decentraland / catalyst-storage / 27284293273

95%

push

github

fix: deduplicate concurrent decompressions of the same gzip (#104)

* fix: deduplicate concurrent decompressions of the same gzip

The in-flight guard read inflightDecompressions, then `await`ed the gzip
lookup, and only then registered the promise. Two concurrent cold-cache
range requests for the same file therefore both saw "not in flight",
both decompressed it, and both added its size to totalCacheSize — so the
guard never actually deduplicated (verified: 16 concurrent requests ran
16 decompressions), wasting CPU/IO and drifting the cache-size accounting
upward (causing premature eviction over time).

Create and register the promise synchronously (no await between the get
and the set), with the gzip lookup moved inside it, so concurrent callers
share a single decompression. Adds a regression test asserting the file is
decompressed exactly once under 16 concurrent range requests.

* test: derive shard path instead of hardcoding the hash prefix

Addresses review feedback: compute the storage shard from sha1(id) the way
the component does, rather than hardcoding "9584", so the race test isn't
coupled to the internal sharding constant.

106 of 119 branches covered (89.08%)

Branch coverage included in aggregate %.

16 of 16 new or added lines in 1 file covered. (100.0%)

370 of 377 relevant lines covered (98.14%)

38.4 hits per line