Upgrade to MySQL 8

[rossjrw]

Resolves #62

• [x] Upgrade test runner
• [x] Make tests pass
• [x] Run the prod database on MySQL 8 to see if it works
• [x] Update codebase documentation to refer to MySQL 8

Then if everything looks good:

• [x] Create a Graviton EC2 instance
• [x] Put MySQL 8 on it
• [x] See if it works
• [x] Migrate the database to it
• [x] Point the other microservices to the new instance

If everything works as expected:

• [x] Terminate the MySQL 5.7 instance
• [x] Delete its data
• [x] Update user documentation to refer to MySQL 8, if mentioned at all Not mentioned

[rossjrw]

Which instance type should I use?

Noting the following criteria for selecting an instance type:

• t4g.nano is a direct cheaper alternative to the current type, t3.nano
• For any burstable instance type (t*), anticipate constant usage of unlimited burst mode - add $0.04 to the hourly cost when comparing with other types
• The database will not make use of multiple threads (we only run 1 query at a time, and in fact MySQL does not support concurrent queries beyond extremely limited use cases anyway) so do not prioritise instance types offering >1 vCPU
• I do not know how much memory is needed (#61 will help)

(Note: after #61, it looks like I am using about 700 MB of memory)

Current on-demand pricing for a selection of instance types:

Type	Price	vCPUs	Memory	Notes
c6g.medium	$0.0404	1	2 GiB	Compute optimised
m6g.medium	$0.0444	1	4 GiB	General purpose non-burstable
t4g.nano	$0.0447	2	0.5 GiB	Direct upgrade
t3.nano	$0.0459	2	0.5 GiB	Current
r6g.medium	$0.0592	1	8 GiB	Most expensive single-vCPU instance type
t4g.medium	$0.0776	2	4 GiB	Burstable medium equivalent for comparison

Based on the above I conclude:

• Burstable types have a terrible price-performance ratio compared to similarly-specced alternative instance types (provided they are running in unlimited burst mode) and I am scamming myself so long as I continue to use one
• If there were nano versions of non-burstable instance types, they would be the best - but there aren't
• It is possible that paying more to get surplus performance will result in a net saving, if the reduction in processing time is enough
  • But it doesn't look like that is even an option - you pay more for more vCPUs (i.e. threads), not better vCPUs

My limiting factor is CPU so c6g.medium is probably the best bet