Fix MEMORY USAGE command

[tomerqodo]

Benchmark PR from agentic-review-benchmarks#5

[augmentcode]

🤖 Augment PR Summary

Summary: Fixes MEMORY USAGE reporting to better account for the actual in-memory footprint of key/value objects.

Changes:

• Renamed and updated the size estimator to kvobjComputeSize, aiming to include kv-object header + embedded key (and embedded value where applicable).
• Adjusted per-type memory accounting for strings, lists, sets, zsets, hashes, streams, and module values.
• Updated MEMORY USAGE command implementation to use the new estimator.
• Refreshed unit-test coverage comments and added a new string-focused MEMORY USAGE test.

Technical Notes: The estimator relies on allocator-reported allocation sizes and sampling to approximate nested container memory usage.

_{🤖 Was this summary useful? React with 👍 or 👎}

[augmentcode]

Review completed. 3 suggestions posted.

Comment augment review to trigger a new review at any time.

[augmentcode]

kvobjComputeSize uses malloc_usable_size(), which may not be available / may not match the allocator Redis is built with (e.g., jemalloc uses je_malloc_usable_size via zmalloc_*). This can break portability and/or make MEMORY USAGE inconsistent with zmalloc_size-based accounting used elsewhere.

_{🤖 Was this useful? React with 👍 or 👎}

[augmentcode]

In the quicklist branch, asize += (double)elesize/samples*ql->count appears to mix “per-node” averaging (/samples) with “total element count” scaling (*ql->count), which can massively mis-estimate list memory usage. It looks like this should be normalized consistently (e.g., per-element vs per-node) before scaling.

_{🤖 Was this useful? React with 👍 or 👎}

[augmentcode]

In the stream branch, asize += lpsize * s->rax->numele is followed by adding the actual size of the last listpack, which seems to double-count the last listpack (the old code used numele-1 before adding the last). This likely inflates MEMORY USAGE results for streams.

_{🤖 Was this useful? React with 👍 or 👎}