Class: Beacon::LRU

Inherits:
Object
  • Object
show all
Defined in:
lib/beacon/lru.rb

Overview

Bounded LRU cache, Mutex-guarded.

Exploits the fact that Ruby’s Hash preserves insertion order: on a read hit we delete + re-insert (moves the key to the tail), and on overflow we shift the oldest entry. No linked list, no external dependency.

The critical section is intentionally minimal — [], []=, and compute_if_absent all run a handful of Hash operations under the Mutex. The middleware’s per-request path cache lives behind this class, so the overhead is on the hot path.

Footgun: this LRU does NOT distinguish a stored nil from a miss. ‘lru` returns nil in both cases. Callers that need to cache nil should store a sentinel (e.g. :__missing__) instead.

Instance Method Summary collapse

Constructor Details

#initialize(max:) ⇒ LRU

Returns a new instance of LRU.

Raises:

  • (ArgumentError) 


17
18
19
20
21
22
 
# File 'lib/beacon/lru.rb', line 17

def initialize(max:)
  raise ArgumentError, "max must be positive" unless max.is_a?(Integer) && max > 0
  @max   = max
  @store = {}
  @mutex = Mutex.new
end

Instance Method Details

#[](key) ⇒ Object 



24
25
26
27
28
29
30
31
 
# File 'lib/beacon/lru.rb', line 24

def [](key)
  @mutex.synchronize do
    return nil unless @store.key?(key)
    value = @store.delete(key)
    @store[key] = value
    value
  end
end

#[]=(key, value) ⇒ Object 



33
34
35
36
37
38
39
40
41
42
 
# File 'lib/beacon/lru.rb', line 33

def []=(key, value)
  @mutex.synchronize do
    @store.delete(key) if @store.key?(key)
    @store[key] = value
    # A single []= can grow the store by at most one entry, so an
    # `if` is sufficient — no need for `while`.
    @store.shift if @store.size > @max
    value
  end
end

#clear ⇒ Object 



63
64
65
 
# File 'lib/beacon/lru.rb', line 63

def clear
  @mutex.synchronize { @store.clear }
end

#compute_if_absent(key) ⇒ Object

Read-or-compute primitive. On hit, returns the cached value. On miss, yields the key, stores the result, and returns it.

The compute block runs OUTSIDE the mutex. Two threads concurrently missing on the same key will both run the block — the second []= then overwrites the first. Use this only when the block is idempotent and cheap (e.g. pure transforms like PathNormalizer). Callers that need exactly-once compute under contention should build their own guard around this class.



53
54
55
56
57
 
# File 'lib/beacon/lru.rb', line 53

def compute_if_absent(key)
  hit = self[key]
  return hit unless hit.nil?
  self[key] = yield(key)
end

#size ⇒ Object 



59
60
61
 
# File 'lib/beacon/lru.rb', line 59

def size
  @mutex.synchronize { @store.size }
end