Remove Sidekiq jobstr from logs and Sentry

When an exception happens in Sidekiq, Sentry gets a hash of the job
details, and then the same hash as a JSON-encoded string. That could be
useful if we had issues decoding the JSON, but that is very rare.
Sending this has a couple of problems:

1. It's wasteful.
2. It makes it hard to process any sensitive information out of the job,
   as we'd have to do so in both the hash and the string.

By removing the string, we only have to remove sensitive information
from the hash.

lib/gitlab/error_tracking.rb

@@ -26,6 +26,7 @@ module Gitlab
 
           # Sanitize fields based on those sanitized from Rails.
           config.sanitize_fields = Rails.application.config.filter_parameters.map(&:to_s)
+          config.processors << ::Gitlab::ErrorTracking::Processor::SidekiqProcessor
           # Sanitize authentication headers
           config.sanitize_http_headers = %w[Authorization Private-Token]
           config.tags = { program: Gitlab.process_name }

lib/gitlab/error_tracking/processor/sidekiq_processor.rb

+# frozen_string_literal: true
+
+module Gitlab
+  module ErrorTracking
+    module Processor
+      class SidekiqProcessor < ::Raven::Processor
+        def process(value, key = nil)
+          sidekiq = value.dig(:extra, :sidekiq)
+
+          return value unless sidekiq
+
+          sidekiq = sidekiq.dup
+          sidekiq.delete(:jobstr)
+
+          value[:extra][:sidekiq] = sidekiq
+
+          value
+        end
+      end
+    end
+  end
+end

lib/gitlab/sidekiq_logging/json_formatter.rb

@@ -18,6 +18,9 @@ module Gitlab
         when String
           output[:message] = data
         when Hash
+          # jobstr is redundant and can include information we wanted to
+          # exclude (like arguments)
+          data.delete(:jobstr)
           convert_to_iso8601!(data)
           convert_retry_to_integer!(data)
           stringify_args!(data)

spec/lib/gitlab/error_tracking/processor/sidekiq_processor_spec.rb

+# frozen_string_literal: true
+
+require 'fast_spec_helper'
+require 'rspec-parameterized'
+
+require 'raven'
+
+RSpec.describe Gitlab::ErrorTracking::Processor::SidekiqProcessor do
+  describe '#process' do
+    context 'when there is Sidekiq data' do
+      it 'removes a jobstr field if present' do
+        value = {
+          job: { 'args' => [1] },
+          jobstr: { 'args' => [1] }.to_json
+        }
+
+        expect(subject.process(extra_sidekiq(value)))
+          .to eq(extra_sidekiq(value.except(:jobstr)))
+      end
+
+      it 'does nothing with no jobstr' do
+        value = { job: { 'args' => [1] } }
+
+        expect(subject.process(extra_sidekiq(value)))
+          .to eq(extra_sidekiq(value))
+      end
+    end
+
+    context 'when there is no Sidekiq data' do
+      it 'does nothing' do
+        value = {
+          request: {
+            method: 'POST',
+            data: { 'key' => 'value' }
+          }
+        }
+
+        expect(subject.process(value)).to eq(value)
+      end
+    end
+
+    def extra_sidekiq(hash)
+      { extra: { sidekiq: hash } }
+    end
+  end
+end

spec/lib/gitlab/sidekiq_logging/json_formatter_spec.rb

@@ -42,6 +42,12 @@ describe Gitlab::SidekiqLogging::JSONFormatter do
       expect(subject).to eq(expected_output)
     end
 
+    it 'removes jobstr from the hash' do
+      hash_input[:jobstr] = 'job string'
+
+      expect(subject).not_to include('jobstr')
+    end
+
     context 'when the job args are bigger than the maximum allowed' do
       it 'keeps args from the front until they exceed the limit' do
         half_limit = Gitlab::Utils::LogLimitedArray::MAXIMUM_ARRAY_LENGTH / 2