Ensure json.extra.sidekiq conforms to logging schema

lib/gitlab/exception_log_formatter.rb

@@ -14,6 +14,21 @@ module Gitlab
 
       payload.delete('extra.server')
 
+      # The raven extra context is populated by Raven::SidekiqCleanupMiddleware.
+      #
+      # It contains the full sidekiq job which consists of mixed types and nested
+      # objects. That causes a bunch of issues when trying to ingest logs into
+      # Elasticsearch.
+      #
+      # We apply a stricter schema here that forces the args to be an array of
+      # strings. This same logic exists in Gitlab::SidekiqLogging::JSONFormatter.
+      payload['extra.sidekiq'].tap do |value|
+        if value.is_a?(Hash) && value.key?('args')
+          value = value.dup
+          payload['extra.sidekiq']['args'] = Gitlab::Utils::LogLimitedArray.log_limited_array(value['args'].try(:map, &:to_s))
+        end
+      end
+
       if exception.backtrace
         payload['exception.backtrace'] = Gitlab::BacktraceCleaner.clean_backtrace(exception.backtrace)
       end

spec/lib/gitlab/error_tracking_spec.rb

@@ -179,5 +179,16 @@ describe Gitlab::ErrorTracking do
         described_class.track_exception(exception, extra_info)
       end
     end
+
+    context 'with sidekiq args' do
+      let(:extra) { { sidekiq: { 'args' => [1, { 'id' => 2, 'name' => 'hello' }, 'some-value', 'another-value'] } } }
+
+      it 'ensures extra.sidekiq.args is a string' do
+        expect(Gitlab::ErrorTracking::Logger).to receive(:error).with(
+          hash_including({ 'extra.sidekiq' => { 'args' => ['1', '{"id"=>2, "name"=>"hello"}', 'some-value', 'another-value'] } }))
+
+        described_class.track_exception(exception, extra)
+      end
+    end
   end
 end