arize-phoenix 8.32.1__tar.gz → 9.0.1__tar.gz

{arize_phoenix-8.32.1 → arize_phoenix-9.0.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: arize-phoenix
-Version: 8.32.1
+Version: 9.0.1
 Summary: AI Observability and Evaluation
 Project-URL: Documentation, https://docs.arize.com/phoenix/
 Project-URL: Issues, https://github.com/Arize-ai/phoenix/issues
@@ -49,7 +49,7 @@ Requires-Dist: scipy
 Requires-Dist: sqlalchemy[asyncio]<3,>=2.0.4
 Requires-Dist: sqlean-py>=3.45.1
 Requires-Dist: starlette>=0.46.0
-Requires-Dist: strawberry-graphql>=0.262.0
+Requires-Dist: strawberry-graphql==0.267.0
 Requires-Dist: tqdm
 Requires-Dist: typing-extensions>=4.6
 Requires-Dist: uvicorn
@@ -71,7 +71,7 @@ Requires-Dist: opentelemetry-sdk; extra == 'container'
 Requires-Dist: opentelemetry-semantic-conventions; extra == 'container'
 Requires-Dist: prometheus-client; extra == 'container'
 Requires-Dist: py-grpc-prometheus; extra == 'container'
-Requires-Dist: strawberry-graphql[opentelemetry]==0.262.5; extra == 'container'
+Requires-Dist: strawberry-graphql[opentelemetry]==0.267.0; extra == 'container'
 Requires-Dist: umap-learn; extra == 'container'
 Requires-Dist: uvloop; (platform_system != 'Windows') and extra == 'container'
 Provides-Extra: dev
@@ -102,7 +102,7 @@ Requires-Dist: pytest-postgresql; extra == 'dev'
 Requires-Dist: pytest-xdist; extra == 'dev'
 Requires-Dist: pytest==8.3.3; extra == 'dev'
 Requires-Dist: ruff==0.6.9; extra == 'dev'
-Requires-Dist: strawberry-graphql[debug-server,opentelemetry]==0.262.5; extra == 'dev'
+Requires-Dist: strawberry-graphql[debug-server,opentelemetry]==0.267.0; extra == 'dev'
 Requires-Dist: tabulate; extra == 'dev'
 Requires-Dist: tox-uv==1.11.3; extra == 'dev'
 Requires-Dist: tox==4.18.1; extra == 'dev'
@@ -172,7 +172,7 @@ Phoenix is an open-source AI observability platform designed for experimentation
 - [**_Playground_**](https://docs.arize.com/phoenix/prompt-engineering/overview-prompts)- Optimize prompts, compare models, adjust parameters, and replay traced LLM calls.
 - [**_Prompt Management_**](https://docs.arize.com/phoenix/prompt-engineering/overview-prompts/prompt-management)- Manage and test prompt changes systematically using version control, tagging, and experimentation.
 
-Phoenix is vendor and language agnostic with out-of-the-box support for popular frameworks (🦙[LlamaIndex](https://docs.arize.com/phoenix/tracing/integrations-tracing/llamaindex), 🦜⛓[LangChain](https://docs.arize.com/phoenix/tracing/integrations-tracing/langchain), [Haystack](https://docs.arize.com/phoenix/tracing/integrations-tracing/haystack), 🧩[DSPy](https://docs.arize.com/phoenix/tracing/integrations-tracing/dspy), 🤗[smolagents](https://docs.arize.com/phoenix/tracing/integrations-tracing/hfsmolagents)) and LLM providers ([OpenAI](https://docs.arize.com/phoenix/tracing/integrations-tracing/openai), [Bedrock](https://docs.arize.com/phoenix/tracing/integrations-tracing/bedrock), [MistralAI](https://docs.arize.com/phoenix/tracing/integrations-tracing/mistralai), [VertexAI](https://docs.arize.com/phoenix/tracing/integrations-tracing/vertexai), [LiteLLM](https://docs.arize.com/phoenix/tracing/integrations-tracing/litellm), and more). For details on auto-instrumentation, check out the [OpenInference](https://github.com/Arize-ai/openinference) project.
+Phoenix is vendor and language agnostic with out-of-the-box support for popular frameworks (🦙[LlamaIndex](https://docs.arize.com/phoenix/tracing/integrations-tracing/llamaindex), 🦜⛓[LangChain](https://docs.arize.com/phoenix/tracing/integrations-tracing/langchain), [Haystack](https://docs.arize.com/phoenix/tracing/integrations-tracing/haystack), 🧩[DSPy](https://docs.arize.com/phoenix/tracing/integrations-tracing/dspy), 🤗[smolagents](https://docs.arize.com/phoenix/tracing/integrations-tracing/hfsmolagents)) and LLM providers ([OpenAI](https://docs.arize.com/phoenix/tracing/integrations-tracing/openai), [Bedrock](https://docs.arize.com/phoenix/tracing/integrations-tracing/bedrock), [MistralAI](https://docs.arize.com/phoenix/tracing/integrations-tracing/mistralai), [VertexAI](https://docs.arize.com/phoenix/tracing/integrations-tracing/vertexai), [LiteLLM](https://docs.arize.com/phoenix/tracing/integrations-tracing/litellm), [Google GenAI](https://docs.arize.com/phoenix/tracing/integrations-tracing/google-genai) and more). For details on auto-instrumentation, check out the [OpenInference](https://github.com/Arize-ai/openinference) project.
 
 Phoenix runs practically anywhere, including your local machine, a Jupyter notebook, a containerized deployment, or in the cloud.
 

{arize_phoenix-8.32.1 → arize_phoenix-9.0.1}/README.md

@@ -42,7 +42,7 @@ Phoenix is an open-source AI observability platform designed for experimentation
 - [**_Playground_**](https://docs.arize.com/phoenix/prompt-engineering/overview-prompts)- Optimize prompts, compare models, adjust parameters, and replay traced LLM calls.
 - [**_Prompt Management_**](https://docs.arize.com/phoenix/prompt-engineering/overview-prompts/prompt-management)- Manage and test prompt changes systematically using version control, tagging, and experimentation.
 
-Phoenix is vendor and language agnostic with out-of-the-box support for popular frameworks (🦙[LlamaIndex](https://docs.arize.com/phoenix/tracing/integrations-tracing/llamaindex), 🦜⛓[LangChain](https://docs.arize.com/phoenix/tracing/integrations-tracing/langchain), [Haystack](https://docs.arize.com/phoenix/tracing/integrations-tracing/haystack), 🧩[DSPy](https://docs.arize.com/phoenix/tracing/integrations-tracing/dspy), 🤗[smolagents](https://docs.arize.com/phoenix/tracing/integrations-tracing/hfsmolagents)) and LLM providers ([OpenAI](https://docs.arize.com/phoenix/tracing/integrations-tracing/openai), [Bedrock](https://docs.arize.com/phoenix/tracing/integrations-tracing/bedrock), [MistralAI](https://docs.arize.com/phoenix/tracing/integrations-tracing/mistralai), [VertexAI](https://docs.arize.com/phoenix/tracing/integrations-tracing/vertexai), [LiteLLM](https://docs.arize.com/phoenix/tracing/integrations-tracing/litellm), and more). For details on auto-instrumentation, check out the [OpenInference](https://github.com/Arize-ai/openinference) project.
+Phoenix is vendor and language agnostic with out-of-the-box support for popular frameworks (🦙[LlamaIndex](https://docs.arize.com/phoenix/tracing/integrations-tracing/llamaindex), 🦜⛓[LangChain](https://docs.arize.com/phoenix/tracing/integrations-tracing/langchain), [Haystack](https://docs.arize.com/phoenix/tracing/integrations-tracing/haystack), 🧩[DSPy](https://docs.arize.com/phoenix/tracing/integrations-tracing/dspy), 🤗[smolagents](https://docs.arize.com/phoenix/tracing/integrations-tracing/hfsmolagents)) and LLM providers ([OpenAI](https://docs.arize.com/phoenix/tracing/integrations-tracing/openai), [Bedrock](https://docs.arize.com/phoenix/tracing/integrations-tracing/bedrock), [MistralAI](https://docs.arize.com/phoenix/tracing/integrations-tracing/mistralai), [VertexAI](https://docs.arize.com/phoenix/tracing/integrations-tracing/vertexai), [LiteLLM](https://docs.arize.com/phoenix/tracing/integrations-tracing/litellm), [Google GenAI](https://docs.arize.com/phoenix/tracing/integrations-tracing/google-genai) and more). For details on auto-instrumentation, check out the [OpenInference](https://github.com/Arize-ai/openinference) project.
 
 Phoenix runs practically anywhere, including your local machine, a Jupyter notebook, a containerized deployment, or in the cloud.
 

{arize_phoenix-8.32.1 → arize_phoenix-9.0.1}/pyproject.toml

@@ -29,7 +29,7 @@ dependencies = [
   "starlette>=0.46.0",
   "uvicorn",
   "psutil",
-  "strawberry-graphql>=0.262.0",  # resolve security vulnerability https://github.com/Arize-ai/phoenix/security/dependabot/45
+  "strawberry-graphql==0.267.0",  # avoid this issue: https://github.com/strawberry-graphql/strawberry/issues/3854
   "pyarrow",
   "typing-extensions>=4.6",
   "scipy",
@@ -88,7 +88,7 @@ dev = [
   "pytest-postgresql",
   "asyncpg",
   "psycopg[binary,pool]",
-  "strawberry-graphql[debug-server,opentelemetry]==0.262.5",  # need to pin version because we're monkey-patching
+  "strawberry-graphql[debug-server,opentelemetry]==0.267.0",  # avoid this issue: https://github.com/strawberry-graphql/strawberry/issues/3854
   "pre-commit",
   "arize[AutoEmbeddings, LLM_Evaluation]",
   "llama-index>=0.10.3",
@@ -137,7 +137,7 @@ container = [
   "opentelemetry-instrumentation-sqlalchemy",
   "opentelemetry-instrumentation-grpc",
   "py-grpc-prometheus",
-  "strawberry-graphql[opentelemetry]==0.262.5",  # need to pin version because we're monkey-patching
+  "strawberry-graphql[opentelemetry]==0.267.0",  # avoid this issue: https://github.com/strawberry-graphql/strawberry/issues/3854
   "uvloop; platform_system != 'Windows'",
   "fast-hdbscan>=0.2.0",
   "numba>=0.60.0",  # https://github.com/astral-sh/uv/issues/6281

arize_phoenix-9.0.1/src/phoenix/db/constants.py

@@ -0,0 +1 @@
+DEFAULT_PROJECT_TRACE_RETENTION_POLICY_ID = 0

{arize_phoenix-8.32.1 → arize_phoenix-9.0.1}/src/phoenix/db/facilitator.py

@@ -9,6 +9,7 @@ from typing import Optional
 
 from sqlalchemy import (
     distinct,
+    exists,
     insert,
     select,
 )
@@ -27,7 +28,13 @@ from phoenix.config import (
     get_env_default_admin_initial_password,
 )
 from phoenix.db import models
+from phoenix.db.constants import DEFAULT_PROJECT_TRACE_RETENTION_POLICY_ID
 from phoenix.db.enums import COLUMN_ENUMS, UserRole
+from phoenix.db.types.trace_retention import (
+    MaxDaysRule,
+    TraceRetentionCronExpression,
+    TraceRetentionRule,
+)
 from phoenix.server.email.types import WelcomeEmailSender
 from phoenix.server.types import DbSessionFactory
 
@@ -57,6 +64,7 @@ class Facilitator:
             _ensure_user_roles,
             _get_system_user_id,
             partial(_ensure_admins, email_sender=self._email_sender),
+            _ensure_default_project_trace_retention_policy,
         ):
             await fn(self._db)
 
@@ -205,3 +213,50 @@ async def _ensure_admins(
     ):
         if isinstance(exc, Exception):
             logger.error(f"Failed to send welcome email: {exc}")
+
+
+async def _ensure_default_project_trace_retention_policy(db: DbSessionFactory) -> None:
+    """
+    Ensures the default trace retention policy (id=1) exists in the database. Default policy
+    applies to all projects without a specific policy (i.e. foreign key is null).
+
+    This function checks for the presence of the default trace retention policy and
+    creates it if missing. The default trace retention policy:
+
+        - Has ID=0
+        - Is named "Default"
+        - Runs every Sunday at midnight UTC (cron: "0 0 * * 0")
+        - Retains traces indefinitely
+
+    If the default policy already exists, this function makes no changes.
+
+    Args:
+        db (DbSessionFactory): An async SQLAlchemy session factory.
+
+    Returns:
+        None
+    """
+    assert DEFAULT_PROJECT_TRACE_RETENTION_POLICY_ID == 0
+    async with db() as session:
+        if await session.scalar(
+            select(
+                exists().where(
+                    models.ProjectTraceRetentionPolicy.id
+                    == DEFAULT_PROJECT_TRACE_RETENTION_POLICY_ID
+                )
+            )
+        ):
+            return
+        cron_expression = TraceRetentionCronExpression(root="0 0 * * 0")
+        rule = TraceRetentionRule(root=MaxDaysRule(max_days=0))
+        await session.execute(
+            insert(models.ProjectTraceRetentionPolicy),
+            [
+                {
+                    "id": DEFAULT_PROJECT_TRACE_RETENTION_POLICY_ID,
+                    "name": "Default",
+                    "cron_expression": cron_expression,
+                    "rule": rule,
+                }
+            ],
+        )

{arize_phoenix-8.32.1 → arize_phoenix-9.0.1}/src/phoenix/db/insertion/document_annotation.py

@@ -24,9 +24,17 @@ _SpanRowId: TypeAlias = int
 _DocumentPosition: TypeAlias = int
 _AnnoRowId: TypeAlias = int
 _NumDocs: TypeAlias = int
+_Identifier: TypeAlias = str
 
-_Key: TypeAlias = tuple[_Name, _SpanId, _DocumentPosition]
-_UniqueBy: TypeAlias = tuple[_Name, _SpanRowId, _DocumentPosition]
+
+class _Key(NamedTuple):
+    annotation_name: _Name
+    annotation_identifier: _Identifier
+    span_id: _SpanId
+    document_position: _DocumentPosition
+
+
+_UniqueBy: TypeAlias = tuple[_Name, _SpanRowId, _DocumentPosition, _Identifier]
 _Existing: TypeAlias = tuple[
     _SpanRowId,
     _SpanId,
@@ -34,6 +42,7 @@ _Existing: TypeAlias = tuple[
     Optional[_AnnoRowId],
     Optional[_Name],
     Optional[_DocumentPosition],
+    Optional[_Identifier],
     Optional[datetime],
 ]
 
@@ -46,7 +55,8 @@ class DocumentAnnotationQueueInserter(
         DocumentAnnotationDmlEvent,
     ],
     table=models.DocumentAnnotation,
-    unique_by=("name", "span_rowid", "document_position"),
+    unique_by=("name", "span_rowid", "document_position", "identifier"),
+    constraint_name="uq_document_annotations_name_span_rowid_document_pos_identifier",
 ):
     async def _events(
         self,
@@ -77,12 +87,14 @@ class DocumentAnnotationQueueInserter(
             e.span_id: _SpanAttr(e.span_rowid, e.num_docs) for e in existing
         }
         existing_annos: Mapping[_Key, _AnnoAttr] = {
-            (e.name, e.span_id, e.document_position): _AnnoAttr(e.span_rowid, e.id, e.updated_at)
+            _Key(
+                annotation_name=e.name,
+                annotation_identifier=e.identifier,
+                span_id=e.span_id,
+                document_position=e.document_position,
+            ): _AnnoAttr(e.span_rowid, e.id, e.updated_at)
             for e in existing
-            if e.id is not None
-            and e.name is not None
-            and e.document_position is not None
-            and e.updated_at is not None
+            if e.id is not None and e.name is not None and e.updated_at is not None
         }
 
         for p in parcels:
@@ -129,13 +141,13 @@ class DocumentAnnotationQueueInserter(
         anno = self.table
         span = (
             select(models.Span.id, models.Span.span_id, num_docs_col(self._db.dialect))
-            .where(models.Span.span_id.in_({span_id for _, span_id, *_ in keys}))
+            .where(models.Span.span_id.in_({k.span_id for k in keys}))
             .cte()
         )
         onclause = and_(
             span.c.id == anno.span_rowid,
-            anno.name.in_({name for name, *_ in keys}),
-            tuple_(anno.name, span.c.span_id, anno.document_position).in_(keys),
+            anno.name.in_({k.annotation_name for k in keys}),
+            tuple_(anno.name, anno.identifier, span.c.span_id, anno.document_position).in_(keys),
         )
         return select(
             span.c.id.label("span_rowid"),
@@ -144,6 +156,7 @@ class DocumentAnnotationQueueInserter(
             anno.id,
             anno.name,
             anno.document_position,
+            anno.identifier,
             anno.updated_at,
         ).outerjoin_from(span, anno, onclause)
 
@@ -160,11 +173,16 @@ class _AnnoAttr(NamedTuple):
 
 
 def _key(p: Received[Precursors.DocumentAnnotation]) -> _Key:
-    return p.item.obj.name, p.item.span_id, p.item.document_position
+    return _Key(
+        annotation_name=p.item.obj.name,
+        annotation_identifier=p.item.obj.identifier,
+        span_id=p.item.span_id,
+        document_position=p.item.document_position,
+    )
 
 
 def _unique_by(p: Received[Insertables.DocumentAnnotation]) -> _UniqueBy:
-    return p.item.obj.name, p.item.span_rowid, p.item.document_position
+    return p.item.obj.name, p.item.span_rowid, p.item.document_position, p.item.identifier
 
 
 def _time(p: Received[Any]) -> datetime:

{arize_phoenix-8.32.1 → arize_phoenix-9.0.1}/src/phoenix/db/insertion/evaluation.py

@@ -86,13 +86,15 @@ async def _insert_trace_evaluation(
         explanation=explanation,
         metadata_={},  # `metadata_` must match ORM
         annotator_kind="LLM",
+        identifier="",
+        source="API",
     )
     await session.execute(
         insert_on_conflict(
             values,
             dialect=dialect,
             table=models.TraceAnnotation,
-            unique_by=("name", "trace_rowid"),
+            unique_by=("name", "trace_rowid", "identifier"),
         )
     )
     return TraceEvaluationInsertionEvent(project_rowid, evaluation_name)
@@ -128,13 +130,15 @@ async def _insert_span_evaluation(
         explanation=explanation,
         metadata_={},  # `metadata_` must match ORM
         annotator_kind="LLM",
+        identifier="",
+        source="API",
     )
     await session.execute(
         insert_on_conflict(
             values,
             dialect=dialect,
             table=models.SpanAnnotation,
-            unique_by=("name", "span_rowid"),
+            unique_by=("name", "span_rowid", "identifier"),
         )
     )
     return SpanEvaluationInsertionEvent(project_rowid, evaluation_name)
@@ -179,13 +183,21 @@ async def _insert_document_evaluation(
         explanation=explanation,
         metadata_={},  # `metadata_` must match ORM
         annotator_kind="LLM",
+        identifier="",
+        source="API",
     )
     await session.execute(
         insert_on_conflict(
             values,
             dialect=dialect,
             table=models.DocumentAnnotation,
-            unique_by=("name", "span_rowid", "document_position"),
+            unique_by=(
+                "name",
+                "span_rowid",
+                "document_position",
+                "identifier",
+            ),
+            constraint_name="uq_document_annotations_name_span_rowid_document_pos_identifier",  # The name of the unique constraint is specified manually since the auto-generated name is longer than the Postgres limit of 63 characters  # noqa: E501
         )
     )
     return DocumentEvaluationInsertionEvent(project_rowid, evaluation_name)

{arize_phoenix-8.32.1 → arize_phoenix-9.0.1}/src/phoenix/db/insertion/helpers.py

@@ -36,6 +36,7 @@ def insert_on_conflict(
     unique_by: Sequence[str],
     on_conflict: OnConflict = OnConflict.DO_UPDATE,
     set_: Optional[Mapping[str, Any]] = None,
+    constraint_name: Optional[str] = None,
 ) -> Insert:
     """
     Dialect specific insertion statement using ON CONFLICT DO syntax.
@@ -50,7 +51,7 @@ def insert_on_conflict(
             unique_records.append(v)
             seen.add(k)
         records = tuple(reversed(unique_records))
-    constraint = "_".join(("uq", table.__tablename__, *unique_by))
+    constraint = constraint_name or "_".join(("uq", table.__tablename__, *unique_by))
     if dialect is SupportedSQLDialect.POSTGRESQL:
         stmt_postgresql = insert_postgresql(table).values(records)
         if on_conflict is OnConflict.DO_NOTHING:

{arize_phoenix-8.32.1 → arize_phoenix-9.0.1}/src/phoenix/db/insertion/span_annotation.py

@@ -21,15 +21,23 @@ from phoenix.server.dml_event import SpanAnnotationDmlEvent
 _Name: TypeAlias = str
 _SpanId: TypeAlias = str
 _SpanRowId: TypeAlias = int
+_Identifier: TypeAlias = str
 _AnnoRowId: TypeAlias = int
 
-_Key: TypeAlias = tuple[_Name, _SpanId]
-_UniqueBy: TypeAlias = tuple[_Name, _SpanRowId]
+
+class _Key(NamedTuple):
+    annotation_name: _Name
+    annotation_identifier: _Identifier
+    span_id: _SpanId
+
+
+_UniqueBy: TypeAlias = tuple[_Name, _SpanRowId, _Identifier]
 _Existing: TypeAlias = tuple[
     _SpanRowId,
     _SpanId,
     Optional[_AnnoRowId],
     Optional[_Name],
+    Optional[_Identifier],
     Optional[datetime],
 ]
 
@@ -42,7 +50,7 @@ class SpanAnnotationQueueInserter(
         SpanAnnotationDmlEvent,
     ],
     table=models.SpanAnnotation,
-    unique_by=("name", "span_rowid"),
+    unique_by=("name", "span_rowid", "identifier"),
 ):
     async def _events(
         self,
@@ -73,7 +81,11 @@ class SpanAnnotationQueueInserter(
             e.span_id: _SpanAttr(e.span_rowid) for e in existing
         }
         existing_annos: Mapping[_Key, _AnnoAttr] = {
-            (e.name, e.span_id): _AnnoAttr(e.span_rowid, e.id, e.updated_at)
+            _Key(
+                annotation_name=e.name,
+                annotation_identifier=e.identifier,
+                span_id=e.span_id,
+            ): _AnnoAttr(e.span_rowid, e.id, e.updated_at)
             for e in existing
             if e.id is not None and e.name is not None and e.updated_at is not None
         }
@@ -119,19 +131,20 @@ class SpanAnnotationQueueInserter(
         anno = self.table
         span = (
             select(models.Span.id, models.Span.span_id)
-            .where(models.Span.span_id.in_({span_id for _, span_id in keys}))
+            .where(models.Span.span_id.in_({k.span_id for k in keys}))
             .cte()
         )
         onclause = and_(
             span.c.id == anno.span_rowid,
-            anno.name.in_({name for name, _ in keys}),
-            tuple_(anno.name, span.c.span_id).in_(keys),
+            anno.name.in_({k.annotation_name for k in keys}),
+            tuple_(anno.name, anno.identifier, span.c.span_id).in_(keys),
         )
         return select(
             span.c.id.label("span_rowid"),
             span.c.span_id,
             anno.id,
             anno.name,
+            anno.identifier,
             anno.updated_at,
         ).outerjoin_from(span, anno, onclause)
 
@@ -147,11 +160,15 @@ class _AnnoAttr(NamedTuple):
 
 
 def _key(p: Received[Precursors.SpanAnnotation]) -> _Key:
-    return p.item.obj.name, p.item.span_id
+    return _Key(
+        annotation_name=p.item.obj.name,
+        annotation_identifier=p.item.obj.identifier,
+        span_id=p.item.span_id,
+    )
 
 
 def _unique_by(p: Received[Insertables.SpanAnnotation]) -> _UniqueBy:
-    return p.item.obj.name, p.item.span_rowid
+    return p.item.obj.name, p.item.span_rowid, p.item.identifier
 
 
 def _time(p: Received[Any]) -> datetime:

{arize_phoenix-8.32.1 → arize_phoenix-9.0.1}/src/phoenix/db/insertion/trace_annotation.py

@@ -22,9 +22,16 @@ _Name: TypeAlias = str
 _TraceId: TypeAlias = str
 _TraceRowId: TypeAlias = int
 _AnnoRowId: TypeAlias = int
+_Identifier: TypeAlias = str
 
-_Key: TypeAlias = tuple[_Name, _TraceId]
-_UniqueBy: TypeAlias = tuple[_Name, _TraceRowId]
+
+class _Key(NamedTuple):
+    annotation_name: _Name
+    annotation_identifier: _Identifier
+    trace_id: _TraceId
+
+
+_UniqueBy: TypeAlias = tuple[_Name, _TraceRowId, _Identifier]
 _Existing: TypeAlias = tuple[
     _TraceRowId,
     _TraceId,
@@ -42,7 +49,7 @@ class TraceAnnotationQueueInserter(
         TraceAnnotationDmlEvent,
     ],
     table=models.TraceAnnotation,
-    unique_by=("name", "trace_rowid"),
+    unique_by=("name", "trace_rowid", "identifier"),
 ):
     async def _events(
         self,
@@ -73,7 +80,11 @@ class TraceAnnotationQueueInserter(
             e.trace_id: _TraceAttr(e.trace_rowid) for e in existing
         }
         existing_annos: Mapping[_Key, _AnnoAttr] = {
-            (e.name, e.trace_id): _AnnoAttr(e.trace_rowid, e.id, e.updated_at)
+            _Key(
+                annotation_name=e.name,
+                annotation_identifier=e.identifier,
+                trace_id=e.trace_id,
+            ): _AnnoAttr(e.trace_rowid, e.id, e.updated_at)
             for e in existing
             if e.id is not None and e.name is not None and e.updated_at is not None
         }
@@ -119,19 +130,20 @@ class TraceAnnotationQueueInserter(
         anno = self.table
         trace = (
             select(models.Trace.id, models.Trace.trace_id)
-            .where(models.Trace.trace_id.in_({trace_id for _, trace_id in keys}))
+            .where(models.Trace.trace_id.in_({k.trace_id for k in keys}))
             .cte()
         )
         onclause = and_(
             trace.c.id == anno.trace_rowid,
-            anno.name.in_({name for name, _ in keys}),
-            tuple_(anno.name, trace.c.trace_id).in_(keys),
+            anno.name.in_({k.annotation_name for k in keys}),
+            tuple_(anno.name, anno.identifier, trace.c.trace_id).in_(keys),
         )
         return select(
             trace.c.id.label("trace_rowid"),
             trace.c.trace_id,
             anno.id,
             anno.name,
+            anno.identifier,
             anno.updated_at,
         ).outerjoin_from(trace, anno, onclause)
 
@@ -147,11 +159,15 @@ class _AnnoAttr(NamedTuple):
 
 
 def _key(p: Received[Precursors.TraceAnnotation]) -> _Key:
-    return p.item.obj.name, p.item.trace_id
+    return _Key(
+        annotation_name=p.item.obj.name,
+        annotation_identifier=p.item.obj.identifier,
+        trace_id=p.item.trace_id,
+    )
 
 
 def _unique_by(p: Received[Insertables.TraceAnnotation]) -> _UniqueBy:
-    return p.item.obj.name, p.item.trace_rowid
+    return p.item.obj.name, p.item.trace_rowid, p.item.identifier
 
 
 def _time(p: Received[Any]) -> datetime:

{arize_phoenix-8.32.1 → arize_phoenix-9.0.1}/src/phoenix/db/insertion/types.py

@@ -50,14 +50,17 @@ class Postponed(Received[_AnyT]):
 class QueueInserter(ABC, Generic[_PrecursorT, _InsertableT, _RowT, _DmlEventT]):
     table: type[_RowT]
     unique_by: Sequence[str]
+    constraint_name: Optional[str] = None
 
     def __init_subclass__(
         cls,
         table: type[_RowT],
         unique_by: Sequence[str],
+        constraint_name: Optional[str] = None,
     ) -> None:
         cls.table = table
         cls.unique_by = unique_by
+        cls.constraint_name = constraint_name
 
     def __init__(
         self,
@@ -109,6 +112,7 @@ class QueueInserter(ABC, Generic[_PrecursorT, _InsertableT, _RowT, _DmlEventT]):
             *records,
             table=self.table,
             unique_by=self.unique_by,
+            constraint_name=self.constraint_name,
             dialect=self._db.dialect,
         )
 
@@ -219,6 +223,7 @@ class Insertables(ABC):
     @dataclass(frozen=True)
     class SpanAnnotation(Precursors.SpanAnnotation):
         span_rowid: int
+        identifier: str = ""
         id_: Optional[int] = None
 
         @property
@@ -232,6 +237,7 @@ class Insertables(ABC):
     @dataclass(frozen=True)
     class TraceAnnotation(Precursors.TraceAnnotation):
         trace_rowid: int
+        identifier: str = ""
         id_: Optional[int] = None
 
         @property
@@ -245,6 +251,7 @@ class Insertables(ABC):
     @dataclass(frozen=True)
     class DocumentAnnotation(Precursors.DocumentAnnotation):
         span_rowid: int
+        identifier: str = ""
         id_: Optional[int] = None
 
         @property