Merge pull request #2106 from Jdubrick/add-provider-id

RHIDP-15394: add optional id field for inference providers

Author: tisnik

docs/config.html

@@ -2105,7 +2105,11 @@ <h2 id="unifiedinferenceprovider">UnifiedInferenceProvider</h2>
 values).</p>
     <p>Attributes: type: Canonical provider identifier. Vendor-neutral so it
 survives a future backend change; each backend-specific synthesizer maps
-it to its own provider vocabulary. api_key_env: Name of the environment
+it to its own provider vocabulary. id: Optional identifier emitted as
+the Llama Stack provider_id. When omitted, synthesized as type with
+underscores hyphenated. If set, must be non-empty after stripping
+whitespace and may contain only lowercase letters, digits, underscores,
+and hyphens. api_key_env: Name of the environment
 variable holding the provider API key. Emitted verbatim as
 <code>${env.&lt;name&gt;}</code> so the secret never lands on disk
 resolved. allowed_models: Optional allow-list of model identifiers
@@ -2134,18 +2138,26 @@ <h2 id="unifiedinferenceprovider">UnifiedInferenceProvider</h2>
 Stack provider_type by the synthesizer.</td>
         </tr>
         <tr class="even">
+          <td>id</td>
+          <td>string</td>
+          <td>Optional identifier emitted as the Llama Stack provider_id. When
+omitted, synthesized as type with underscores hyphenated. If set, must
+be non-empty after stripping whitespace and may contain only lowercase
+letters, digits, underscores, and hyphens.</td>
+        </tr>
+        <tr class="odd">
           <td>api_key_env</td>
           <td>string</td>
           <td>Name of the environment variable holding the provider API key.
 Emitted as a ${env.name} reference so the secret is never written to
 disk in resolved form.</td>
         </tr>
-        <tr class="odd">
+        <tr class="even">
           <td>allowed_models</td>
           <td>array</td>
           <td>Optional allow-list of model identifiers for this provider.</td>
         </tr>
-        <tr class="even">
+        <tr class="odd">
           <td>extra</td>
           <td>object</td>
           <td>Additional provider-config keys merged verbatim into the synthesized

docs/config.json

@@ -458,7 +458,7 @@
       },
       "CompactionConfiguration": {
         "additionalProperties": false,
-        "description": "Configuration for conversation history compaction.\n\nCompaction summarizes older conversation turns when their estimated\ntoken count approaches the context window limit, keeping the\nconversation usable instead of failing with HTTP 413. The\nconfiguration here controls when compaction triggers and how much\nrecent context is preserved verbatim.\n\nAttributes:\n    enabled: Master switch. When False, compaction never triggers\n        and other fields are inert.\n    threshold_ratio: Trigger compaction when estimated input tokens\n        exceed this fraction of the model's context window\n        (clamped to 0.0..1.0).\n    token_floor: Minimum estimated token count before compaction\n        can trigger, regardless of threshold_ratio. Prevents\n        triggering on very small context windows.\n    buffer_turns: Initial number of recent turns to keep verbatim.\n        The runtime applies a degrading guard — if these turns\n        exceed the available budget, it reduces buffer_turns by\n        one repeatedly until the budget fits, down to zero.\n    buffer_max_ratio: Hard cap on the fraction of the context\n        window the buffer zone may occupy, regardless of\n        buffer_turns.",
+        "description": "Configuration for conversation history compaction.\n\nCompaction summarizes older conversation turns when their estimated\ntoken count approaches the context window limit, keeping the\nconversation usable instead of failing with HTTP 413. The\nconfiguration here controls when compaction triggers and how much\nrecent context is preserved verbatim.\n\nAttributes:\n    enabled: Master switch. When False, compaction never triggers\n        and other fields are inert.\n    threshold_ratio: Trigger compaction when estimated input tokens\n        exceed this fraction of the model's context window\n        (clamped to 0.0..1.0).\n    token_floor: Minimum estimated token count before compaction\n        can trigger, regardless of threshold_ratio. Prevents\n        triggering on very small context windows.\n    buffer_turns: Initial number of recent turns to keep verbatim.\n        The runtime applies a degrading guard \u2014 if these turns\n        exceed the available budget, it reduces buffer_turns by\n        one repeatedly until the budget fits, down to zero.\n    buffer_max_ratio: Hard cap on the fraction of the context\n        window the buffer zone may occupy, regardless of\n        buffer_turns.",
         "properties": {
           "enabled": {
             "default": false,
@@ -575,7 +575,7 @@
           },
           "compaction": {
             "$ref": "#/components/schemas/CompactionConfiguration",
-            "description": "Controls when conversation history is summarized to keep the model's input below the context window limit. Disabled by default — when disabled, requests that exceed the window continue to surface as HTTP 413.",
+            "description": "Controls when conversation history is summarized to keep the model's input below the context window limit. Disabled by default \u2014 when disabled, requests that exceed the window continue to surface as HTTP 413.",
             "title": "Conversation compaction configuration"
           },
           "approvals": {
@@ -885,7 +885,7 @@
               "minimum": 0,
               "type": "integer"
             },
-            "description": "Map of fully-qualified model identifier (e.g., \"openai/gpt-4o-mini\") to context window size in tokens. Used by the conversation compaction trigger to decide when older turns must be summarized before the input exceeds the window. Models absent from this map have no registered window — callers fall back to their own default or skip the token-based trigger.",
+            "description": "Map of fully-qualified model identifier (e.g., \"openai/gpt-4o-mini\") to context window size in tokens. Used by the conversation compaction trigger to decide when older turns must be summarized before the input exceeds the window. Models absent from this map have no registered window \u2014 callers fall back to their own default or skip the token-based trigger.",
             "title": "Per-model context window sizes (tokens)",
             "type": "object"
           },
@@ -1719,7 +1719,7 @@
       },
       "UnifiedInferenceProvider": {
         "additionalProperties": false,
-        "description": "A high-level inference provider entry for unified-mode synthesis.\n\nOperators describe inference providers at this high level (backend-agnostic\nvocabulary) instead of authoring raw Llama Stack provider blocks. The\nsynthesizer (`apply_high_level_inference`) expands each entry into a Llama\nStack `providers.inference` entry, mapping `type` to a `provider_type` and\nemitting `${env.<VAR>}` references for secrets (never literal values).\n\nAttributes:\n    type: Canonical provider identifier. Vendor-neutral so it survives a\n        future backend change; each backend-specific synthesizer maps it to\n        its own provider vocabulary.\n    api_key_env: Name of the environment variable holding the provider API\n        key. Emitted verbatim as `${env.<name>}` so the secret never lands\n        on disk resolved.\n    allowed_models: Optional allow-list of model identifiers passed through\n        to the synthesized provider config.\n    extra: Additional provider-config keys merged verbatim into the\n        synthesized provider's `config` block — an escape hatch for\n        provider-specific knobs not modeled here.",
+        "description": "A high-level inference provider entry for unified-mode synthesis.\n\nOperators describe inference providers at this high level (backend-agnostic\nvocabulary) instead of authoring raw Llama Stack provider blocks. The\nsynthesizer (`apply_high_level_inference`) expands each entry into a Llama\nStack `providers.inference` entry, mapping `type` to a `provider_type` and\nemitting `${env.<VAR>}` references for secrets (never literal values).\n\nAttributes:\n    type: Canonical provider identifier. Vendor-neutral so it survives a\n        future backend change; each backend-specific synthesizer maps it to\n        its own provider vocabulary.\n    id: Optional identifier emitted as the Llama Stack provider_id. When\n        omitted, synthesized as type with underscores hyphenated. If set,\n        must be non-empty after stripping whitespace and may contain only\n        lowercase letters, digits, underscores, and hyphens.\n    api_key_env: Name of the environment variable holding the provider API\n        key. Emitted verbatim as `${env.<name>}` so the secret never lands\n        on disk resolved.\n    allowed_models: Optional allow-list of model identifiers passed through\n        to the synthesized provider config.\n    extra: Additional provider-config keys merged verbatim into the\n        synthesized provider's `config` block \u2014 an escape hatch for\n        provider-specific knobs not modeled here.",
         "properties": {
           "type": {
             "description": "Canonical, backend-agnostic provider identifier mapped to a Llama Stack provider_type by the synthesizer.",
@@ -1737,6 +1737,13 @@
             "title": "Provider type",
             "type": "string"
           },
+          "id": {
+            "type": "string",
+            "nullable": true,
+            "default": null,
+            "description": "Optional identifier emitted as the Llama Stack provider_id. When omitted, synthesized as type with underscores hyphenated. If set, must be non-empty after stripping whitespace and may contain only lowercase letters, digits, underscores, and hyphens.",
+            "title": "Provider ID"
+          },
           "api_key_env": {
             "type": "string",
             "nullable": true,

docs/config.md

@@ -794,6 +794,10 @@ Attributes:
     type: Canonical provider identifier. Vendor-neutral so it survives a
         future backend change; each backend-specific synthesizer maps it to
         its own provider vocabulary.
+    id: Optional identifier emitted as the Llama Stack provider_id. When
+        omitted, synthesized as type with underscores hyphenated. If set,
+        must be non-empty after stripping whitespace and may contain only
+        lowercase letters, digits, underscores, and hyphens.
     api_key_env: Name of the environment variable holding the provider API
         key. Emitted verbatim as `${env.<name>}` so the secret never lands
         on disk resolved.
@@ -807,6 +811,7 @@ Attributes:
 | Field          | Type   | Description                                                                                                                                                  |
 |----------------|--------|--------------------------------------------------------------------------------------------------------------------------------------------------------------|
 | type           | string | Canonical, backend-agnostic provider identifier mapped to a Llama Stack provider_type by the synthesizer.                                                    |
+| id             | string | Optional identifier emitted as the Llama Stack provider_id. When omitted, synthesized as type with underscores hyphenated. If set, must be non-empty after stripping whitespace and may contain only lowercase letters, digits, underscores, and hyphens. |
 | api_key_env    | string | Name of the environment variable holding the provider API key. Emitted as a ${env.<name>} reference so the secret is never written to disk in resolved form. |
 | allowed_models | array  | Optional allow-list of model identifiers for this provider.                                                                                                  |
 | extra          | object | Additional provider-config keys merged verbatim into the synthesized provider's config block.                                                                |

docs/config.puml

@@ -303,7 +303,8 @@ class "UnifiedInferenceProvider" as src.models.config.UnifiedInferenceProvider {
   allowed_models : Optional[list[str]]
   api_key_env : Optional[str]
   extra : dict[str, object]
-  type : Literal['openai', 'sentence_transformers', 'azure', 'vertexai', 'watsonx', 'vllm_rhaiis', 'vllm_rhel_ai']
+  id : str | None
+  type : Literal['openai', 'ollama', 'vllm', 'sentence_transformers', 'azure', 'vertexai', 'watsonx', 'vllm_rhaiis', 'vllm_rhel_ai']
 }
 class "UnifiedLlamaStackConfig" as src.models.config.UnifiedLlamaStackConfig {
   baseline : Literal['default', 'empty']

docs/openapi.json

@@ -20355,6 +20355,18 @@
                         "title": "Provider type",
                         "description": "Canonical, backend-agnostic provider identifier mapped to a Llama Stack provider_type by the synthesizer."
                     },
+                    "id": {
+                        "anyOf": [
+                            {
+                                "type": "string"
+                            },
+                            {
+                                "type": "null"
+                            }
+                        ],
+                        "title": "Provider ID",
+                        "description": "Optional identifier emitted as the Llama Stack provider_id. When omitted, synthesized as type with underscores hyphenated. If set, must be non-empty after stripping whitespace and may contain only lowercase letters, digits, underscores, and hyphens."
+                    },
                     "api_key_env": {
                         "anyOf": [
                             {
@@ -20395,7 +20407,7 @@
                     "type"
                 ],
                 "title": "UnifiedInferenceProvider",
-                "description": "A high-level inference provider entry for unified-mode synthesis.\n\nOperators describe inference providers at this high level (backend-agnostic\nvocabulary) instead of authoring raw Llama Stack provider blocks. The\nsynthesizer (`apply_high_level_inference`) expands each entry into a Llama\nStack `providers.inference` entry, mapping `type` to a `provider_type` and\nemitting `${env.<VAR>}` references for secrets (never literal values).\n\nAttributes:\n    type: Canonical provider identifier. Vendor-neutral so it survives a\n        future backend change; each backend-specific synthesizer maps it to\n        its own provider vocabulary.\n    api_key_env: Name of the environment variable holding the provider API\n        key. Emitted verbatim as `${env.<name>}` so the secret never lands\n        on disk resolved.\n    allowed_models: Optional allow-list of model identifiers passed through\n        to the synthesized provider config.\n    extra: Additional provider-config keys merged verbatim into the\n        synthesized provider's `config` block \u2014 an escape hatch for\n        provider-specific knobs not modeled here."
+                "description": "A high-level inference provider entry for unified-mode synthesis.\n\nOperators describe inference providers at this high level (backend-agnostic\nvocabulary) instead of authoring raw Llama Stack provider blocks. The\nsynthesizer (`apply_high_level_inference`) expands each entry into a Llama\nStack `providers.inference` entry, mapping `type` to a `provider_type` and\nemitting `${env.<VAR>}` references for secrets (never literal values).\n\nAttributes:\n    type: Canonical provider identifier. Vendor-neutral so it survives a\n        future backend change; each backend-specific synthesizer maps it to\n        its own provider vocabulary.\n    id: Optional identifier emitted as the Llama Stack provider_id. When\n        omitted, synthesized as type with underscores hyphenated. If set,\n        must be non-empty after stripping whitespace and may contain only\n        lowercase letters, digits, underscores, and hyphens.\n    api_key_env: Name of the environment variable holding the provider API\n        key. Emitted verbatim as `${env.<name>}` so the secret never lands\n        on disk resolved.\n    allowed_models: Optional allow-list of model identifiers passed through\n        to the synthesized provider config.\n    extra: Additional provider-config keys merged verbatim into the\n        synthesized provider's `config` block \u2014 an escape hatch for\n        provider-specific knobs not modeled here."
             },
             "UnifiedLlamaStackConfig": {
                 "properties": {

examples/profiles/inline-faiss.yaml

@@ -11,6 +11,13 @@
 #     providers:
 #       - type: openai
 #         api_key_env: OPENAI_API_KEY
+#       # Multiple backends of the same type need distinct id values:
+#       # - type: vllm
+#       #   id: vllm-prod
+#       #   api_key_env: VLLM_PROD_KEY
+#       # - type: vllm
+#       #   id: vllm-staging
+#       #   api_key_env: VLLM_STAGING_KEY
 #   llama_stack:
 #     use_as_library_client: true
 #     config:

src/client.py

@@ -64,6 +64,12 @@ async def _load_library_client(self, config: LlamaStackConfiguration) -> None:
         """
         logger.info("Using Llama stack as library client")
 
+        # Configure logging before synthesis/enrichment so INFO lines from those
+        # steps are not dropped. Without handlers, Python's lastResort only
+        # emits WARNING and above — which is why replace logs disappeared after
+        # moving from warning to info.
+        setup_logging()
+
         if config.library_client_config_path is not None and config.config is None:
             self._config_path = self._enrich_library_config(
                 config.library_client_config_path

src/llama_stack_configuration.py

@@ -718,12 +718,14 @@ def apply_high_level_inference(
 
     Each high-level provider is mapped to a Llama Stack ``providers.inference``
     entry via :data:`PROVIDER_TYPE_MAP`. The emitted ``provider_id`` is the
-    provider ``type`` with underscores hyphenated, so an inline embedder declared
-    as ``sentence_transformers`` becomes ``sentence-transformers`` and matches the
+    optional explicit high-level ``id`` when set; otherwise the provider ``type``
+    with underscores hyphenated, so an inline embedder declared as
+    ``sentence_transformers`` becomes ``sentence-transformers`` and matches the
     baseline's ecosystem convention (e.g. the default embedding model reference).
-    An entry whose ``provider_id`` already exists in the baseline is replaced; new
-    ones are appended. Secrets are emitted as ``${env.<VAR>}`` references, never
-    resolved values (R6).
+    An entry whose ``provider_id`` already exists in the baseline (or was emitted
+    by an earlier high-level entry) is replaced with an info log; new ones are
+    appended. Secrets are emitted as ``${env.<VAR>}`` references, never resolved
+    values (R6).
 
     Parameters:
         ls_config: The Llama Stack configuration being synthesized (modified in
@@ -743,7 +745,7 @@ def apply_high_level_inference(
 
     for provider in providers:
         provider_type = provider["type"]
-        emitted_id = provider_type.replace("_", "-")
+        emitted_id = provider.get("id") or provider_type.replace("_", "-")
         ls_provider_type = PROVIDER_TYPE_MAP[provider_type]
         entry: dict[str, Any] = {
             "provider_id": emitted_id,
@@ -764,6 +766,11 @@ def apply_high_level_inference(
         # Replace a baseline provider with the same id, else append.
         for index, existing in enumerate(inference_list):
             if isinstance(existing, dict) and existing.get("provider_id") == emitted_id:
+                logger.info(
+                    "Replacing existing inference provider with "
+                    "provider_id=%r; a later high-level entry overwrote it",
+                    emitted_id,
+                )
                 inference_list[index] = entry
                 break
         else: