Resolve live trace read race condition using a retry

This commit makes it possible to retry a trace read operation when an
empty chunk data slice is detected, possibly due to a race condition
between a read operation and trace archive worker migrating data.

lib/gitlab/ci/trace.rb

@@ -80,11 +80,9 @@ module Gitlab
         job.trace_chunks.any? || current_path.present? || old_trace.present?
       end
 
-      def read(should_retry: true, &block)
+      def read(&block)
         read_stream(&block)
-      rescue Errno::ENOENT
-        raise unless should_retry
-
+      rescue Errno::ENOENT, ChunkedIO::FailedToGetChunkError
         job.reset
         read_stream(&block)
       end

spec/lib/gitlab/ci/trace_spec.rb

@@ -33,6 +33,16 @@ RSpec.describe Gitlab::Ci::Trace, :clean_gitlab_redis_shared_state, factory_defa
 
       expect(artifact2.job.trace.raw).to eq(test_data)
     end
+
+    it 'reloads the trace in case of a chunk error' do
+      chunk_error = described_class::ChunkedIO::FailedToGetChunkError
+
+      allow_any_instance_of(described_class::Stream)
+        .to receive(:raw).and_raise(chunk_error)
+
+      expect(build).to receive(:reset).and_return(build)
+      expect { trace.raw }.to raise_error(chunk_error)
+    end
   end
 
   context 'when live trace feature is disabled' do