lex-llm-azure-foundry 0.2.10 → 0.2.14

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 7e151d90b17f955ff40511cc8e135f909ca7fd45d54147f357947bce824d04dd
-  data.tar.gz: c6972df094e1e34636cea4762ed40dd75737080231a4af3c21f1f4062676056d
+  metadata.gz: 30c68d11d7549c672c912006d7ca6e843fbd976cd7e5e1845db580b9f86543f8
+  data.tar.gz: 5451587d09659bd54f9f6a5d52c165f1eee58a66a2667797a65a7f2f84c9fbce
 SHA512:
-  metadata.gz: 3354eab641cb8026d825eb32b2a014316daa17b8fbccc659b067efc7d52c2ff999d3a624ecbca54773c50e36a83f0cea3bacc52a7e903c268010dbc81bfe5c2b
-  data.tar.gz: 15e2d75ed42617bdc0a9c49be74e42b09aaa70a5e9c45f12cae16320f78d33180058a233f64c043e012433b76021d69dd2dfdcd078502aa3727bfa203b64efb2
+  metadata.gz: 0a9509a27ea507bfa9cf0a68564cb56cafad2ca30fe8f158a2a1a33dcdb701dc3b25289ce0707a524c9488f54f12421005702b31b3bf6d2b35eb869bc79f77e8
+  data.tar.gz: 0dd91d0a237fa15c9a6fcb32f8c77d81c15274e97fc525d08a254787e91f410449b01448d43e677f9f785343738f33e0de6a0bb07076af45dc7c2ccebbf67cf6

data/CHANGELOG.md

@@ -1,5 +1,31 @@
 # Changelog
 
+## [0.2.14] - 2026-07-03
+
+### Fixed
+- Emit relative request paths from `path_for` (no leading slash). `Connection` builds Faraday with `api_base` as the base URL; on the `openai_v1` surface that base carries the `/openai/v1` path, and a leading-slash path was treated as absolute and dropped it — 404ing live discovery (empty offerings) and chat. Paths are now relative so the base path survives on both surfaces.
+- Resolve `models_url`/`health_url` per surface: `models` on `openai_v1`, `models/info` on `model_inference`. Previously always `info`, which 404s on the `openai_v1` surface.
+
+## [0.2.13] - 2026-06-20
+
+### Fixed
+- Stub shared registry publishing through `RegistryPublisher#schedule` in specs so async availability-event coverage stays stable after the shared publisher moved off raw `Thread.new`.
+
+## [0.2.12] - 2026-06-20
+
+### Changed
+- Align Azure Foundry instance discovery with the shared `lex-llm` contract by preserving explicit tier overrides while defaulting unconfigured instances to `:cloud`.
+- Restore offline deployment-backed offering discovery and carry the configured provider instance id through Azure offering metadata.
+- Normalize Azure Foundry capability and health metadata to the current shared offering contract.
+
+## [0.2.11] - 2026-06-19
+
+### Changed
+- Adopt `Legion::Extensions::Llm::Inventory::ScopedRefresher` mixin (lex-llm 0.6.0). Discovery
+  refresh actors now write directly to the live `Inventory` catalog via `Inventory.write_lane`.
+- Pin `lex-llm >= 0.6.0` and `legion-llm >= 0.14.0` in gemspec.
+- Standard `weight: 100` default added to provider instance settings schema.
+
 ## 0.2.10 - 2026-06-16
 
 - Dependency updates and code quality improvements.

data/lex-llm-azure-foundry.gemspec

@@ -27,5 +27,5 @@ Gem::Specification.new do |spec|
   spec.add_dependency 'legion-logging', '>= 1.3.2'
   spec.add_dependency 'legion-settings', '>= 1.3.14'
   spec.add_dependency 'legion-transport', '>= 1.4.14'
-  spec.add_dependency 'lex-llm', '>= 0.5.0'
+  spec.add_dependency 'lex-llm', '>= 0.6.0'
 end

data/lib/legion/extensions/llm/azure_foundry/actors/discovery_refresh.rb

@@ -1,11 +1,19 @@
 # frozen_string_literal: true
 
+require 'digest'
+
 begin
   require 'legion/extensions/actors/every'
 rescue LoadError => e
   warn(e.message) if $VERBOSE
 end
 
+begin
+  require 'legion/extensions/llm/inventory/scoped_refresher'
+rescue LoadError => e
+  warn(e.message) if $VERBOSE
+end
+
 return unless defined?(Legion::Extensions::Actors::Every)
 
 module Legion
@@ -13,10 +21,16 @@ module Legion
     module Llm
       module AzureFoundry
         module Actor
-          class DiscoveryRefresh < Legion::Extensions::Actors::Every # rubocop:disable Style/Documentation
+          class DiscoveryRefresh < Legion::Extensions::Actors::Every # rubocop:disable Style/Documentation,Metrics/ClassLength
             include Legion::Logging::Helper
 
-            REFRESH_INTERVAL = 1800
+            if defined?(Legion::Extensions::Llm::Inventory::ScopedRefresher)
+              include Legion::Extensions::Llm::Inventory::ScopedRefresher
+            end
+
+            EMBED_TYPES = %i[embed embedding].freeze
+
+            def self.every_seconds = 3600
 
             def runner_class    = self.class
             def runner_function = 'manual'
@@ -26,13 +40,15 @@ module Legion
             def generate_task?  = false
 
             def time
-              return REFRESH_INTERVAL unless defined?(Legion::Settings)
+              return self.class.every_seconds unless defined?(Legion::Settings)
 
-              Legion::Settings.dig(:extensions, :llm, :azure_foundry, :discovery_interval) || REFRESH_INTERVAL
+              Legion::Settings.dig(:extensions, :llm, :azure_foundry, :discovery_interval) || self.class.every_seconds
             end
 
-            def manual
+            def manual # rubocop:disable Metrics/CyclomaticComplexity
               log.debug('[azure_foundry][discovery_refresh] refreshing model list')
+              tick if respond_to?(:tick)
+
               return unless defined?(Legion::LLM::Discovery)
 
               Legion::LLM::Discovery.refresh_discovered_models!(provider: :azure_foundry)
@@ -46,6 +62,109 @@ module Legion
             rescue StandardError => e
               handle_exception(e, level: :warn, handled: true, operation: 'azure_foundry.actor.discovery_refresh')
             end
+
+            def scope_key(**)
+              { provider: :azure_foundry }
+            end
+
+            def compute_lanes_for_scope(**)
+              return [] unless defined?(Legion::LLM::Call::Registry)
+
+              instances = Legion::LLM::Call::Registry.all_instances.select do |e|
+                (e[:provider] || '').to_sym == :azure_foundry
+              end
+
+              lanes = []
+              instances.each { |entry| collect_instance_lanes(entry, lanes) }
+              lanes
+            rescue StandardError => e
+              handle_exception(e, level: :warn, handled: true,
+                                  operation: 'azure_foundry.compute_lanes_for_scope')
+              []
+            end
+
+            def credential_hash(**)
+              settings = Legion::Settings.dig(:extensions, :llm, :azure_foundry) || {}
+              ::Digest::SHA256.hexdigest(settings[:api_key].to_s + settings[:instances].to_s)[0, 16]
+            rescue StandardError
+              'unknown'
+            end
+
+            private
+
+            def collect_instance_lanes(instance_entry, lanes)
+              adapter = instance_entry[:adapter]
+              return unless adapter.respond_to?(:discover_offerings)
+
+              Array(adapter.discover_offerings(live: true)).each do |offering|
+                build_and_append_lanes(offering, instance_entry, lanes)
+              end
+            rescue StandardError => e
+              handle_exception(e, level: :warn, handled: true,
+                                  operation: 'azure_foundry.compute_lanes_for_scope.instance')
+            end
+
+            def build_and_append_lanes(offering, instance_entry, lanes)
+              raw = offering.respond_to?(:to_h) ? offering.to_h : offering
+              return unless raw.is_a?(Hash)
+
+              lane = build_lane(raw, instance_entry)
+              lanes << lane
+              lanes << lane.merge(id: fleet_id_for(lane), tier: :fleet) if fleet_enabled? && lane[:type] == :inference
+            end
+
+            def build_lane(raw, instance_entry) # rubocop:disable Metrics/AbcSize,Metrics/CyclomaticComplexity,Metrics/PerceivedComplexity
+              model       = raw[:model] || raw['model']
+              instance_id = raw[:instance_id] || raw['instance_id'] ||
+                            instance_entry[:instance] || instance_entry[:instance_id] ||
+                            instance_entry[:id] || :default
+              pf          = raw[:provider_family] || raw['provider_family'] || :azure_foundry
+              type        = resolve_type(raw)
+              tier        = (raw[:tier] || raw['tier'] || :cloud).to_sym
+              lane_id     = compose_lane_id(tier: tier, provider_family: pf, instance_id: instance_id,
+                                            type: type, model: model)
+
+              {
+                id: lane_id,
+                tier: tier,
+                provider_family: pf,
+                instance_id: instance_id,
+                model: model,
+                canonical_model_alias: raw[:canonical_model_alias] || raw['canonical_model_alias'],
+                type: type,
+                capabilities: normalize_capabilities(raw[:capabilities] || raw['capabilities'] || []),
+                limits: raw[:limits] || raw['limits'] || {},
+                enabled: raw.fetch(:enabled, raw.fetch('enabled', true)),
+                cost: raw[:cost] || raw['cost'] || {}
+              }
+            end
+
+            def resolve_type(raw)
+              val = raw[:type] || raw['type'] || raw[:usage_type] || raw['usage_type']
+              EMBED_TYPES.include?(val&.to_sym) ? :embedding : :inference
+            end
+
+            def normalize_capabilities(raw)
+              if defined?(Legion::Extensions::Llm::Inventory::Capabilities)
+                Legion::Extensions::Llm::Inventory::Capabilities.normalize(raw)
+              else
+                Array(raw)
+              end
+            end
+
+            def compose_lane_id(**fields)
+              Legion::Extensions::Llm::Inventory::ScopedRefresher.compose_id(fields)
+            end
+
+            def fleet_id_for(lane)
+              compose_lane_id(tier: :fleet, provider_family: lane[:provider_family],
+                              instance_id: lane[:instance_id], type: lane[:type], model: lane[:model])
+            end
+
+            def fleet_enabled?
+              settings = Legion::Settings.dig(:extensions, :llm, :azure_foundry) || {}
+              settings[:fleet]&.dig(:dispatch, :enabled)
+            end
           end
         end
       end

data/lib/legion/extensions/llm/azure_foundry/provider.rb

@@ -121,6 +121,10 @@ module Legion
 
           def stream_usage_supported? = true
 
+          def settings
+            AzureFoundry.default_settings.dig(:instances, :default)
+          end
+
           def api_base
             endpoint = config.azure_foundry_endpoint.to_s.sub(%r{/*\z}, '')
             return "#{endpoint}/openai/v1" if surface == OPENAI_V1_SURFACE && !endpoint.end_with?('/openai/v1')
@@ -139,33 +143,21 @@ module Legion
           def completion_url = path_for('chat/completions')
           def chat_url = completion_url
           def stream_url = completion_url
-          def models_url = path_for('info')
+          def models_url = surface == MODEL_INFERENCE_SURFACE ? path_for('info') : path_for('models')
           def embedding_url(**) = path_for('embeddings')
           def health_url = models_url
 
-          def discover_offerings(live: false, **filters)
-            log.info { "discovering offerings live=#{live} from #{api_base}" }
-            offerings = filter_offerings(allowed_offerings, **filters)
-            return offerings unless live
-
-            offerings.map do |offering|
-              with_live_metadata(offering)
-            rescue StandardError => e
-              handle_exception(e, level: :warn, handled: true, operation: 'azure_foundry.discover_offerings')
-              with_health(offering, ready: false, checked: true, error: e)
-            end
-          end
-
-          def offering_for(model:, model_family: nil, canonical_model_alias: nil, instance_id: :default, # rubocop:disable Metrics/ParameterLists
+          def offering_for(model:, model_family: nil, canonical_model_alias: nil, instance_id: nil, # rubocop:disable Metrics/ParameterLists, Metrics/AbcSize
                            usage_type: nil, **metadata)
             deployment = self.class.deployment_config(model, config:)
             model_id = self.class.resolve_model_id(model, config:)
             configured_family = value_for(deployment, :model_family)
             configured_alias = value_for(deployment, :canonical_model_alias)
+            resolved_instance_id = instance_id || provider_instance_id
 
             build_offering(
               model: model_id,
-              instance_id: instance_id,
+              instance_id: resolved_instance_id,
               model_family: normalize_family(model_family || configured_family || infer_model_family(model_id)),
               canonical_model_alias: canonical_model_alias || configured_alias,
               usage_type: usage_type || value_for(deployment, :usage_type) || usage_type_for(model_id),
@@ -179,10 +171,12 @@ module Legion
             return baseline.merge(checked: false) unless live
 
             response = connection.get(health_url)
-            baseline.merge(checked: true, model_info: response.body)
+            baseline.merge(checked: true, ready: true, status: 'healthy', circuit_state: 'closed',
+                           raw: response.body)
           rescue StandardError => e
             handle_exception(e, level: :warn, handled: true, operation: 'azure_foundry.health')
-            baseline.merge(checked: true, ready: false, error: e.class.name, message: e.message)
+            baseline.merge(checked: true, ready: false, status: 'unhealthy', circuit_state: 'open',
+                           error: e.class.name, message: e.message)
           end
 
           def readiness(live: false)
@@ -258,11 +252,15 @@ module Legion
           end
 
           def health_baseline(live)
+            ready = configured?
             {
               provider: :azure_foundry,
-              configured: configured?,
-              ready: configured?,
+              instance_id: provider_instance_id,
+              configured: ready,
+              ready: ready,
               live: live,
+              status: ready ? 'healthy' : 'unhealthy',
+              circuit_state: ready ? 'closed' : 'open',
               api_base: api_base,
               surface: surface
             }
@@ -287,10 +285,14 @@ module Legion
             config.azure_foundry_api_version || DEFAULT_API_VERSION
           end
 
+          # Paths MUST be relative (no leading slash). Faraday builds the
+          # connection with api_base as the base URL — on the openai_v1 surface
+          # that base carries the /openai/v1 path, and a leading-slash path would
+          # be treated as absolute and drop it, 404ing discovery and chat.
           def path_for(path)
-            prefix = surface == MODEL_INFERENCE_SURFACE ? '/models' : ''
+            prefix = surface == MODEL_INFERENCE_SURFACE ? 'models/' : ''
             suffix = surface == MODEL_INFERENCE_SURFACE ? "?api-version=#{api_version}" : ''
-            "#{prefix}/#{path}#{suffix}"
+            "#{prefix}#{path}#{suffix}"
           end
 
           def bearer_header
@@ -298,6 +300,20 @@ module Legion
             token ? "Bearer #{token}" : nil
           end
 
+          def discover_offerings(live: false, raise_on_unreachable: false, **filters)
+            offerings = allowed_offerings
+            return filter_offerings(offerings, **filters) unless live
+
+            resolved = offerings.map { |offering| with_live_metadata(offering) }
+            filter_offerings(resolved, **filters)
+          rescue Faraday::ConnectionFailed, Faraday::TimeoutError => e
+            log.warn("[#{slug}] instance=#{provider_instance_id} unreachable: #{e.message}")
+            raise if raise_on_unreachable
+
+            []
+          end
+          public :discover_offerings
+
           def configured_deployments
             self.class.normalize_deployments(config.azure_foundry_deployments)
           end
@@ -315,16 +331,34 @@ module Legion
           end
 
           def offering_from_config(deployment)
-            deployment_name = value_for(deployment, :deployment) || value_for(deployment, :model)
-            return nil if deployment_name.to_s.empty?
-
             offering_for(
-              model: deployment_name,
+              model: value_for(deployment, :deployment) || value_for(deployment, :model),
               model_family: value_for(deployment, :model_family),
               canonical_model_alias: value_for(deployment, :canonical_model_alias),
-              instance_id: value_for(deployment, :instance_id) || :default,
-              usage_type: value_for(deployment, :usage_type),
-              configured: true
+              instance_id: provider_instance_id,
+              usage_type: value_for(deployment, :usage_type) || :inference,
+              metadata: deployment_metadata(deployment)
+            )
+          end
+
+          def offering_from_model(model_info, _health: nil) # rubocop:disable Metrics/AbcSize, Metrics/CyclomaticComplexity, Metrics/PerceivedComplexity
+            model_id = model_info.respond_to?(:id) ? model_info.id : model_info['id']
+            name = model_info.respond_to?(:name) ? model_info.name : model_info['name'] || model_id
+            family = model_info.respond_to?(:family) ? model_info.family : model_info['model_family']
+            usage = model_info.respond_to?(:embedding?) && model_info.embedding? ? :embedding : :inference
+            meta = if model_info.respond_to?(:metadata)
+                     model_info.metadata
+                   else
+                     model_info.is_a?(Hash) ? model_info : {}
+                   end
+
+            offering_for(
+              model: model_id,
+              model_family: family,
+              canonical_model_alias: name,
+              instance_id: provider_instance_id,
+              usage_type: usage,
+              metadata: meta
             )
           end
 
@@ -376,7 +410,7 @@ module Legion
 
           def resolve_capability_policy(model, usage_type)
             if usage_type.to_sym == :embedding
-              return { capabilities: %i[embeddings], sources: { embeddings: { value: true, source: :model_metadata } } }
+              return { capabilities: %i[embedding], sources: { embedding: { value: true, source: :model_metadata } } }
             end
 
             real_caps = real_capabilities_for(model)
@@ -420,7 +454,7 @@ module Legion
 
             cfg.except(:instances, 'instances')
           rescue StandardError => e
-            handle_exception(e, level: :debug, handled: true, operation: 'azure_foundry.provider_level_config')
+            handle_exception(e, level: :warn, handled: true, operation: 'azure_foundry.provider_level_config')
             {}
           end
 

data/lib/legion/extensions/llm/azure_foundry/version.rb

@@ -4,7 +4,7 @@ module Legion
   module Extensions
     module Llm
       module AzureFoundry
-        VERSION = '0.2.10'
+        VERSION = '0.2.14'
       end
     end
   end

data/lib/legion/extensions/llm/azure_foundry.rb

@@ -66,7 +66,7 @@ module Legion
                      cfg['api_base']
           return if endpoint.nil? || endpoint.to_s.strip.empty?
 
-          instances[:settings] = normalize_instance_config(cfg).merge(tier: :cloud)
+          instances[:settings] = normalize_instance_config(cfg).merge(tier: cfg[:tier] || cfg['tier'] || :cloud)
         end
 
         def self.discover_named_instances(instances)
@@ -86,7 +86,8 @@ module Legion
                      config[:api_base] || config['api_base']
           return if endpoint.nil? || endpoint.to_s.strip.empty?
 
-          instances[name.to_sym] = normalize_instance_config(config).merge(tier: :cloud)
+          instances[name.to_sym] =
+            normalize_instance_config(config).merge(tier: config[:tier] || config['tier'] || :cloud)
         end
 
         def self.normalize_instance_config(config) # rubocop:disable Metrics/AbcSize, Metrics/CyclomaticComplexity, Metrics/PerceivedComplexity

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: lex-llm-azure-foundry
 version: !ruby/object:Gem::Version
-  version: 0.2.10
+  version: 0.2.14
 platform: ruby
 authors:
 - LegionIO
@@ -71,14 +71,14 @@ dependencies:
     requirements:
     - - ">="
       - !ruby/object:Gem::Version
-        version: 0.5.0
+        version: 0.6.0
   type: :runtime
   prerelease: false
   version_requirements: !ruby/object:Gem::Requirement
     requirements:
     - - ">="
       - !ruby/object:Gem::Version
-        version: 0.5.0
+        version: 0.6.0
 description: Azure AI Foundry and Azure OpenAI hosted provider integration for LegionIO
   LLM routing.
 email: