Page MenuHomePhabricator

Test silverpop export with utf8mb4 tables
Closed, ResolvedPublic

Description

As we are converting the database tables to charset=utf8mb4, we want to test and rule out any issues in processes that use the DB. We need to test the export of the silverpop data to see if there are any differences in the export after charset conversion. None are anticipated, but it should be confirmed.

Current testing idea:

  • Copy the silverpop db/tables to a second instance.
  • Convert the db/tables in the second instance to utf8mb4
  • Change view silverpop export uses to point at converted data
  • Run silverpop export
  • Compare output CSV from converted data to that of unconverted
  • Change view back to original

Event Timeline

The silverpop DB has been backed up and then restored as dev_silverpop on frdb1001. dev_silverpop and all of it's tables were then converted to utf8mb4. It is ready for testing the export script.

Looking at the table space, everything has been converted to utf8mb4 except the two views: dev_silverpop.silverpop_export_view and dev_silverpop.silverpop_export_view_full

That's kinda what we are testing here and they may need to be recreated to pick up the altered charsets as the conversion script didn't address views specifically.

Change 670000 had a related patch set uploaded (by Dwisehaupt; owner: Dwisehaupt):
[wikimedia/fundraising/tools@master] Shift from utf8 to utf8mb4 encoding for the DB tables

https://gerrit.wikimedia.org/r/670000

Change 670000 merged by Eileen:
[wikimedia/fundraising/tools@master] Shift from utf8 to utf8mb4 encoding for the DB tables

https://gerrit.wikimedia.org/r/670000

Change 670632 had a related patch set uploaded (by Eileen; owner: Eileen):
[wikimedia/fundraising/tools@master] Fix a few more utf8s in silverpop

https://gerrit.wikimedia.org/r/670632

I just did some testing and found just a few rows that were not switched to utf8mb4 - presumably the inconsistent caching caused them to slip through - I put up a patch for that (& hence moving this into review)

Regarding the view - that is constantly recreated with slightly different date parameters

Change 670632 merged by jenkins-bot:
[wikimedia/fundraising/tools@master] Fix a few more utf8s in silverpop

https://gerrit.wikimedia.org/r/670632

Change 671217 had a related patch set uploaded (by Cstone; owner: Cstone):
[wikimedia/fundraising/tools@master] Switch general to unicode.

https://gerrit.wikimedia.org/r/671217

Change 671217 merged by Cstone:
[wikimedia/fundraising/tools@master] Switch general to unicode.

https://gerrit.wikimedia.org/r/671217

The work has been completed and silverpop runs are occurring. We are tracking an accuracy task in T277366 but there is currently nothing left to do on the ops side for the utf8mb4 conversion at the database or code for base functionality.