pgtk 0.32.7 → 0.33.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 54a200f1f191d63b73c0137b1a1ace4aa34667f041d141365b47d9c90d96f191
-  data.tar.gz: 1518e5031b71a69e3955cabe77397d05917398fb87a12b94ee5c3aa57ab107eb
+  metadata.gz: 7acb299c9e4e6cd1e272651041704fd89da2858bd1e4b2aa935eed9724815419
+  data.tar.gz: 1260a6fb42a11e34d1a733eb9d1bff8ba6a28f81724457e12986adccd134ebc5
 SHA512:
-  metadata.gz: 28bbce0d3b2cfcbe2347f376738938e0e714f22b89fd2696052c56ecf7be83c242e9f0a87c76c235f109ec2008a45a1471ea157fc78f3ac9a8b8b71641ac2556
-  data.tar.gz: e94795ee7e4de524c77b5fddefb7a0d2ea9e85ec05beb09ebf81000919afec9a7b54d8a0fb480d9e24c16990fed7b7b142769caefddaefdd32d504f8086f5f46
+  metadata.gz: 29aad3a51e77675418fc5eaffa72d7e0015d4e6d1464c87cc118f3ab94e223b0ce4461cffba91e7f4e2d22b594bb0a61f2d4b299db1055d2637d9afca5a0533d
+  data.tar.gz: 8483ec44d6da87cd0e57f57e2ca14373965269a5463bc74ba7e38ac4693b4acf35299a64b6ec91a17fc04812160ca31e6826313fce660cdd3e210c8cc3007c81

data/Gemfile.lock

@@ -36,7 +36,7 @@ GEM
       tago (~> 0.1)
     ellipsized (0.3.0)
     joined (0.4.0)
-    json (2.19.9)
+    json (2.20.0)
     language_server-protocol (3.17.0.5)
     lint_roller (1.1.0)
     logger (1.7.0)
@@ -95,7 +95,7 @@ GEM
       tago (~> 0.0)
     regexp_parser (2.12.0)
     rexml (3.4.4)
-    rubocop (1.87.0)
+    rubocop (1.88.1)
       json (~> 2.3)
       language_server-protocol (~> 3.17.0.2)
       lint_roller (~> 1.1.0)
@@ -109,7 +109,7 @@ GEM
     rubocop-ast (1.49.1)
       parser (>= 3.3.7.2)
       prism (~> 1.7)
-    rubocop-elegant (0.6.0)
+    rubocop-elegant (0.7.0)
       lint_roller (~> 1.1)
       rubocop (~> 1.75)
     rubocop-minitest (0.39.1)

data/lib/pgtk/stash.rb

@@ -6,6 +6,7 @@
 require 'concurrent-ruby'
 require 'joined'
 require 'loog'
+require 'objspace'
 require 'tago'
 require_relative '../pgtk'
 
@@ -15,6 +16,10 @@ require_relative '../pgtk'
 # the cache when tables are modified. Read queries are cached while write
 # queries bypass the cache and invalidate related cached entries.
 #
+# Tables that are too write-heavy to benefit from caching can be listed
+# via the +volatile:+ constructor parameter; any read whose FROM/JOIN
+# set touches one of them bypasses the cache entirely.
+#
 # Thread-safe with read-write locking.
 #
 # The implementation is very naive! Use it at your own risk.
@@ -34,6 +39,7 @@ class Pgtk::Stash
     REINDEX TRUNCATE CREATE ALTER DROP SET START
   ].freeze
   MODS_RE = Regexp.new("\\A(#{MODS.join('|')})(\\s|$)")
+  WITH_RE = /\AWITH(\s|$)/
 
   IDENT = '[a-z_][a-z0-9_]*'
 
@@ -48,7 +54,7 @@ class Pgtk::Stash
     CLOCK_TIMESTAMP)(?:\s*\(|(?=[^a-z0-9_]|$))
   /ix
 
-  private_constant :MODS, :ALTS, :IDENT, :MODS_RE, :ALTS_RE, :READS_RE, :NONDETERMINISTIC
+  private_constant :MODS, :ALTS, :IDENT, :MODS_RE, :WITH_RE, :ALTS_RE, :READS_RE, :NONDETERMINISTIC
 
   # Initialize a new Stash with query caching.
   #
@@ -66,11 +72,13 @@ class Pgtk::Stash
   # @param [Float] retirement Interval in seconds between retirement tasks
   # @param [Loog] loog Logger instance
   # @param [Concurrent::ReentrantReadWriteLock] entrance Read-write lock for thread-safe access
+  # @param [Array<String>] volatile Table names that must never be cached
   def initialize(
     pool,
     stash: { queries: {}, tables: {}, table_mod: {}, table_inflight: {} },
     loog: Loog::NULL,
     entrance: Concurrent::ReentrantReadWriteLock.new,
+    volatile: [],
     refill: 16,
     delay: 0,
     maxqueue: 128,
@@ -88,6 +96,7 @@ class Pgtk::Stash
     end
     @loog = loog
     @entrance = entrance
+    @volatile = Array(volatile).map(&:to_s).freeze
     @refill = refill
     @delay = delay
     @maxqueue = maxqueue
@@ -130,9 +139,9 @@ class Pgtk::Stash
   # @return [PG::Result] Query result object
   def exec(query, params = [], result = 0)
     pure = (query.is_a?(Array) ? query.join(' ') : query).gsub(/\s+/, ' ').strip
-    if MODS_RE.match?(pure)
+    if MODS_RE.match?(pure) || (WITH_RE.match?(pure) && ALTS_RE.match?(pure))
       modify(pure, params, result)
-    elsif /(^|\s)pg_[a-z_]+\(/.match?(pure)
+    elsif /(^|\s)pg_[a-z_]+\(/.match?(pure) || (!@volatile.empty? && @volatile.intersect?(pure.scan(READS_RE).flatten))
       @pool.exec(pure, params, result)
     else
       select(pure, params, result)
@@ -145,7 +154,7 @@ class Pgtk::Stash
   # @return [Object] The result of the block
   def transaction
     @pool.transaction do |t|
-      yield(Pgtk::Stash.new(t, stash: @stash, loog: @loog, entrance: @entrance))
+      yield(Pgtk::Stash.new(t, stash: @stash, loog: @loog, entrance: @entrance, volatile: @volatile))
     end
   end
 
@@ -155,7 +164,7 @@ class Pgtk::Stash
   # @return [Object] The result of the block
   def session
     @pool.session do |t|
-      yield(Pgtk::Stash.new(t, stash: @stash, loog: @loog, entrance: @entrance))
+      yield(Pgtk::Stash.new(t, stash: @stash, loog: @loog, entrance: @entrance, volatile: @volatile))
     end
   end
 
@@ -184,6 +193,7 @@ class Pgtk::Stash
         '  Not launched yet'
       end,
       "  #{cached} queries cached (#{cached > @cap ? 'above' : 'below'} the cap)",
+      "  ~#{footprint} bytes of RAM occupied by cache",
       "  #{@stash[:tables].count} tables in cache",
       "  #{list.sum { |a| a[:s] }} stale queries in cache:",
       stale(list),
@@ -326,6 +336,33 @@ class Pgtk::Stash
     end
   end
 
+  # Estimate the total heap footprint, in bytes, of everything held in
+  # +@stash[:queries]+.
+  #
+  # The count of cached entries alone is misleading: a single query with many
+  # parameter combinations keeps a distinct +PG::Result+ per combination, and
+  # the +cap+ limits the number of entries, not their weight. This gauge sums
+  # +ObjectSpace.memsize_of+ over each cached entry and its +:ret+, +:params+,
+  # and key strings, so an operator can tell whether the cache is about to
+  # exhaust memory. +Marshal+ is not an option here since neither +PG::Result+
+  # nor +Concurrent::AtomicFixnum+ is marshallable.
+  #
+  # The result is approximate: +memsize_of+ is shallow, so nested contents
+  # (rows inside a +PG::Result+) are only partially accounted for. A rough
+  # order of magnitude is enough.
+  #
+  # @return [Integer] Approximate total bytes of RAM held by the cache
+  def footprint
+    @entrance.with_read_lock do
+      @stash[:queries].sum do |q, kk|
+        ObjectSpace.memsize_of(q) +
+          kk.sum do |params, entry|
+            [params, entry, entry[:ret], entry[:params]].sum { |o| ObjectSpace.memsize_of(o) }
+          end
+      end
+    end
+  end
+
   # Discover ON DELETE CASCADE / ON UPDATE CASCADE foreign keys so that a
   # modify on the parent table also invalidates cached queries on children.
   def cascade!

data/lib/pgtk/version.rb

@@ -10,5 +10,5 @@ require_relative '../pgtk'
 # Copyright:: Copyright (c) 2019-2026 Yegor Bugayenko
 # License:: MIT
 module Pgtk
-  VERSION = '0.32.7' unless defined?(VERSION)
+  VERSION = '0.33.0' unless defined?(VERSION)
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: pgtk
 version: !ruby/object:Gem::Version
-  version: 0.32.7
+  version: 0.33.0
 platform: ruby
 authors:
 - Yegor Bugayenko