lex-llm-azure-foundry 0.2.17 → 0.3.5

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 433a2650edf79d39694f77ff2ebdabe3e97aa4e6c4ed5aa23935290748789e04
-  data.tar.gz: d1953642f8ea39987493f634488dc8480be56642e60369c6696ac2c0213ac7c3
+  metadata.gz: 49e5d8dc9a03ec7e353324aaf22050bf23a28ff085da22a8f779c7b01b28ffe0
+  data.tar.gz: 3c8c342a9bc24b2133a2060f9cc9dc9a9743d4d69bda1205e26db02f6d5a4f59
 SHA512:
-  metadata.gz: ddf30f8e148302d3fc8c897dbc2b55c7df6b93b507860cc20ff4f6a8f4bd47a30df502e5aa9ae95832e4755e3331ebc54673085c2ba44dd3e0ca124d091315f2
-  data.tar.gz: 3187ac2e0b1e0fbddcd9fa31ba85fbc55219bc275a14220861e488940657fa8111a03df0ee773520110bd2bb6a18626a8125a98334c3746e0258fd868e3fc879
+  metadata.gz: ba398b8cdb7dcd17fa6138c9edccc482b66abb201c8ce198d4373d83ced0de5f8c1c756be4b55ec3b3a64921278db4d3b6673d146a77ed2121d6e7317622be79
+  data.tar.gz: d1c2877ae02270b74dba8be1daa23c9c676070c4d705194fa35933d31206d3c9bd5eccb0d2837c80ffe2fe278153766f0865d1e53d077cd4b9e84ce67c5fdef0

data/.rubocop.yml

@@ -16,3 +16,11 @@ Metrics/MethodLength:
   Enabled: false
 RSpec/MultipleExpectations:
   Enabled: false
+RSpec/ExampleLength:
+  Enabled: false
+RSpec/MultipleMemoizedHelpers:
+  Enabled: false
+RSpec/DescribeClass:
+  Exclude:
+    - spec/legion/extensions/llm/azure_foundry/capability_policy_spec.rb
+    - spec/legion/extensions/llm/azure_foundry/context_window_spec.rb

data/CHANGELOG.md

@@ -1,5 +1,100 @@
 # Changelog
 
+## [0.3.5] - 2026-08-19
+
+### Changed
+- **Write-time SSOT lane weights** — Build Azure deployment drafts with the shared
+  four-component weight pair using the deployment name as the provider-native
+  offering identity, and reconcile weight-only changes atomically on the existing
+  discovery cadence. Initial/recovery activation now rebuilds from current settings,
+  removal wins readiness races, and dormant configured weights log once per absence
+  period without any Settings lifecycle callback.
+- **Callable-path system conformance** — Capture the rendered Azure/OpenAI-compatible
+  HTTP payload and verify a folded leading system message remains in the dialect-native
+  `role: system` message position.
+- **Dependency floor** — Raise `lex-llm` to `>= 0.7.6` for `WeightSchema`,
+  `WeightReconciler`, and the immutable record weight pair. The `legion-settings`
+  dependency and provider/operator workflow are unchanged.
+
+### Fixed
+- **Malformed-weight startup cleanup** — Validate and build weighted offering
+  drafts before constructing or claiming any callable scope, so invalid weight
+  values cannot leave an orphaned initializing Registry publication. A later
+  corrected discovery pass claims and activates the instance normally without
+  requiring a restart or operator cleanup.
+- **Complete offering replacement comparison** — Ordinary discovery now compares
+  every authoritative `OfferingDraft` field while ignoring only evidence
+  observation timestamps. Deployment order does not churn, duplicate counts stay
+  significant, and native-key, evidence, quota, metadata, publication-source,
+  tier, or weight drift advances the Registry snapshot exactly once.
+
+## [0.3.4] - 2026-08-18
+
+### Fixed
+- **Default instance labels** — Accept `default` as an ordinary operator
+  `InstanceKey` label, matching the shared `lex-llm` identity contract.
+
+## [0.3.3] - 2026-08-16
+
+### Changed
+- **Instance identity is the operator's config name** — the discovery actor now
+  publishes `InstanceKey.instance_id` as the config NAME (the key the router
+  resolves `instances.<name>` settings — per-instance tuning, `enable_*`
+  overrides — by), restoring the baseline semantics. The previously-derived
+  `host:port[/ak:fingerprint]` id is kept as the **secondary** `InstanceKey`
+  `physical_id` field (dedup/diagnostics only; it never participates in
+  equality, hashing, or registry-scope identity). Two config names pointing at
+  the same endpoint stay distinct instances; the credential fingerprint still
+  distinguishes same-name configs in diagnostics. `physical_id` is threaded
+  through every `Inventory::Publisher` lifecycle call and onto offering
+  metadata.
+- **Authoritative operation evidence for embedding deployments** — an embedding
+  deployment now publishes `chat: :unsupported` and `stream_chat: :unsupported`
+  (with `embed: :supported`), matching how bedrock authoritatively excludes
+  embedding models, so a plain chat request can no longer misroute to an
+  embedding-only deployment.
+- **Dependency floor** — `lex-llm` raised to `>= 0.7.1` (the
+  `InstanceKey#physical_id` secondary field).
+
+### Fixed
+- **Single actor registration** — the provider module no longer extends `::Legion::Extensions::Core` at file level, so the boot-time submodule walk skips it and the gem's own top-level extension load is the sole actor registration — eliminating the double-claim / `FencedPublisherError` twin.
+
+## [0.3.2] - 2026-08-13
+
+### Fixed
+- **§1 — zero rubocop:disable** — Removed every `rubocop:disable` / `rubocop:enable` inline directive from `lib/` and `spec/`. Refactored `azure_foundry.rb`, `provider.rb`, and the fleet-worker spec to comply without suppression: extracted `ProviderClassMethods`, `ProviderDispatchMethods`, `ProviderOfferingHelpers`, `ProviderOfferingMetadata`, `ProviderCapabilityHelpers`, `Capabilities` modules and new private helpers (`extract_endpoint_from_cfg`, `extract_tier_from_cfg`, `promote_endpoint_aliases`, `promote_provider_aliases`, `resolve_usage_type`, `resolve_model_family`, `build_offering_record`, `build_offering_metadata`).
+- **§9 — :default substitution removed** — Deleted `build_default_instance` and `default_instance_config` from `InstanceConfigHelpers`; `configured_instances` now reads `settings[:instances]` directly and skips entries with nil/empty endpoints. Removed `settings[:endpoint] || settings[:azure_foundry_endpoint]` `||` guard.
+- **§1 — swallowed rescue** — `extract_host_port` in `InstanceIdentityHelpers` now calls `handle_exception` and re-raises on `URI::InvalidURIError` instead of silently returning `'unknown:0'`.
+- **§1 — credentials .dig removed** — `apply_auth_headers` and `derive_instance_id` in the discovery actor now read only from `instance_cfg[:azure_foundry_api_key]`; the `instance_cfg.dig(:credentials, :api_key)` fallback was removed (all configs pass through `normalize_instance_config` which promotes this key).
+- **§6 — circuit_state vocabulary** — Removed `circuit_state:` field from `Provider#health_baseline` and from all spec assertions; availability is probe-cleared only, not modelled as a circuit breaker.
+- **spec path** — Renamed `spec/…/actors/fleet_worker_spec.rb` to `spec/…/actor/fleet_worker_spec.rb` to match the `Actor::FleetWorker` module path and satisfy `RSpec/SpecFilePathFormat`.
+- **INSTANCE_DEFAULTS constant** — Extracted default instance settings hash to `AzureFoundry::INSTANCE_DEFAULTS` to reduce `default_settings` line count below the `Metrics/ModuleLength` limit.
+
+## [0.3.1] - 2026-08-13
+
+### Fixed
+- **§1 / §8 compliance sweep** — Removed all `rubocop:disable` comments. Extracted helper modules (`HealthCheckHelpers`, `OfferingBuilderHelpers`, `InstanceConfigHelpers`, `InstanceIdentityHelpers`, `ProbeHelpers`, `ShutdownHelpers`) from `DiscoveryRefresh` to bring `Metrics/ClassLength` and `Metrics/AbcSize` into conformance.
+- **§8 health firewall** — `AzureFoundryCallable#classify_server_error` now maps 503 to `:instance_unavailable` ONLY when the Azure `EndpointDeactivated` error code appears in the response body; all other 5xx (including plain 503 overload) remain `:overloaded` or `:provider_error`. `Faraday::ConnectionFailed` and `Faraday::TimeoutError` remain request-local and never promote the instance to unavailable.
+- **§9 / settings guards** — Removed `def settings` override in `Provider` that called `.dig(:instances, :default)`. Removed `settings[:discovery_interval] || ...` and all `settings.dig(...)` patterns in `DiscoveryRefresh`; replaced with direct canonical path access (`settings[:discovery][:interval_seconds]`, `settings[:instances][:default][...]`).
+- **§2 / single publication engine** — Removed `registry_publisher.publish_readiness_async` call from `Provider#readiness` and `registry_publisher.publish_models_async` call from `Provider#list_models`; publication is now exclusively through `Inventory::Publisher` in the discovery actor.
+- Vision capability evidence in `OfferingBuilderHelpers` now emits `:unknown` / `:default_false` instead of inferring support from model name regex (which is not authoritative proof).
+- Discovery interval default overridden to 3600 s via `provider_settings(discovery: { interval_seconds: 3600 })` to match the previous actor behaviour.
+
+## [0.3.0] - 2026-08-13
+
+### Changed
+- **SSOT v3 migration** — Rewrote `DiscoveryRefresh` actor to use `Inventory::Publisher` for instance lifecycle (claim/activate/replace/remove) instead of the legacy `ScopedRefresher` mixin and `Legion::LLM::Call::Registry`.
+- Added `AzureFoundryCallable` class implementing `disconnect`, `disconnected?`, `normalize_dispatch_error(error:)`, and dispatch operations (`chat`, `stream_chat`, `embed`, `count_tokens`).
+- Fleet worker now passes `registry: Legion::Extensions::Llm::Inventory::Registry` for exact fleet execution.
+- Instance identity derived from normalized endpoint host:port plus API key fingerprint (first 6 hex of SHA256).
+- Offerings publish deployment-scoped quota domains (`azure:deployment:<name>`).
+- Operation evidence covers all `Taxonomies::OPERATIONS`: chat/stream_chat supported, embed conditional on deployment type, count_tokens/image/transcribe/translate/speak/moderate unsupported.
+- Bumped `lex-llm` dependency floor to `>= 0.7.0`.
+
+### Removed
+- Removed `ScopedRefresher` include and all references to `Legion::LLM::Call::Registry`.
+- Removed legacy `compute_lanes_for_scope`, `scope_key`, `credential_hash` methods.
+
 ## [0.2.17] - 2026-08-07
 
 ### Fixed

data/Gemfile

@@ -2,6 +2,8 @@
 
 source 'https://rubygems.org'
 
+gem 'lex-llm'
+
 gemspec
 
 group :development do

data/lex-llm-azure-foundry.gemspec

@@ -25,7 +25,7 @@ Gem::Specification.new do |spec|
 
   spec.add_dependency 'legion-json', '>= 1.2.1'
   spec.add_dependency 'legion-logging', '>= 1.3.2'
-  spec.add_dependency 'legion-settings', '>= 1.3.14'
+  spec.add_dependency 'legion-settings', '>= 1.4.2'
   spec.add_dependency 'legion-transport', '>= 1.4.14'
-  spec.add_dependency 'lex-llm', '>= 0.6.0'
+  spec.add_dependency 'lex-llm', '>= 0.7.6'
 end