submitted a task within a given period on production.

Remediation

In the case that production data has been entered into a training instance, care must be used to ensure data is not lost or overwritten. These steps assume you’re on a Linux command line and have jq, curl and cat installed. The task is to carefully download all docs created after a certain date on a training, clean up extra docs not needed and then upload them to a production instance:

1. Set the aliases and environment variables:

   alias ts='date -u +"%Y-%m-%dT%H:%M:%SZ"'
   alias curlj="curl -H 'content-type:application/json'"
   alias curljz="curl --compressed -H 'content-type:application/json'"
   export COUCH_URL=https://admin:pass@dev.example.org
   

2. Ensure affected CHWs have synced their devices to the development instance one last time. If they have been offline for some time and gathered production data, signing out or deleting the app before synchronizing may result in lost data.

3. Find the date when CHWs should have started using production. Convert this to epoch with milliseconds. For example if the time was “Tue, 01 Dec 2020 09:36:50 GMT” then the epoch would be “1606815410000”

4. Using Fauxton, create a view to show all documents created after this date:

   function (doc) {
       if(doc.reported_date > 1606815410000) {
           emit(doc._id, 1);
       }
   }
   

   

5. Download the UUIDs of the documents in the view, and convert it to raw format (uuids-2.txt):

   curl "$COUCH_URL/medic/_design/temp/_view/temp1" > uuids-1.txt
   cat uuids-1.txt | jq -r '.rows[].id' > uuids-2.txt
   

6. Download all the matching docs to docs.json:

   cat uuids-2.txt | jq --raw-input --slurp '{keys: split("\n") }' | curljz -d@- "$COUCH_URL/medic/_all_docs?include_docs=true" | jq '{docs: [ .rows[] | select(.doc).doc ]}' > docs.json
   

7. Change the fields by deleting “_rev” field and also “_attachments” if uploading in a different instance. This can be carefully done by manually editing the JSON locally before uploading.

8. Change the COUCH_URL to be your production instance by setting another export:

   export COUCH_URL=http://admin:pass@app.example.org
   

9. Upload the docs, for example, the following command adds a field (updated=true) to the docs and uploads them:

   cat docs.json | jq '{"docs":[ .docs[] | .updated = true ]}' | curljz -d@- $COUCH_URL/medic/_bulk_docs > results.json