dataloss subcommand based on repository generations

Changes dataloss subcommand to report data loss based on repository
generations. This allows us to report more accurately, including
how many writes a repository is behind at maximum.

The report is not 100% accurate as generations map to mutator calls and
some mutator calls do not indicate unreplicated writes. These are mutators
are mainly for repository housekeeping and do make any changes visible to the
client. The accuracy can be improved by tagging these housekeeping mutators
later as such and ignoring them when it comes to incrementing the generation
of a repository.

Generations guarantee the repository was at least on the stored generation.
As such, the tool might produce false positives in case a replication was
successful but the database record failed to be updated. In case of a failover,
some of the repositories reported as outdated may contain the latest changes.

Since generations are only recorded for repositories that have received
new mutator RPCs, this won't report data loss from repositories that have
failed replication prior to the generations being introduced.

1 files changed, 10 insertions, 8 deletions

diff --git a/ruby/proto/gitaly/praefect_pb.rb b/ruby/proto/gitaly/praefect_pb.rb
index 1ad2ce827..cd8fb9abb 100644
--- a/ruby/proto/gitaly/praefect_pb.rb
+++ b/ruby/proto/gitaly/praefect_pb.rb
@@ -22,15 +22,16 @@ Google::Protobuf::DescriptorPool.generated_pool.build do
     optional :virtual_storage, :string, 1
   end
   add_message "gitaly.DatalossCheckResponse" do
-    optional :virtual_storage, :string, 1
-    optional :previous_writable_primary, :string, 2
-    optional :current_primary, :string, 3
-    optional :is_read_only, :bool, 4
-    repeated :outdated_nodes, :message, 5, "gitaly.DatalossCheckResponse.Nodes"
+    optional :primary, :string, 1
+    repeated :repositories, :message, 2, "gitaly.DatalossCheckResponse.Repository"
   end
-  add_message "gitaly.DatalossCheckResponse.Nodes" do
+  add_message "gitaly.DatalossCheckResponse.Repository" do
     optional :relative_path, :string, 1
-    repeated :nodes, :string, 2
+    repeated :storages, :message, 2, "gitaly.DatalossCheckResponse.Repository.Storage"
+  end
+  add_message "gitaly.DatalossCheckResponse.Repository.Storage" do
+    optional :name, :string, 1
+    optional :behind_by, :int64, 2
   end
   add_message "gitaly.RepositoryReplicasRequest" do
     optional :repository, :message, 1, "gitaly.Repository"
@@ -65,7 +66,8 @@ module Gitaly
   EnableWritesResponse = Google::Protobuf::DescriptorPool.generated_pool.lookup("gitaly.EnableWritesResponse").msgclass
   DatalossCheckRequest = Google::Protobuf::DescriptorPool.generated_pool.lookup("gitaly.DatalossCheckRequest").msgclass
   DatalossCheckResponse = Google::Protobuf::DescriptorPool.generated_pool.lookup("gitaly.DatalossCheckResponse").msgclass
-  DatalossCheckResponse::Nodes = Google::Protobuf::DescriptorPool.generated_pool.lookup("gitaly.DatalossCheckResponse.Nodes").msgclass
+  DatalossCheckResponse::Repository = Google::Protobuf::DescriptorPool.generated_pool.lookup("gitaly.DatalossCheckResponse.Repository").msgclass
+  DatalossCheckResponse::Repository::Storage = Google::Protobuf::DescriptorPool.generated_pool.lookup("gitaly.DatalossCheckResponse.Repository.Storage").msgclass
   RepositoryReplicasRequest = Google::Protobuf::DescriptorPool.generated_pool.lookup("gitaly.RepositoryReplicasRequest").msgclass
   RepositoryReplicasResponse = Google::Protobuf::DescriptorPool.generated_pool.lookup("gitaly.RepositoryReplicasResponse").msgclass
   RepositoryReplicasResponse::RepositoryDetails = Google::Protobuf::DescriptorPool.generated_pool.lookup("gitaly.RepositoryReplicasResponse.RepositoryDetails").msgclass