✨ Reduce downtime of make refresh by swapping the database

[Marigold]

Running make refresh induces downtime of about ~2 minutes. This is annoying when refreshing staging servers. The idea of this PR is to create a _temp database, load it and then swap with the original. Unfortunately, there's no RENAME DATABASE command in MySQL and we have to rename it table by table (and recreate views).

See slack

Other potential solutions:

• switch to PostgreSQL
• how about spinning up another server in a different data directory, loading into that, then shutting down the first mysql server and removing it's data directory and starting the second one?
• The more longterm option would be we give all tables uuids and timestamps where they are missing and then change the refresh code to upsert missing rows and delete obsolete ones

Note that owid user on staging cannot create a database. We have to change its privileges in ops to

mysql> GRANT ALL PRIVILEGES ON *.* TO 'owid'@'%';
mysql> FLUSH PRIVILEGES;

[owidbot]

Quick links (staging server):

Site	Admin	Wizard

Login: ssh owid@staging-site-refresh-swap

• Site-screenshots: https://github.com/owid/site-screenshots/compare/refresh-swap
• SVG tester: https://github.com/owid/owid-grapher-svgs/compare/refresh-swap

SVG tester:

Number of differences (default views): 1264 (f285de) ❌
Number of differences (all views): 531 (b8e7fe) ❌

Edited: 2024-06-26 09:51:44 UTC
Execution time: 1.12 seconds