chore(deletions): Drop sentry_incidentseen and sentry_incidentsubscription leftover tables

[yuvmen]

These two models were already preivously deleted and a previous migration dropped them, but due to an issue with SafeDeleteModel they were not actually dropped. This migration drops them directly.

[github-actions]

This PR has a migration; here is the generated SQL for src/sentry/migrations/1004_drop_legacy_incidentseen_incidentsubscription.py

for 1004_drop_legacy_incidentseen_incidentsubscription in sentry

--
-- Raw SQL operation
--

            DROP TABLE IF EXISTS sentry_incidentseen;
            
--
-- Raw SQL operation
--

            DROP TABLE IF EXISTS sentry_incidentsubscription;

[vgrozdanic]

Do we know why this PR #81623 didn't drop the tables?

[vgrozdanic]

Should we also CASCADE deletes?

Not sure how was this used before, but previous PR that for to me unknown reason did't work generated following SQL:

--
-- Delete model IncidentSeen
--
DROP TABLE "sentry_incidentseen" CASCADE;
--
-- Delete model IncidentSubscription
--
DROP TABLE "sentry_incidentsubscription" CASCADE;

[yuvmen]

hmm yea it seemed wrong to cascade but now I realize the original had it, should probably be okay to cascade here as well - ill add it

[vgrozdanic]

If there are no cascading drops - this can be in-deployment migration, since the drop is quite fast when there are no active transactions on the table

[markstory]

Do we know why this PR #81623 didn't drop the tables?

This could be related to database routing not being able to route the table to a database. When the model class is removed, the silo assignment of the table is also lost. Adding these table names to the historical_silo_assignments could help.

sentry/src/sentry/db/router.py

Lines 68 to 80 in db8dc3d

	historical_silo_assignments = {
	"authidentity_duplicate": SiloMode.CONTROL,
	"authprovider_duplicate": SiloMode.CONTROL,
	"sentry_actor": SiloMode.REGION,
	"sentry_alertruleactivations": SiloMode.REGION,
	"sentry_monitorlocation": SiloMode.REGION,
	"sentry_notificationsetting": SiloMode.CONTROL,
	"sentry_pagerdutyservice": SiloMode.REGION,
	"sentry_projectavatar": SiloMode.REGION,
	"sentry_scheduledjob": SiloMode.CONTROL,
	"sentry_teamavatar": SiloMode.REGION,
	"workflow_engine_workflowaction": SiloMode.REGION,
	}

I suspect that you'll need to update the historical_silo_assignments to get this migration to run correctly as well.

[yuvmen]

@markstory - yea @wedamija and I discussed it, he didnt seem to believe it was necessary anymore and that SafeRunSQL should work without, though we agree thats probably why it failed originally.
theres actually a problem there where if it fails on that it just does a no-op silently because of this if:

sentry/src/sentry/new_migrations/monkey/models.py

Lines 38 to 39 in f6c2bc7

	if self.allow_migrate_model(schema_editor.connection.alias, model):
	schema_editor.delete_model(model, is_safe=True)

@wedamija is looking into preventing this in the future I believe

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@             Coverage Diff             @@
##           master   #103559      +/-   ##
===========================================
+ Coverage   75.42%    80.59%   +5.17%     
===========================================
  Files        9260      9259       -1     
  Lines      395305    395356      +51     
  Branches    25213     25179      -34     
===========================================
+ Hits       298160    318642   +20482     
+ Misses      96716     76284   -20432     
- Partials      429       430       +1     

[github-actions]

This PR has a migration; here is the generated SQL for src/sentry/migrations/1006_drop_legacy_incidentseen_incidentsubscription.py

for 1006_drop_legacy_incidentseen_incidentsubscription in sentry

--
-- Raw SQL operation
--

            DROP TABLE IF EXISTS sentry_incidentseen CASCADE;
            
--
-- Raw SQL operation
--

            DROP TABLE IF EXISTS sentry_incidentsubscription CASCADE;