[Fleet] Fleet server migrations should not block the start lifecycle

[kobelb]

The fleet server migrations should not block the start lifecycle:

kibana/x-pack/plugins/fleet/server/plugin.ts

Line 302 in 3b728b7

await runFleetServerMigration();

The long-term goal is to remove the ability to have asynchronous setup/start/stop lifecycle methods entirely. Currently, plugins have 30 seconds to perform their start lifecycle, and if it takes longer than 30 seconds, Kibana will crash and fail to start. Given the nature of the migrations, and that Elasticsearch might be under heavy load during startup, we're in serious risk of exceeding the 30-second window.

I'd recommend starting the fleet server migrations in the start lifecycle, and passing the unresolved Promise to the services that should wait for the fleet server migrations to complete before proceeding. This allows the rest of Kibana to start-up and not be blocked while fleet specific migrations are performed.

[elasticmachine]

Pinging @elastic/ingest-management (Team:Ingest Management)

[elasticmachine]

Pinging @elastic/fleet (Feature:Fleet)

[jen-huang]

This was fixed by #95481:

kibana/x-pack/plugins/fleet/server/plugin.ts

Line 302 in 78937e3

const fleetServerSetup = startFleetServerSetup();