[release/5.x] Cherry pick: Don't throw error if the ledger directory exists but is empty on node start (#6859) (#6885)

Co-authored-by: Andrea Piccione <[email protected]>

Author: achamayou

src/host/main.cpp

@@ -206,10 +206,12 @@ int main(int argc, char** argv)
   {
     if (config.command.type == StartType::Start)
     {
-      if (files::exists(config.ledger.directory))
+      if (
+        files::exists(config.ledger.directory) &&
+        !fs::is_empty(config.ledger.directory))
       {
         throw std::logic_error(fmt::format(
-          "On start, ledger directory should not exist ({})",
+          "On start, ledger directory should not exist or be empty ({})",
           config.ledger.directory));
       }
       // Count members with public encryption key as only these members will be

src/host/snapshots.h

@@ -158,6 +158,12 @@ namespace asynchost
         continue;
       }
 
+      if (fs::exists(f.path()) && fs::is_empty(f.path()))
+      {
+        LOG_INFO_FMT("Ignoring empty snapshot file {}", file_name);
+        continue;
+      }
+
       auto snapshot_idx = get_snapshot_idx_from_file_name(file_name);
       if (snapshot_idx > latest_committed_snapshot_idx)
       {

tests/e2e_operations.py

@@ -205,6 +205,40 @@ def test_large_snapshot(network, args):
     )
 
 
+def test_empty_snapshot(network, args):
+
+    LOG.info("Check that empty snapshot is ignored")
+
+    with tempfile.TemporaryDirectory() as snapshots_dir:
+        LOG.debug(f"Using {snapshots_dir} as snapshots directory")
+
+        snapshot_name = "snapshot_1000_1500.committed"
+
+        with open(
+            os.path.join(snapshots_dir, snapshot_name), "wb+"
+        ) as temp_empty_snapshot:
+
+            LOG.debug(f"Created empty snapshot {temp_empty_snapshot.name}")
+
+            # Check the file is indeed empty
+            assert (
+                os.stat(temp_empty_snapshot.name).st_size == 0
+            ), temp_empty_snapshot.name
+
+            # Create new node and join network
+            new_node = network.create_node("local://localhost")
+            network.join_node(new_node, args.package, args, snapshots_dir=snapshots_dir)
+            new_node.stop()
+
+            # Check that the empty snapshot is correctly skipped
+            if not new_node.check_log_for_error_message(
+                f"Ignoring empty snapshot file {snapshot_name}"
+            ):
+                raise AssertionError(
+                    f"Expected empty snapshot file {snapshot_name} to be skipped in node logs"
+                )
+
+
 def split_all_ledger_files_in_dir(input_dir, output_dir):
     # A ledger file can only be split at a seqno that contains a signature
     # (so that all files end on a signature that verifies their integrity).
@@ -292,6 +326,7 @@ def run_file_operations(args):
                 test_forced_ledger_chunk(network, args)
                 test_forced_snapshot(network, args)
                 test_large_snapshot(network, args)
+                test_empty_snapshot(network, args)
 
                 primary, _ = network.find_primary()
                 # Scoped transactions are not handled by historical range queries
@@ -589,6 +624,47 @@ def run_service_subject_name_check(args):
             assert cert.subject.rfc4514_string() == "CN=This test service", cert
 
 
+def run_empty_ledger_dir_check(args):
+    with infra.network.network(
+        args.nodes,
+        args.binary_dir,
+        args.debug_nodes,
+        args.perf_nodes,
+        pdb=args.pdb,
+    ) as network:
+        LOG.info("Check that empty ledger directory is handled correctly")
+        with tempfile.TemporaryDirectory() as tmp_dir:
+            LOG.debug(f"Using {tmp_dir} as ledger directory")
+
+            dir_name = os.path.basename(tmp_dir)
+
+            # Check tmp_dir is indeed empty
+            assert len(os.listdir(tmp_dir)) == 0, tmp_dir
+
+            # Start network, this should not fail
+            network.start_and_open(args, ledger_dir=tmp_dir)
+            primary, _ = network.find_primary()
+            network.stop_all_nodes()
+
+            # Now write a file in the directory
+            with open(os.path.join(tmp_dir, "ledger_1000_1500.committed"), "wb") as f:
+                f.write(b"bar")
+
+            # Start new network, this should fail
+            try:
+                network.start(args, ledger_dir=tmp_dir)
+            except Exception:
+                pass
+
+            # Check that the node has failed with the expected error message
+            if not primary.check_log_for_error_message(
+                f"On start, ledger directory should not exist or be empty ({dir_name})"
+            ):
+                raise AssertionError(
+                    f"Expected node error message with non-empty ledger directory {dir_name}"
+                )
+
+
 def run(args):
     run_max_uncommitted_tx_count(args)
     run_file_operations(args)
@@ -599,3 +675,4 @@ def run(args):
     run_preopen_readiness_check(args)
     run_sighup_check(args)
     run_service_subject_name_check(args)
+    run_empty_ledger_dir_check(args)

tests/infra/network.py

@@ -465,6 +465,7 @@ def _start_all_nodes(
                             workspace=args.workspace,
                             label=args.label,
                             common_dir=self.common_dir,
+                            ledger_dir=ledger_dir,
                             members_info=self.consortium.get_members_info(),
                             **forwarded_args_with_overrides,
                             **kwargs,