arize-phoenix 3.16.0__tar.gz → 3.16.2__tar.gz

{arize_phoenix-3.16.0 → arize_phoenix-3.16.2}/PKG-INFO

@@ -1,6 +1,6 @@
-Metadata-Version: 2.1
+Metadata-Version: 2.3
 Name: arize-phoenix
-Version: 3.16.0
+Version: 3.16.2
 Summary: AI Observability and Evaluation
 Project-URL: Documentation, https://docs.arize.com/phoenix/
 Project-URL: Issues, https://github.com/Arize-ai/phoenix/issues
@@ -17,7 +17,6 @@ Classifier: Programming Language :: Python :: 3.10
 Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Requires-Python: <3.13,>=3.8
-Requires-Dist: ddsketch
 Requires-Dist: hdbscan>=0.8.33
 Requires-Dist: jinja2
 Requires-Dist: numpy
@@ -44,6 +43,8 @@ Requires-Dist: typing-extensions>=4.6; python_version >= '3.12'
 Requires-Dist: umap-learn
 Requires-Dist: uvicorn
 Requires-Dist: wrapt
+Provides-Extra: container
+Requires-Dist: prometheus-client; extra == 'container'
 Provides-Extra: dev
 Requires-Dist: anthropic; extra == 'dev'
 Requires-Dist: arize[autoembeddings,llm-evaluation]; extra == 'dev'
@@ -57,6 +58,7 @@ Requires-Dist: llama-index>=0.10.3; extra == 'dev'
 Requires-Dist: nbqa; extra == 'dev'
 Requires-Dist: pandas-stubs<=2.0.2.230605; extra == 'dev'
 Requires-Dist: pre-commit; extra == 'dev'
+Requires-Dist: prometheus-client; extra == 'dev'
 Requires-Dist: pytest-asyncio; extra == 'dev'
 Requires-Dist: pytest-cov; extra == 'dev'
 Requires-Dist: pytest-lazy-fixture; extra == 'dev'
@@ -109,7 +111,7 @@ Phoenix provides MLOps and LLMOps insights at lightning speed with zero-config o
 -   **LLM Traces** - Trace through the execution of your LLM Application to understand the internals of your LLM Application and to troubleshoot problems related to things like retrieval and tool execution.
 -   **LLM Evals** - Leverage the power of large language models to evaluate your generative model or application's relevance, toxicity, and more.
 -   **Embedding Analysis** - Explore embedding point-clouds and identify clusters of high drift and performance degradation.
--   **RAG Analysis** - Visualize your generative application's search and retrieval process to solve improve your retrieval-augmented generation.
+-   **RAG Analysis** - Visualize your generative application's search and retrieval process to identify problems and improve your RAG pipeline.
 -   **Structured Data Analysis** - Statistically analyze your structured data by performing A/B analysis, temporal drift analysis, and more.
 
 **Table of Contents**
@@ -230,7 +232,7 @@ session = px.launch_app()
 
 # Once you have started a Phoenix server, you can start your LangChain application with the OpenInferenceTracer as a callback. To do this, you will have to instrument your LangChain application with the tracer:
 
-from phoenix.trace.langchain import OpenInferenceTracer, LangChainInstrumentor
+from phoenix.trace.langchain import LangChainInstrumentor
 
 # By default, the traces will be exported to the locally running Phoenix server.
 LangChainInstrumentor().instrument()

{arize_phoenix-3.16.0 → arize_phoenix-3.16.2}/README.md

@@ -34,7 +34,7 @@ Phoenix provides MLOps and LLMOps insights at lightning speed with zero-config o
 -   **LLM Traces** - Trace through the execution of your LLM Application to understand the internals of your LLM Application and to troubleshoot problems related to things like retrieval and tool execution.
 -   **LLM Evals** - Leverage the power of large language models to evaluate your generative model or application's relevance, toxicity, and more.
 -   **Embedding Analysis** - Explore embedding point-clouds and identify clusters of high drift and performance degradation.
--   **RAG Analysis** - Visualize your generative application's search and retrieval process to solve improve your retrieval-augmented generation.
+-   **RAG Analysis** - Visualize your generative application's search and retrieval process to identify problems and improve your RAG pipeline.
 -   **Structured Data Analysis** - Statistically analyze your structured data by performing A/B analysis, temporal drift analysis, and more.
 
 **Table of Contents**
@@ -155,7 +155,7 @@ session = px.launch_app()
 
 # Once you have started a Phoenix server, you can start your LangChain application with the OpenInferenceTracer as a callback. To do this, you will have to instrument your LangChain application with the tracer:
 
-from phoenix.trace.langchain import OpenInferenceTracer, LangChainInstrumentor
+from phoenix.trace.langchain import LangChainInstrumentor
 
 # By default, the traces will be exported to the locally running Phoenix server.
 LangChainInstrumentor().instrument()

{arize_phoenix-3.16.0 → arize_phoenix-3.16.2}/pyproject.toml

@@ -40,7 +40,6 @@ dependencies = [
   "wrapt",
   "sortedcontainers",
   "protobuf>=3.20, <5.0",
-  "ddsketch",
   "tqdm",
   "requests",
   "opentelemetry-sdk",
@@ -73,6 +72,7 @@ dev = [
   "litellm>=1.0.3",
   "google-cloud-aiplatform>=1.3",
   "anthropic",
+  "prometheus_client",
 ]
 evals = [
   "arize-phoenix-evals>=0.3.0",
@@ -85,6 +85,9 @@ llama-index = [
   "llama-index-callbacks-arize-phoenix>=0.1.2",
   "openinference-instrumentation-llama-index>=1.2.0",
 ]
+container = [
+  "prometheus-client",
+]
 
 [project.urls]
 Documentation = "https://docs.arize.com/phoenix/"
@@ -145,6 +148,7 @@ dependencies = [
   "types-setuptools",
   "openai>=1.0.0",
   "litellm>=1.0.3",
+  "prometheus_client",
 ]
 
 [tool.hatch.envs.style]

{arize_phoenix-3.16.0 → arize_phoenix-3.16.2}/src/phoenix/core/project.py

@@ -13,13 +13,13 @@ from typing import (
     Mapping,
     Optional,
     Set,
+    Sized,
     Tuple,
     Union,
     cast,
 )
 
 import numpy as np
-from ddsketch import DDSketch
 from google.protobuf.json_format import MessageToDict
 from openinference.semconv.trace import SpanAttributes
 from pandas import DataFrame, Index, MultiIndex
@@ -215,10 +215,15 @@ class _Spans:
         self._start_time_sorted_root_spans: SortedKeyList[WrappedSpan] = SortedKeyList(
             key=lambda span: span.start_time,
         )
+        """
+        A root span is defined to be a span whose parent span is not in our collection.
+        This includes spans whose parent is None and spans whose parent has not arrived
+        (or will not arrive). For spans whose parent is not None, the root span status
+        is temporary and will be revoked when its parent span arrives.
+        """
         self._latency_sorted_root_spans: SortedKeyList[WrappedSpan] = SortedKeyList(
             key=lambda span: span[ComputedAttributes.LATENCY_MS],
         )
-        self._root_span_latency_ms_sketch = DDSketch()
         self._token_count_total: int = 0
         self._last_updated_at: Optional[datetime] = None
 
@@ -284,7 +289,15 @@ class _Spans:
     def root_span_latency_ms_quantiles(self, probability: float) -> Optional[float]:
         """Root span latency quantiles in milliseconds"""
         with self._lock:
-            return self._root_span_latency_ms_sketch.get_quantile_value(probability)
+            spans = self._latency_sorted_root_spans
+            if not (n := len(spans)):
+                return None
+            if probability >= 1:
+                return cast(float, spans[-1][ComputedAttributes.LATENCY_MS])
+            if probability <= 0:
+                return cast(float, spans[0][ComputedAttributes.LATENCY_MS])
+            k = max(0, round(n * probability) - 1)
+            return cast(float, spans[k][ComputedAttributes.LATENCY_MS])
 
     def get_descendant_spans(self, span_id: SpanID) -> Iterator[WrappedSpan]:
         for span in self._get_descendant_spans(span_id):
@@ -352,26 +365,27 @@ class _Spans:
             return
 
         parent_span_id = span.parent_id
-        is_root_span = parent_span_id is None
-        if not is_root_span:
+        if parent_span_id is not None:
             self._child_spans[parent_span_id].add(span)
             self._parent_span_ids[span_id] = parent_span_id
 
+        for child_span in self._child_spans.get(span_id, ()):
+            # A root span is a span whose parent span is not in our collection.
+            # Now that their parent span has arrived, they are no longer root spans.
+            self._start_time_sorted_root_spans.remove(child_span)
+            self._latency_sorted_root_spans.remove(child_span)
+
         # Add computed attributes to span
         start_time = span.start_time
         end_time = span.end_time
-        span[ComputedAttributes.LATENCY_MS] = latency = (
-            end_time - start_time
-        ).total_seconds() * 1000
-        if is_root_span:
-            self._root_span_latency_ms_sketch.add(latency)
+        span[ComputedAttributes.LATENCY_MS] = (end_time - start_time).total_seconds() * 1000
         span[ComputedAttributes.ERROR_COUNT] = int(span.status_code is SpanStatusCode.ERROR)
 
         # Store the new span (after adding computed attributes)
         self._spans[span_id] = span
         self._traces[span.context.trace_id].add(span)
         self._start_time_sorted_spans.add(span)
-        if is_root_span:
+        if parent_span_id is None or parent_span_id not in self._spans:
             self._start_time_sorted_root_spans.add(span)
             self._latency_sorted_root_spans.add(span)
         self._propagate_cumulative_values(span)
@@ -386,9 +400,10 @@ class _Spans:
         span_id = span.context.span_id
         if token_count_update := span.attributes.get(SpanAttributes.LLM_TOKEN_COUNT_TOTAL):
             self._token_count_total += token_count_update
-        if num_documents_update := len(
-            span.attributes.get(SpanAttributes.RETRIEVAL_DOCUMENTS) or ()
-        ):
+        if isinstance(
+            (retrieval_documents := span.attributes.get(SpanAttributes.RETRIEVAL_DOCUMENTS)),
+            Sized,
+        ) and (num_documents_update := len(retrieval_documents)):
             self._num_documents[span_id] += num_documents_update
 
     def _propagate_cumulative_values(self, span: WrappedSpan) -> None:

{arize_phoenix-3.16.0 → arize_phoenix-3.16.2}/src/phoenix/core/traces.py

@@ -46,17 +46,13 @@ class Traces:
                 yield project_id, project_name, project
 
     def archive_project(self, id: int) -> Optional["Project"]:
+        if id == 0:
+            raise ValueError("Cannot archive the default project")
         with self._lock:
-            active_projects = {
-                project_id: project
-                for project_id, _, project in self.get_projects()
-                if not project.is_archived
-            }
-            if len(active_projects) <= 1:
-                return None
-            if project := active_projects.get(id):
-                project.archive()
-                return project
+            for project_id, _, project in self.get_projects():
+                if id == project_id:
+                    project.archive()
+                    return project
         return None
 
     def put(

{arize_phoenix-3.16.0 → arize_phoenix-3.16.2}/src/phoenix/server/api/schema.py

@@ -31,7 +31,7 @@ from .types.Event import create_event_id, unpack_event_id
 from .types.ExportEventsMutation import ExportEventsMutation
 from .types.Functionality import Functionality
 from .types.Model import Model
-from .types.node import GlobalID, Node, from_global_id
+from .types.node import GlobalID, Node, from_global_id, from_global_id_with_expected_type
 from .types.pagination import Connection, ConnectionArgs, Cursor, connection_from_list
 
 
@@ -230,22 +230,16 @@ class Query:
 class Mutation(ExportEventsMutation):
     @strawberry.mutation
     def delete_project(self, info: Info[Context, None], id: GlobalID) -> Query:
-        if (traces := info.context.traces) is None:
-            return Query()
-        type_name, node_id = from_global_id(str(id))
-        if type_name != "Project":
-            return Query()
-        traces.archive_project(node_id)
+        if (traces := info.context.traces) is not None:
+            node_id = from_global_id_with_expected_type(str(id), "Project")
+            traces.archive_project(node_id)
         return Query()
 
     @strawberry.mutation
     def archive_project(self, info: Info[Context, None], id: GlobalID) -> Query:
-        if (traces := info.context.traces) is None:
-            return Query()
-        type_name, node_id = from_global_id(str(id))
-        if type_name != "Project":
-            return Query()
-        traces.archive_project(node_id)
+        if (traces := info.context.traces) is not None:
+            node_id = from_global_id_with_expected_type(str(id), "Project")
+            traces.archive_project(node_id)
         return Query()
 
 

{arize_phoenix-3.16.0 → arize_phoenix-3.16.2}/src/phoenix/server/api/types/Span.py

@@ -2,7 +2,7 @@ import json
 from collections import defaultdict
 from datetime import datetime
 from enum import Enum
-from typing import Any, DefaultDict, Dict, List, Mapping, Optional, Sized, cast
+from typing import Any, DefaultDict, Dict, Iterable, List, Mapping, Optional, Sized, cast
 
 import numpy as np
 import strawberry
@@ -339,12 +339,12 @@ def _hide_embedding_vectors(
     attributes: Mapping[str, Any],
 ) -> Dict[str, Any]:
     _attributes = dict(attributes)
-    if not (embeddings := _attributes.get(EMBEDDING_EMBEDDINGS)):
+    if not isinstance((embeddings := _attributes.get(EMBEDDING_EMBEDDINGS)), Iterable):
         return _attributes
     _embeddings = []
     for embedding in embeddings:
         _embedding = dict(embedding)
-        if vector := _embedding.get(EMBEDDING_VECTOR):
+        if isinstance((vector := _embedding.get(EMBEDDING_VECTOR)), Sized):
             _embedding[EMBEDDING_VECTOR] = f"<{len(vector)} dimensional vector>"
         _embeddings.append(_embedding)
     _attributes[EMBEDDING_EMBEDDINGS] = _embeddings

{arize_phoenix-3.16.0 → arize_phoenix-3.16.2}/src/phoenix/server/api/types/node.py

@@ -30,6 +30,20 @@ def from_global_id(global_id: str) -> Tuple[str, int]:
     return type_name, int(node_id)
 
 
+def from_global_id_with_expected_type(global_id: str, expected_type_name: str) -> int:
+    """
+    Decodes the given global id and return the id, checking that the type
+    matches the expected type.
+    """
+    type_name, node_id = from_global_id(global_id)
+    if type_name != expected_type_name:
+        raise ValueError(
+            f"The node id must correspond to a node of type {expected_type_name}, "
+            f"but instead corresponds to a node of type: {type_name}"
+        )
+    return node_id
+
+
 class GlobalIDValueError(ValueError):
     """GlobalID value error, usually related to parsing or serialization."""
 

{arize_phoenix-3.16.0 → arize_phoenix-3.16.2}/src/phoenix/server/app.py

@@ -151,6 +151,7 @@ def create_app(
     span_store: Optional[SpanStore] = None,
     debug: bool = False,
     read_only: bool = False,
+    enable_prometheus: bool = False,
 ) -> Starlette:
     graphql = GraphQLWithContext(
         schema=schema,
@@ -160,9 +161,16 @@ def create_app(
         export_path=export_path,
         graphiql=True,
     )
+    if enable_prometheus:
+        from phoenix.server.prometheus import PrometheusMiddleware
+
+        prometheus_middlewares = [Middleware(PrometheusMiddleware)]
+    else:
+        prometheus_middlewares = []
     return Starlette(
         middleware=[
             Middleware(HeadersMiddleware),
+            *prometheus_middlewares,
         ],
         debug=debug,
         routes=(

{arize_phoenix-3.16.0 → arize_phoenix-3.16.2}/src/phoenix/server/main.py

@@ -129,6 +129,7 @@ if __name__ == "__main__":
     parser.add_argument("--no-internet", action="store_true")
     parser.add_argument("--umap_params", type=str, required=False, default=DEFAULT_UMAP_PARAMS_STR)
     parser.add_argument("--debug", action="store_false")
+    parser.add_argument("--enable-prometheus", type=bool, default=False)
     subparsers = parser.add_subparsers(dest="command", required=True)
     serve_parser = subparsers.add_parser("serve")
     datasets_parser = subparsers.add_parser("datasets")
@@ -223,6 +224,10 @@ if __name__ == "__main__":
     )
     read_only = args.read_only
     logger.info(f"Server umap params: {umap_params}")
+    if enable_prometheus := args.enable_prometheus:
+        from phoenix.server.prometheus import start_prometheus
+
+        start_prometheus()
     app = create_app(
         export_path=export_path,
         model=model,
@@ -232,6 +237,7 @@ if __name__ == "__main__":
         debug=args.debug,
         read_only=read_only,
         span_store=span_store,
+        enable_prometheus=enable_prometheus,
     )
     host = args.host or get_env_host()
     port = args.port or get_env_port()

arize_phoenix-3.16.2/src/phoenix/server/prometheus.py

@@ -0,0 +1,75 @@
+import time
+from threading import Thread
+
+import psutil
+from prometheus_client import (
+    Counter,
+    Gauge,
+    Histogram,
+    start_http_server,
+)
+from starlette.middleware.base import BaseHTTPMiddleware, RequestResponseEndpoint
+from starlette.requests import Request
+from starlette.responses import Response
+from starlette.routing import Match
+
+REQUESTS_PROCESSING_TIME = Histogram(
+    name="starlette_requests_processing_time_seconds",
+    documentation="Histogram of requests processing time by method and path (in seconds)",
+    labelnames=["method", "path"],
+)
+EXCEPTIONS = Counter(
+    name="starlette_exceptions_total",
+    documentation="Total count of exceptions raised by method, path and exception type",
+    labelnames=["method", "path", "exception_type"],
+)
+RAM_METRIC = Gauge(
+    name="memory_usage_bytes",
+    documentation="Memory usage in bytes",
+    labelnames=["type"],
+)
+CPU_METRIC = Gauge(
+    name="cpu_usage_percent",
+    documentation="CPU usage percent",
+    labelnames=["core"],
+)
+
+
+class PrometheusMiddleware(BaseHTTPMiddleware):
+    async def dispatch(self, request: Request, call_next: RequestResponseEndpoint) -> Response:
+        for route in request.app.routes:
+            match, _ = route.matches(request.scope)
+            if match is Match.FULL:
+                path = route.path
+                break
+        else:
+            return await call_next(request)
+        method = request.method
+        start_time = time.perf_counter()
+        try:
+            response = await call_next(request)
+        except BaseException as e:
+            EXCEPTIONS.labels(method=method, path=path, exception_type=type(e).__name__).inc()
+            raise
+        stop_time = time.perf_counter()
+        REQUESTS_PROCESSING_TIME.labels(method=method, path=path).observe(stop_time - start_time)
+        return response
+
+
+def start_prometheus() -> None:
+    Thread(target=gather_system_data, daemon=True).start()
+    start_http_server(9090)
+
+
+def gather_system_data() -> None:
+    while True:
+        time.sleep(1)
+
+        ram = psutil.virtual_memory()
+        swap = psutil.swap_memory()
+
+        RAM_METRIC.labels(type="virtual").set(ram.used)
+        RAM_METRIC.labels(type="swap").set(swap.used)
+
+        for core, percent in enumerate(psutil.cpu_percent(interval=1, percpu=True)):
+            CPU_METRIC.labels(core=core).set(percent)