fix(balancer) ensure existing healthchecker is stopped when processing target event

[onematchfox]

Summary

Ensures that existing healthchecks are stopped when processing CRUD events on targets.

Full changelog

• Ensure stop_healthchecker is called for existing balancer prior to create_balancer. Follows pattern from do_upstream_event

Notes:

• Haven't added tests for this. Figured I would try copy the existing test from do_upstream_event but it doesn't look like this is tested at present? (Was looking at https://github.com/Kong/kong/blob/2.4.1/spec/01-unit/09-balancer_spec.lua so perhaps Iäm in the wrong place?)
• Opened PR against master. But it seems like the branching strategy in this repo has changed somewhat since I last submitted a PR. Feel free to let me know if you want this rebased to a (release?) branch.

Issues resolved

Fix #7406
Fix #7407

[locao]

Hi @onematchfox, thanks for your submission! This change is correct but needs a sister PR (Kong/lua-resty-healthcheck#74) to fix the issue. We'll merge this after releasing a new version of lua-resty-healthcheck and bumping it in Kong.

[onematchfox]

Hi @onematchfox, thanks for your submission! This change is correct but needs a sister PR (Kong/lua-resty-healthcheck#74) to fix the issue. We'll merge this after releasing a new version of lua-resty-healthcheck and bumping it in Kong.

👍

Was aware of the additional change that was required and had already opened a PR for this (Kong/lua-resty-healthcheck#71). I hadn't removed worker_events.register_weak(self.ev_callback, self.EVENT_SOURCE) not sure I follow why you'd want to remove the registration of the callback as well but respect your expertise on the matter 😄

[locao]

Sorry, I missed that PR, thanks for pointing that out!

About the worker_events.register_weak(self.ev_callback, self.EVENT_SOURCE), I removed it because it was being called twice, by new() and by start() functions.