Apparent memory leaks

TL;DR

I have memory leaks coming from several H5D functions used when I extend and write to a data set, I assume the problem is on my end but my code is pretty boiler plate.

Questions:

  • Is there some returned/duplicated structures I am not cleaning up or is this a know problem?

  • Is this related to my dataset extension/chunking?

    • Is this an expected addition to the data set structure ( represented by dset_id, which I assume is held in memory) as we expand / add chunks?

Full Description

I have a program: it uses HDF5 C API (local cmake build) on windows x64, developed using VS 2019 and MSVC.

I have a method to extend and write to a data set.

However there appears to be a memory leak stemming from my use of HDF5, specifically H5D.
Specifically functions:

  • H5Dwrite
  • H5Dset_extent
  • H5Dget_space

Functions from other parts of the library, eg H5Sselect_hyperslab do not appear to leak in the same way.

Memory profiling:

Using two snapshots in VS 2019:

The code:

	// extend dataset
	cur_dims.front() += bufLen;
	ret = H5Dset_extent(dset_id, &cur_dims.front());
	assert(ret >= 0);

	// Get dataspace of extended dataset
	space_id = H5Dget_space(dset_id);
	assert(space_id >= 0);

	// Select start of hyperslab in the dataset dataspace
	std::vector<hsize_t> slab_start(rank, 0);
	slab_start.front() = cur_dims.front() - bufLen;

	std::vector<hsize_t> slab_count{ bufLen };
	slab_count.insert(std::end(slab_count), std::begin(extra_dims), std::end(extra_dims));

	ret = H5Sselect_hyperslab(space_id, H5S_SELECT_SET, &slab_start.front(), NULL, &slab_count.front(), NULL);
	assert(ret >= 0);

	ret = H5Dwrite(dset_id, mem_datatype, mem_space_id, space_id, H5P_DEFAULT, buf);
	assert(ret >= 0);

Solved:

I was not closing the old data space (spce_id) before calling H5Dget_space on the extended dataset.