Only report MysteryRedisDurationError for shared state Redis

This error reporting (behind a feature flag) was added to investigate
high durations we saw on the shared state Redis on GitLab.com. However,
it did not check the Redis instance. This is a particular problem with
the Sidekiq (queues) Redis instance, which receives many BRPOP commands
with a timeout higher than the reporting threshold for this error.

As a result, when we enabled this flag we got flooded with 'expected'
errors for the queues Redis, which we didn't care about at all. So now
we also check which Redis instance we're using before reporting the
error.

lib/gitlab/instrumentation/redis_interceptor.rb

@@ -41,7 +41,10 @@ module Gitlab
           instrumentation_class.add_call_details(duration, args)
         end
 
-        if duration > DURATION_ERROR_THRESHOLD && Feature.enabled?(:report_on_long_redis_durations, default_enabled: :yaml)
+        if duration > DURATION_ERROR_THRESHOLD &&
+           instrumentation_class == ::Gitlab::Instrumentation::Redis::SharedState &&
+           Feature.enabled?(:report_on_long_redis_durations, default_enabled: :yaml)
+
           Gitlab::ErrorTracking.track_exception(MysteryRedisDurationError.new(caller),
                                                 command: command_from_args(args),
                                                 duration: duration,

spec/lib/gitlab/instrumentation/redis_interceptor_spec.rb

@@ -130,15 +130,25 @@ RSpec.describe Gitlab::Instrumentation::RedisInterceptor, :clean_gitlab_redis_sh
     end
 
     context 'when report_on_long_redis_durations is enabled' do
-      it 'tracks an exception and continues' do
-        expect(Gitlab::ErrorTracking)
-          .to receive(:track_exception)
-                .with(an_instance_of(described_class::MysteryRedisDurationError),
-                      command: 'mget',
-                      duration: be > threshold,
-                      timestamp: a_string_matching(/^\d{4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2}\.\d{5}/))
+      context 'for an instance other than SharedState' do
+        it 'does nothing' do
+          expect(Gitlab::ErrorTracking).not_to receive(:track_exception)
 
-        Gitlab::Redis::SharedState.with { |r| r.mget('foo', 'foo') { sleep threshold + 0.1 } }
+          Gitlab::Redis::Queues.with { |r| r.mget('foo', 'foo') { sleep threshold + 0.1 } }
+        end
+      end
+
+      context 'for the SharedState instance' do
+        it 'tracks an exception and continues' do
+          expect(Gitlab::ErrorTracking)
+            .to receive(:track_exception)
+                  .with(an_instance_of(described_class::MysteryRedisDurationError),
+                        command: 'mget',
+                        duration: be > threshold,
+                        timestamp: a_string_matching(/^\d{4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2}\.\d{5}/))
+
+          Gitlab::Redis::SharedState.with { |r| r.mget('foo', 'foo') { sleep threshold + 0.1 } }
+        end
       end
     end
   end