...
When files are uploaded on bulk upload page, the files are uploaded to some storage and the job jobs are recorded with the status “PENDING“. Then the bulk upload service starts to pick up the first file, and set its status to “IN PROGRESS“, leaving all other following files “PENDING“. Only when the current file is marked with “COMPLETED“, the following “PENDING“ one ones will get processed.
We had a problem with Tis, the system was not doing Bulk upload which
...
Trigger
A file of person update template with 1180 records was uploaded.
On prod, the id (Epoch time) of the file is: 1674667043653 and the record id in
genericupload.ApplicationTypeis 28781.
...
Detection
User’s query on Teams
bulk upload file list on Admins-UI
docker logs on Prod Green
...
Resolution
Get all the person id from the spreadsheet, and query all the emails by person ids from metabase, then compare the emails from DB and emails in spreadsheet and find the discrepancies.
Inform the user the discrepancies and ask for a manual update.
Set the status of the stalled file from “IN_PROGRESS“ to “COMPLETED“.
Restart docker generic upload on Prod Green and find the service is picking up the following pending files on Admins-UI or docker logs.
...
17:17 user uploaded a file of person update template with 1180 records to update doctor emails.
09:35 users reported that the bulk upload job is still showing in process and all the following files were stuck.
10:00 the team discussed the user queries on standup.
11:10 most of the records in the stalled file were found to be alrady already updated on Prod.
12:34 a PR merged to retry the stalled file.
schema_versiongot the new installed version, andjobStartTimeinApplicationTypewas updated for the job, but there were no logs found on TCS.14:30ish a manual update on status of the job in progress to “PENDING“ was done, but it skipped over that spreadsheet.
15:00ish tcs cloudwatch logs were re-checked by Pepe & Yafang together and they agreed the spreadsheet had already been process.
16:00ish as expected, only one email was found not to be updated for that spreadsheet.
16:16 user informed with the status of the job
16:20 a manual update on status of the job from “IN_PROGRESS“ to “COMPLETED“ and then a docker restart was done. Generic upload service was resumed.
...
Root Cause(s)
A file was processing and hit an issue that stopped it completing. The other files were pending because it queues them up.
The logs show that the first file was started but not finished but the TCS logs indicated the first & last people in the file were saved, which took a number of minutes. Not initially, but later we later confirmed that all expected records had been updated. The problem occurred between the data being received by TCS and Bulk Upload generating the results.
There was likely a timeout because of the time taken to process all the records in the file and the limits of connection between Bulk Upload & TCS.
For person update, Generic Upload service assembles all the DTOs together, and sends them to TCS at all once.
Then in TCS, it verifies all the gmc/gdc, person details, contact details, right to work, roles, trainerApprovals, etc on existing person entity together with the data from spreadsheet. It took very long time if there’re many records are uploaded in one file.
After the validation, TCS saves the data into several DB tables one record by one record, which is also time-consuming.
Only when all the processing is done in TCS, it sends a response back to Generic Upload service. So there possibly have been timeouts between TCS and Generic Upload.
When initially developed, there were very few bits of stateful validation with lookups but there have been additions which don’t take minutes to process for “normal”/smaller numbers of updates.
There is a hard limit to the maximum timeout for requests without “keep-alive” packets in some newer infrastructure we are now using.
...
Action Items
Action Items | Comments | Owner |
|---|---|---|
Add a new endpoint in TCS for single person validation&update, and modify Person Update bulk upload to send person one by one. | When we re-design bulk upload service, this should be taken into account. | |
...