legion-llm 0.12.2 → 0.12.3

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: b5c5db81b61e68e29fd4a0ebdb612b8032145bac6ef1e2e70a6578b4dff3c9e4
-  data.tar.gz: 6ea03a18b9e7e448607d168af2c8c54bbccee3bb942670cd6adc5d1e673ca115
+  metadata.gz: 733b0a06bf37557dfe7c661d6d63596e30ef5cde4b01c9ae12e801c5257b684e
+  data.tar.gz: f24cfe5b32ed137a47efaac7adf57f348d911cb1f511fbdfc5a769ce54d7c32a
 SHA512:
-  metadata.gz: 13ec651f801eede51e8389244609bd4d74a38ecd385815aeca5bb22ef9cb9a6df35ef12ab0bb99d92e1fa96f08a730d72d03302d5a268631db38c96fdf5ba938
-  data.tar.gz: e9798e55b838b440ec1608ddb25424124bf36885a7fef46c3c445fc662361a7c5a82951aa38aedc63934242d363fb67ebdad6ff13060eee272a234b6ef450da7
+  metadata.gz: c006671e6a1dc02bff77b18e3c31dc4cb771ffd1ba5a6b773e2d7a62a9e886b76e95fb1e9780b0c1285e0165a218112f239fe088ceae9296da440918bd682648
+  data.tar.gz: 14eaff3d563eb80e058f699bca53c79308642f745710c1a546b7e00ecfde049129d90e91a52cea5e13a0e19c39afb1282daf1228640680b8f7b94db777676f81

data/CHANGELOG.md

@@ -1,5 +1,10 @@
 # Legion LLM Changelog
 
+## [0.12.3] - 2026-06-02
+
+### Fixed
+- **Escalation retries all instances of a broken provider** — When a `StandardError` fires during an escalation attempt (e.g. a code bug in the provider adapter), the loop previously retried every other registered instance of the same `provider+model` combination before giving up. New `skip_all_provider_model_instances!` marks all chain entries with the same `provider+model` as tried on the first `StandardError`, so escalation immediately advances to a different provider. Transient error paths (rate limit, auth, context overflow) are unaffected (inference/executor.rb)
+
 ## [0.12.2] - 2026-06-02
 
 ### Fixed

data/lib/legion/llm/inference/executor.rb

@@ -790,7 +790,7 @@ module Legion
                                     handled:   true)
           false
         rescue StandardError => e
-          tried << { provider: resolution.provider, instance: resolution.instance, model: resolution.model }
+          skip_all_provider_model_instances!(resolution, tried)
           record_escalation_failure(e, resolution, start_time,
                                     outcome:   :error,
                                     operation: 'llm.pipeline.escalation_attempt')
@@ -935,6 +935,20 @@ module Legion
           end
         end
 
+        def skip_all_provider_model_instances!(failed_resolution, tried)
+          chain = @escalation_chain
+          return unless chain.respond_to?(:each)
+
+          chain.each do |r|
+            next if r.provider != failed_resolution.provider || r.model != failed_resolution.model
+            next if tried.any? { |t| t[:provider] == r.provider && t[:instance] == r.instance && t[:model] == r.model }
+
+            log.warn "[llm][escalation] action=skip_provider_model provider=#{r.provider} model=#{r.model} " \
+                     "instance=#{r.instance} reason=provider_error"
+            tried << { provider: r.provider, instance: r.instance, model: r.model }
+          end
+        end
+
         def escalation_attempt_hash(resolution, outcome:, failures:, duration_ms:)
           attempt = { model: resolution.model, provider: resolution.provider, tier: resolution.tier,
                       outcome: outcome, failures: failures, duration_ms: duration_ms }

data/lib/legion/llm/version.rb

@@ -2,6 +2,6 @@
 
 module Legion
   module LLM
-    VERSION = '0.12.2'
+    VERSION = '0.12.3'
   end
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: legion-llm
 version: !ruby/object:Gem::Version
-  version: 0.12.2
+  version: 0.12.3
 platform: ruby
 authors:
 - Esity