Currently, BIGGEST_SANE_USER_ALLOC is overused and prevents user code from allocating medium-large amounts of memory, or large unreserved slabs (memtable-like techniques). We should split this into multiple thresholds and use the appropriate one in each case. The 4GB limit is only really relevant to leaf-level allocations, and then only if we intend to use libcrunch-style bounds checking for them.
Currently,
BIGGEST_SANE_USER_ALLOCis overused and prevents user code from allocating medium-large amounts of memory, or large unreserved slabs (memtable-like techniques). We should split this into multiple thresholds and use the appropriate one in each case. The 4GB limit is only really relevant to leaf-level allocations, and then only if we intend to use libcrunch-style bounds checking for them.