Make Redis instrumentation spec more robust

When run as the first spec in a run, this would fail. It would then pass
on retry. That was because on the first Redis call, we SELECT the
correct database for the test environment.

To work around this, we perform a Redis command - also doing the SELECT
if needed - and then reset the request store data so that the expected
counts can remain exact.

spec/lib/gitlab/instrumentation/redis_spec.rb

@@ -28,6 +28,13 @@ RSpec.describe Gitlab::Instrumentation::Redis do
 
   describe '.payload', :request_store do
     before do
+      # If this is the first spec in a spec run that uses Redis, there
+      # will be an extra SELECT command to choose the right database. We
+      # don't want to make the spec less precise, so we force that to
+      # happen (if needed) first, then clear the counts.
+      Gitlab::Redis::Cache.with { |redis| redis.info }
+      RequestStore.clear!
+
       Gitlab::Redis::Cache.with { |redis| redis.set('cache-test', 321) }
       Gitlab::Redis::SharedState.with { |redis| redis.set('shared-state-test', 123) }
     end