lex-apollo 0.4.24 → 0.4.26

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 2dceee29e5b2af12bab5833beac8b837d2366eaf9509c7e4ce04e88b44a5c28f
-  data.tar.gz: 97d62a6385f375b19a6f48b1881fef8525e5949c4348bfe0cd07581d9ff710f2
+  metadata.gz: fe40d4ba4efef56b2bd1cec166f525ca3c165989453e455bdacc14a2a5377d16
+  data.tar.gz: e4f7f0daabec002d031c00a6936308df68d75dd2200b1f49f7264de6f2833153
 SHA512:
-  metadata.gz: 90065729231f3322eeaaeaa0e44e03b9b79350157674bec30f174fa6447d74bc53b4794e3385eee9d617c13c62cb22198de39fc21ebece7d2d512e7b1dd09ba6
-  data.tar.gz: f1c1b607fd9d9d244c0937b654ea95ca361980f5732c6e513719a6163004595ce0aaa9baab4181db56c8e6e8d1c511c1d6695c9110905e2094a252ba06e5a2c6
+  metadata.gz: 7d0995197fb2c7191a81a91bba476792570baaf5e750df8b0464de096beabc82a4fbec8fabb8b09fda967d5c834c7bf682041d8775ce24c602d1790e332692c7
+  data.tar.gz: 6c0277a64242c3c37795dde7229db4d2f0c27d191a7ad5d252266db1c4b433538ec5741f57624a3e262d1ad1b38ca9abf9c0ec1b2463a7bd69b9fac2126cb190

data/CHANGELOG.md

@@ -1,5 +1,16 @@
 # Changelog
 
+## [0.4.26] - 2026-05-11
+
+### Fixed
+- Handle `Sequel::UniqueConstraintViolation` in `create_candidate_entry` gracefully — a race condition during concurrent knowledge ingestion can cause two threads to pass the content_hash dedup check simultaneously and both attempt to insert the same row. On collision, the rescue block now looks up the existing winner row by content_hash (excluding archived) and returns its ID so the caller continues normally (access log, contradiction detection, etc.) instead of propagating a database error.
+- Added `Sequel::UniqueConstraintViolation` stub to the test-only Sequel shim so the race-condition rescue path is exercisable in unit tests without a live database.
+
+## [0.4.25] - 2026-05-08
+
+### Fixed
+- Entity watchdog now remembers processed task-log text during the process lifetime so unchanged logs do not trigger repeated structured LLM extraction on every watchdog interval.
+
 ## [0.4.24] - 2026-05-07
 
 ### Fixed

data/lib/legion/extensions/apollo/actors/entity_watchdog.rb

@@ -1,5 +1,6 @@
 # frozen_string_literal: true
 
+require 'digest'
 require 'legion/extensions/actors/every'
 require_relative '../runners/knowledge'
 require_relative '../runners/entity_extractor'
@@ -40,6 +41,8 @@ module Legion
 
             ingested = 0
             texts.each do |text|
+              next if task_log_text_processed?(text)
+
               result = extract_entities(
                 text:           text,
                 entity_types:   entity_types,
@@ -47,6 +50,7 @@ module Legion
               )
               next unless result[:success]
 
+              mark_task_log_text_processed(text) unless result[:source] == :unavailable
               result[:entities].each do |entity|
                 next if entity_exists_in_apollo?(entity)
 
@@ -124,6 +128,28 @@ module Legion
           def dedup_similarity_threshold
             settings[:entity_watchdog][:dedup_threshold].to_f
           end
+
+          def task_log_text_processed?(text)
+            processed_task_log_hashes.key?(task_log_text_hash(text))
+          end
+
+          def mark_task_log_text_processed(text)
+            hashes = processed_task_log_hashes
+            hashes[task_log_text_hash(text)] = true
+            hashes.shift while hashes.size > processed_task_log_hash_limit
+          end
+
+          def processed_task_log_hashes
+            @processed_task_log_hashes ||= {}
+          end
+
+          def processed_task_log_hash_limit
+            [settings[:entity_watchdog][:log_limit].to_i * 4, 100].max
+          end
+
+          def task_log_text_hash(text)
+            Digest::SHA256.hexdigest(text.to_s)
+          end
         end
       end
     end

data/lib/legion/extensions/apollo/runners/knowledge.rb

@@ -469,6 +469,22 @@ module Legion
             )
             log.info("Apollo Knowledge.handle_ingest created entry_id=#{new_entry.id} status=candidate domain=#{metadata[:domain]} source_agent=#{metadata[:source_agent]}") # rubocop:disable Layout/LineLength
             new_entry.id
+          rescue Sequel::UniqueConstraintViolation => e
+            # Race condition: another thread/process inserted the same content_hash between our
+            # dedup check and this insert. Fetch and return the winner's id so the caller can
+            # continue normally (access log, contradiction detection, etc.).
+            winner = Helpers::DataModels.apollo_entry
+                                        .where(content_hash: content_hash)
+                                        .exclude(status: 'archived')
+                                        .first
+            if winner
+              log.warn("Apollo Knowledge.create_candidate_entry race_dedup entry_id=#{winner.id} content_hash=#{content_hash} source_agent=#{metadata[:source_agent]}") # rubocop:disable Layout/LineLength
+              winner.id
+            else
+              handle_exception(e, level: :warn, handled: true, operation: 'apollo.knowledge.create_candidate_entry',
+                                  content_hash: content_hash)
+              nil
+            end
           end
 
           def browse_query?(query)

data/lib/legion/extensions/apollo/version.rb

@@ -3,7 +3,7 @@
 module Legion
   module Extensions
     module Apollo
-      VERSION = '0.4.24'
+      VERSION = '0.4.26'
     end
   end
 end

data/spec/legion/extensions/apollo/actors/entity_watchdog_spec.rb

@@ -67,6 +67,13 @@ RSpec.describe Legion::Extensions::Apollo::Actor::EntityWatchdog do
       expect(actor).to have_received(:publish_entity_ingest).once
     end
 
+    it 'does not extract entities from the same task log text twice' do
+      actor.scan_and_ingest
+      actor.scan_and_ingest
+
+      expect(actor).to have_received(:extract_entities).once
+    end
+
     context 'when entity already exists in Apollo (high similarity)' do
       let(:existing_match) do
         { success: true, entries: [{ id: 42, content: 'lex-synapse', distance: 0.02 }], count: 1 }

data/spec/legion/extensions/apollo/runners/knowledge_spec.rb

@@ -321,6 +321,30 @@ RSpec.describe Legion::Extensions::Apollo::Runners::Knowledge do
           expect(result[:deduped]).to be true
           expect(result[:entry_id]).to eq('uuid-existing')
         end
+
+        it 'recovers gracefully when a concurrent ingest wins the content_hash unique constraint race' do
+          # Simulate: dedup check passes (nil — no existing entry yet), then .create
+          # raises UniqueConstraintViolation (another thread inserted between check and insert).
+          # create_candidate_entry must rescue and return the existing entry's id so the
+          # caller succeeds rather than propagating a database error.
+          race_entry = double('race_entry', id: 'uuid-race-winner')
+
+          allow(mock_entry_class).to receive(:create)
+            .and_raise(Sequel::UniqueConstraintViolation, 'duplicate key value violates unique constraint "idx_apollo_content_hash"')
+
+          collision_dataset = double('collision_dataset')
+          allow(mock_entry_class).to receive(:where).with(content_hash: anything).and_return(collision_dataset)
+          allow(collision_dataset).to receive(:exclude).with(status: 'archived').and_return(collision_dataset)
+          # First call: dedup pre-check returns nil (not yet in DB).
+          # Second call: post-collision lookup returns the winner inserted by the other thread.
+          allow(collision_dataset).to receive(:first).and_return(nil, race_entry)
+
+          result = host.handle_ingest(content: 'concurrent content', content_type: 'fact',
+                                      source_agent: 'agent-1',
+                                      content_hash: 'd3861b2862454c5a6a9e480829333841')
+          expect(result[:success]).to be true
+          expect(result[:entry_id]).to eq('uuid-race-winner')
+        end
       end
     end
 

data/spec/spec_helper.rb

@@ -13,6 +13,7 @@ require 'legion/transport'
 unless defined?(Sequel)
   module Sequel
     class Error < StandardError; end
+    class UniqueConstraintViolation < Error; end
 
     def self.pg_array(arr) = arr
     def self.lit(str, *) = str

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: lex-apollo
 version: !ruby/object:Gem::Version
-  version: 0.4.24
+  version: 0.4.26
 platform: ruby
 authors:
 - Esity