arize-phoenix 3.16.1__py3-none-any.whl → 3.16.2__py3-none-any.whl

{arize_phoenix-3.16.1.dist-info → arize_phoenix-3.16.2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: arize-phoenix
-Version: 3.16.1
+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'

{arize_phoenix-3.16.1.dist-info → arize_phoenix-3.16.2.dist-info}/RECORD

@@ -4,13 +4,13 @@ phoenix/datetime_utils.py,sha256=D955QLrkgrrSdUM6NyqbCeAu2SMsjhR5rHVQEsVUdng,277
 phoenix/exceptions.py,sha256=X5k9ipUDfwSCwZB-H5zFJLas86Gf9tAx0W4l5TZxp5k,108
 phoenix/py.typed,sha256=AbpHGcgLb-kRsJGnwFEktk7uzpZOCcBY74-YBdrKVGs,1
 phoenix/services.py,sha256=f6AeyKTuOpy9RCcTCjVH3gx5nYZhbTMFOuv1WSUOB5o,4992
-phoenix/version.py,sha256=RQZ-prBEeYUaDo2wVPl3j8O3CLeHe2r0kT17Sc411m0,23
+phoenix/version.py,sha256=6An59m8khxMeeR51SwvjJubBK8eqW_S7vJrujMumRKc,23
 phoenix/core/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 phoenix/core/embedding_dimension.py,sha256=zKGbcvwOXgLf-yrJBpQyKtd-LEOPRKHnUToyAU8Owis,87
 phoenix/core/model.py,sha256=C-kDATyJEgP-oqYVKOiQM76Ljs66F6VZdT93_b8kTGk,4725
 phoenix/core/model_schema.py,sha256=lQaTvKS34yurHOJ53YD020uURLfgG3dqKC1NLQftOjA,50222
 phoenix/core/model_schema_adapter.py,sha256=3GkyzqUST4fYi-Bgs8qAam5hwMCdQRZTDLjZ9Bnzdm4,8268
-phoenix/core/project.py,sha256=VjfiWH-51FFzgL4EBGEXduCnoAHICw4I-Frr5b5Swxo,24905
+phoenix/core/project.py,sha256=wtpfifivrQUOb3Tj91wBzk6q2_aKFA1FGjz9tRIVptY,25805
 phoenix/core/traces.py,sha256=qAlsDmQJhS9Pkl1IFm-jne-8xS7MEqtw5Q1Ohv9TT2w,3432
 phoenix/datasets/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 phoenix/datasets/dataset.py,sha256=scKVZ7zc6Dpc_ntt-pWhzY-KWqOJEwKePuyNnKSVTGE,30515
@@ -55,8 +55,9 @@ phoenix/pointcloud/pointcloud.py,sha256=4zAIkKs2xOUbchpj4XDAV-iPMXrfAJ15TG6rlIYG
 phoenix/pointcloud/projectors.py,sha256=zO_RrtDYSv2rqVOfIP2_9Cv11Dc8EmcZR94xhFcBYPU,1057
 phoenix/pointcloud/umap_parameters.py,sha256=lJsEOrbSuSiqI7g4Yt6xj7kgYxEqoep4ZHWLr6VWBqw,1760
 phoenix/server/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-phoenix/server/app.py,sha256=7yGzmItFLoTFa5CBZwz4Qb2VZbVbczh_3qRqTTtkaKw,6888
-phoenix/server/main.py,sha256=wrKegfYm-5APU0CqPqpEE_3MUlSgU2LPfpEkYSd870E,9457
+phoenix/server/app.py,sha256=kika5W-3Uy17iqNdRVB-6KdtBEiIrCB_pQRC4LobroI,7170
+phoenix/server/main.py,sha256=zhSu11gEAADD86T2QimmRrIyXlMKrOR2UAkVT_0BlUU,9718
+phoenix/server/prometheus.py,sha256=YQXwXZt3kxXN5JqGKDMH_jI9r4IL23xn6c8Bdrvg15A,2368
 phoenix/server/thread_server.py,sha256=dP6cm6Cf08jNhDA1TRlVZpziu1YgtPDmaeIJMm725eI,2154
 phoenix/server/api/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 phoenix/server/api/context.py,sha256=wjCzq4QlszKG1iN-xgu5rRLYPqdvTFqX02aFYPipNoQ,512
@@ -127,7 +128,7 @@ phoenix/server/static/apple-touch-icon-76x76.png,sha256=CT_xT12I0u2i0WU8JzBZBuOQ
 phoenix/server/static/apple-touch-icon.png,sha256=fOfpjqGpWYbJ0eAurKsyoZP1EAs6ZVooBJ_SGk2ZkDs,3801
 phoenix/server/static/favicon.ico,sha256=bY0vvCKRftemZfPShwZtE93DiiQdaYaozkPGwNFr6H8,34494
 phoenix/server/static/index.css,sha256=KKGpx4iwF91VGRm0YN-4cn8oC-oIqC6HecoPf0x3ZM8,1885
-phoenix/server/static/index.js,sha256=ama86RKEPj-mNPhSYzsYXDngLW73aExeCgI3I5ByjO0,3182471
+phoenix/server/static/index.js,sha256=Bcnr7c5XdUC2kyU57-MIsPpNcDfMCNQP1DIYBROQAa0,3182685
 phoenix/server/static/modernizr.js,sha256=mvK-XtkNqjOral-QvzoqsyOMECXIMu5BQwSVN_wcU9c,2564
 phoenix/server/templates/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 phoenix/server/templates/index.html,sha256=lO2wGA5XsftPg03rw_VcyaYf_4vegtlWbIT5ms4fA_c,1982
@@ -154,7 +155,7 @@ phoenix/trace/trace_dataset.py,sha256=RpHIfZLbMmULOIb-fKXJkQLhIdC0sJlAOTjlyJppMY
 phoenix/trace/utils.py,sha256=7LurVGXn245cjj4MJsc7v6jq4DSJkpK6YGBfIaSywuw,1307
 phoenix/trace/dsl/__init__.py,sha256=WIQIjJg362XD3s50OsPJJ0xbDsGp41bSv7vDllLrPuA,144
 phoenix/trace/dsl/filter.py,sha256=paLpcSMnHdgCfcvcroaqOoCe2retAZ5ocp_5cNTnv9s,14167
-phoenix/trace/dsl/helpers.py,sha256=TG8EFZAjvRwjXpxitEGAc4QpF3vn4jpqhI_Tcwp5mE4,2134
+phoenix/trace/dsl/helpers.py,sha256=urPNmk_D2ZpoKxYWS5DQsQt0nb38O0CoG3wNjOit-yM,2610
 phoenix/trace/dsl/missing.py,sha256=BWPOHr2_tBkPDgVeq8GVXXVbNbJiBelu4NtwHBg6mTE,1435
 phoenix/trace/dsl/query.py,sha256=k0guhWBEo6L7ZJH5FJs2-iGSnWXdUUqu09gd-8M4CGg,14783
 phoenix/trace/langchain/__init__.py,sha256=F37GfD1pd5Kuw7R7iRUM1zXXpO8xEcycNZh5dwqBXNk,109
@@ -171,8 +172,8 @@ phoenix/utilities/error_handling.py,sha256=7b5rpGFj9EWZ8yrZK1IHvxB89suWk3lggDayU
 phoenix/utilities/logging.py,sha256=lDXd6EGaamBNcQxL4vP1au9-i_SXe0OraUDiJOcszSw,222
 phoenix/utilities/project.py,sha256=qWsvKnG1oKhOFUowXf9qiOL2ia7jaFe_ijFFHEt8GJo,431
 phoenix/utilities/span_store.py,sha256=13UK0rE4wQd70yl___WsDRnH0ru-xErng9_Ml7zfEwE,978
-arize_phoenix-3.16.1.dist-info/METADATA,sha256=k1zzlIBTgts8dU1rEDWSEDBP8_kondb0f81P0ipb9R4,29181
-arize_phoenix-3.16.1.dist-info/WHEEL,sha256=bq9SyP5NxIRA9EpQgMCd-9RmPHWvbH-4lTDGwxgIR64,87
-arize_phoenix-3.16.1.dist-info/licenses/IP_NOTICE,sha256=JBqyyCYYxGDfzQ0TtsQgjts41IJoa-hiwDrBjCb9gHM,469
-arize_phoenix-3.16.1.dist-info/licenses/LICENSE,sha256=HFkW9REuMOkvKRACuwLPT0hRydHb3zNg-fdFt94td18,3794
-arize_phoenix-3.16.1.dist-info/RECORD,,
+arize_phoenix-3.16.2.dist-info/METADATA,sha256=woqWRuz22Oi3v7sZupUCnEuztlm4ahUYgjEwJADYOwM,29287
+arize_phoenix-3.16.2.dist-info/WHEEL,sha256=bq9SyP5NxIRA9EpQgMCd-9RmPHWvbH-4lTDGwxgIR64,87
+arize_phoenix-3.16.2.dist-info/licenses/IP_NOTICE,sha256=JBqyyCYYxGDfzQ0TtsQgjts41IJoa-hiwDrBjCb9gHM,469
+arize_phoenix-3.16.2.dist-info/licenses/LICENSE,sha256=HFkW9REuMOkvKRACuwLPT0hRydHb3zNg-fdFt94td18,3794
+arize_phoenix-3.16.2.dist-info/RECORD,,

phoenix/core/project.py

@@ -20,7 +20,6 @@ from typing import (
 )
 
 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
@@ -216,10 +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
 
@@ -285,7 +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):
@@ -353,26 +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)

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=(

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()

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)