Elamur Groupantsmova-serve-cache
The groupant beside tools/mova-serve/src/cache.rs in the Mova repository: what the code computes, what it never does, how to go back from its outputs, its invariants, cost and checks.
Facts
Implementation material. A Mova code model describes this code at one source revision: partial models, finite checks and their stated scope. It is not a reviewed general exposition, and its labels are shown as written, never raised.
Components
| component | label | where |
|---|---|---|
| Approximant | partial | Approximant |
| Ast | partial | syntax.json |
| Cost | symbolic source derived | Cost |
| Descriptant | present | Descriptant |
| Invariant | partial | Invariant |
| Tests | executed bounded checks | Tests |
Obernants
The typed relations around the forward model: the behaviour complement (what the code never does), the reverse reconstruction (every input compatible with an output) and specialised views.
| relation | type | label | where |
|---|---|---|---|
| Behavior complement | behavior complement | partial source derived model | Behavior complement |
| Reverse reconstruction | reverse reconstruction | partial source derived model | Reverse reconstruction |
| Model presence | model presence | partial source derived model | Model presence |
Model text
Descriptant
A byte-bounded cache with W-TinyLFU admission (Einziger, Friedman & Manes 2017): a 1% window LRU, a segmented main LRU (probation 20%, protected 80%), and a count-min sketch of 4-bit counters halved every 10 × width accesses.
Approximant
insert → window; window overflow → candidate; admit iff freq(candidate) > freq(victim); get: probation hit → protected; protected overflow → probation.
Invariant
Bytes ≤ capacity after every call; each key in exactly one segment; the eviction order is a function of the call sequence (ticks, fixed seeds, ordered maps).
Cost
O(log n) a call; a sketch reset O(width) every 10 × width accesses.
Behavior complement
No expiry by time, no thread safety of its own (the caller holds a mutex), no doorkeeper filter.
Reverse reconstruction
top recovers the protected-first order of keys; the stats recover hits, misses, admissions, rejections and
evictions.
Model presence
Cache::lru is the same structure with admission and the protected segment off: the negative control.
Tests
cache::tests: a one-off flood of 5 000 unique keys keeps all 20 popular keys under W-TinyLFU and evicts all 20
under the plain LRU control; the size bound over a skewed stream of 20 000 calls and two identical runs; aging after
five halvings; oversized values refused.
Provenance
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/tools/mova-serve/src/cache.rs.groupant/