fix(eventstore): Use more correct start date for eventstore (#67070)

Clickhouse behaves oddly when the normal epoch is used for dates:


ClickHouse/ClickHouse#9335 (comment)

This was working before by coincedence: an obsolete piece of code (query
splitters) in Snuba was clamping the date to a non-epoch value. When the
query
splitters were removed, this started breaking.

Instead of simply asking for "all time" use a date that aligns with how
our
data is actually stored.

This is also not a new issue, see
#30912

src/sentry/eventstore/snuba/backend.py

@@ -426,7 +426,9 @@ def get_adjacent_event_ids(self, event, filter):
         prev_filter = deepcopy(filter)
         prev_filter.conditions = prev_filter.conditions or []
         prev_filter.conditions.extend(get_before_event_condition(event))
-        prev_filter.start = datetime.fromtimestamp(0)
+
+        # We only store 90 days of data, add a few extra days just in case
+        prev_filter.start = event.datetime - timedelta(days=100)
         # the previous event can have the same timestamp, add 1 second
         # to the end condition since it uses a less than condition
         prev_filter.end = event.datetime + timedelta(seconds=1)

tests/sentry/eventstore/snuba/test_backend.py

@@ -1,7 +1,5 @@
 from unittest import mock
 
-import pytest
-
 from sentry.eventstore.base import Filter
 from sentry.eventstore.models import Event
 from sentry.eventstore.snuba.backend import SnubaEventStorage
@@ -221,7 +219,6 @@ def test_get_event_beyond_retention(self):
             event = self.eventstore.get_event_by_id(self.project2.id, "d" * 32)
             assert event is None
 
-    @pytest.mark.xfail(reason="fails in Snuba CI")
     def test_get_adjacent_event_ids(self):
         event = self.eventstore.get_event_by_id(self.project2.id, "b" * 32)