From 914d9998cdf551bdddf4ad850dde48e66b72998e Mon Sep 17 00:00:00 2001 From: Timo Heister Date: Wed, 1 Jun 2022 12:05:00 -0400 Subject: [PATCH] add comment --- source/base/data_out_base.cc | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-) diff --git a/source/base/data_out_base.cc b/source/base/data_out_base.cc index a7bbb7d98c..b7ea069b18 100644 --- a/source/base/data_out_base.cc +++ b/source/base/data_out_base.cc @@ -7786,9 +7786,11 @@ DataOutInterface::write_vtu_in_parallel( } } - // Make sure we sync to disk. This ensures that the change in file - // size caused by the write_at() for the footer is visible for all - // ranks. Otherwise, the footer is sometimes lost. + // Make sure we sync to disk. As written in the standard, + // MPI_File_close() actually already implies a sync but there seems + // to be a bug on at least one configuration (running with multiple + // nodes using OpenMPI 4.1) that requires it. Without this call, the + // footer is sometimes missing. ierr = MPI_File_sync(fh); AssertThrowMPI(ierr); -- 2.39.5