What is total_allocations?

[peekxc]

What is total allocations?

For whatever reason, this only shows up in the JSON report (see #620).

Suppose I get an output like this:

{'total_num_allocations': 2844284,
 'total_bytes_allocated': 526180830,
 'allocation_size_histogram': [...],
 'allocator_type_distribution': {'PYMALLOC_MALLOC': 2721820,
  'PYMALLOC_REALLOC': 122350,
  'REALLOC': 86,
  'PYMALLOC_CALLOC': 22,
  'MALLOC': 6},
 'top_allocations_by_size': [...],
 'metadata': {
  'total_allocations': 5686611,
  'total_frames': 46,
  'peak_memory': 4073520,
  'python_allocator': 'pymalloc',
  'has_native_traces': False,
  'trace_python_allocators': True,
  'file_format': 0}}

Notice that total_num_allocations doesn't match total_allocations. It would be nice in memray included somewhere on its website a description of:

• what total_num_allocations is
• what total_allocations is
• what total_frames is
• what precisely peak_memory is

[godlygeek]

Yeah, that's misleading.

total_allocations is the number of allocations and deallocations that Memray observed. That's because it's counting the number of records of type AllocationRecord or NativeAllocationRecord that our record writer has processed, and we use the same record type for free as for malloc. This is why it's within 1% of double total_num_allocations:

5686611 / (2844284 * 2) = 0.99966

total_num_allocations is only counting allocations, not deallocations.

Similarly to total_allocations, total_frames is counting the number of distinct Python frame records that Memray's record writer processed, which is certainly not interesting to you. It's not a count of distinct Python frame objects (the same object might be double counted if we see allocations from multiple lines within that frame). It's a statistic that might be mildly interesting to someone who is trying to interpret or optimize Memray's binary record format, but which certainly isn't interesting to end users of Memray.

peak_memory is simply tracking the maximum number of bytes that were allocated and not yet freed while Memray tracking was active. At any given point in time, we say that the heap memory usage is the sum of the sizes of all allocations that were made since Memray tracking started and that have not yet been freed. Peak memory usage is the maximum value of heap memory usage over the entire tracking session.