[humble] Fix for a false negative integration test with bag split in …

…recorder (backport #1743) (#1751)

* Fix for a false negative integration test with bag split in recorder (#1743)

- In
record_end_to_end_with_splitting_bagsize_split_is_at_least_specified_size
publisher was starting after terminating recorder.

Signed-off-by: Michael Orlov <[email protected]>
(cherry picked from commit da1acb2)

# Conflicts:
#	rosbag2_tests/test/rosbag2_tests/test_rosbag2_record_end_to_end.cpp

* Address merge conflicts

Signed-off-by: Michael Orlov <[email protected]>

---------

Signed-off-by: Michael Orlov <[email protected]>
Co-authored-by: Michael Orlov <[email protected]>

rosbag2_tests/test/rosbag2_tests/test_rosbag2_record_end_to_end.cpp

@@ -326,13 +326,13 @@ TEST_F(RecordFixture, record_end_to_end_with_splitting_bagsize_split_is_at_least
   ASSERT_TRUE(pub_manager.wait_for_matched(topic_name)) <<
     "Expected find rosbag subscription";
 
+  pub_manager.run_publishers();
+
   wait_for_db();
 
   stop_execution(process_handle);
   cleanup_process_handle.cancel();
 
-  pub_manager.run_publishers();
-
   rosbag2_storage::MetadataIo metadata_io;
 
 #ifdef _WIN32
@@ -357,10 +357,7 @@ TEST_F(RecordFixture, record_end_to_end_with_splitting_bagsize_split_is_at_least
   const auto metadata = metadata_io.read_metadata(root_bag_path_.string());
   const auto actual_splits = static_cast<int>(metadata.files.size());
 
-  // TODO(zmichaels11): Support reliable sync-to-disk for more accurate splits.
-  // The only guarantee with splits right now is that they will not occur until
-  // a bagfile is at least the specified max_bagfile_size.
-  EXPECT_GT(actual_splits, 0);
+  EXPECT_EQ(actual_splits, expected_splits);
 
   // Don't include the last bagfile since it won't be full
   for (int i = 0; i < actual_splits - 1; ++i) {