quantpad-data 0.2.0__tar.gz → 0.2.1__tar.gz

quantpad_data-0.2.1/.gitignore

@@ -0,0 +1,16 @@
+__pycache__/
+*.py[cod]
+*$py.class
+.env
+.env.local
+.venv/
+venv/
+*.egg-info/
+.mypy_cache/
+.ruff_cache/
+.pytest_cache/
+
+# Throwaway investigation probes. Named so they sort together and are
+# obviously not part of the service; there are usually a lot of them
+# mid-investigation and none of them should ever be committed.
+_qp_*

{quantpad_data-0.2.0 → quantpad_data-0.2.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: quantpad-data
-Version: 0.2.0
+Version: 0.2.1
 Summary: Official Python client for the QuantPad market-data API
 Project-URL: Documentation, https://api.quantpad.ai/external/docs
 Author: QuantPad

{quantpad_data-0.2.0 → quantpad_data-0.2.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "quantpad-data"
-version = "0.2.0"
+version = "0.2.1"
 description = "Official Python client for the QuantPad market-data API"
 readme = "README.md"
 requires-python = ">=3.10"

{quantpad_data-0.2.0 → quantpad_data-0.2.1}/quantpad_data/client.py

@@ -24,7 +24,7 @@ from .errors import (
     ValidationError,
 )
 
-VERSION = "0.2.0"
+VERSION = "0.2.1"
 USER_AGENT = f"quantpad-data-python/{VERSION}"
 TICK_SCHEMAS = (
     "trades", "mbp-1", "mbp-10", "cmbp-1", "tcbbo", "cbbo-1s",
@@ -32,6 +32,28 @@ TICK_SCHEMAS = (
 )
 ROLL_ADJUST_MODES = ("none", "back", "ratio")
 
+#: ``(connect, read)`` for ``/v1/bars``, matching what ``/v1/ticks`` has
+#: always used.
+#:
+#: The default 120s read budget is wrong for this route and was wrong in
+#: the direction that hurts most.  ``/v1/bars`` does not stream: the
+#: server materialises the whole answer before the first byte, so the
+#: read timeout is a cap on total server time rather than a stall
+#: detector.  Measured over three hours of production traffic, 24.6% of
+#: bars requests took longer than 120s -- 37% of those from the notebook
+#: sandbox, and 75% of every ``30s`` request -- against 2 of 1,924 tick
+#: requests exceeding the 1800s ticks already allowed.
+#:
+#: Retrying made it worse rather than better.  The server finishes an
+#: abandoned request (it notices the client left on 1% of them), so a
+#: timeout frees nothing and the three retries turn one slow call into
+#: four concurrent copies of it, each holding one of the eight stream
+#: slots a key is allowed.  That is a self-inflicted 429.
+#:
+#: This is a floor under the symptom, not a fix: the tail beyond 1800s is
+#: server latency and belongs in the engine, not in a client's patience.
+BARS_TIMEOUT = (30, 1800)
+
 
 class Client:
     def __init__(
@@ -164,14 +186,18 @@ class Client:
             "format": "arrow",
             "roll_adjust": self._roll_adjust(symbol, roll_adjust),
         }
-        response = self._request("GET", "/v1/bars", params=params, stream=True)
+        response = self._request(
+            "GET", "/v1/bars", params=params, timeout=BARS_TIMEOUT, stream=True
+        )
         try:
             table = ipc.open_stream(response.content).read_all()
             frame = table.to_pandas()
         except Exception:
             response.close()
             params["format"] = "csv"
-            fallback = self._request("GET", "/v1/bars", params=params, stream=True)
+            fallback = self._request(
+                "GET", "/v1/bars", params=params, timeout=BARS_TIMEOUT, stream=True
+            )
             try:
                 frame = pd.read_csv(io.StringIO(fallback.text))
             finally:

{quantpad_data-0.2.0 → quantpad_data-0.2.1}/tests/test_client.py

@@ -66,6 +66,44 @@ def test_bars_uses_api_key_and_preserves_aliases():
     assert session.calls[0][2]["params"]["roll_adjust"] == "back"
 
 
+def test_bars_gets_the_same_read_budget_as_ticks():
+    """The default 120s was a cap on total server time, not a stall
+    detector: ``/v1/bars`` materialises the whole answer before the first
+    byte, and 24.6% of production bars requests take longer than that.
+    Timing out did not even save the work -- the server finishes the
+    request regardless, so the three retries just ran four copies of it.
+    """
+    from quantpad_data.client import BARS_TIMEOUT
+
+    table = pa.table({"t": [1_700_000_000_000], "c": [1.5]})
+    session = FakeSession([response(body=arrow_bytes(table))])
+    client = Client(api_key="secret", session=session, max_retries=0)
+
+    client.get_bars("GC.FUT", "30s", 1, 2)
+
+    connect, read = session.calls[0][2]["timeout"]
+    assert (connect, read) == BARS_TIMEOUT
+    assert read >= 1800, "a one-year 30s request does not fit in 120s"
+
+
+def test_the_csv_fallback_keeps_the_long_read_budget():
+    """It re-requests the same window, so inheriting the 120s default
+    would time out exactly where the Arrow attempt just did."""
+    from quantpad_data.client import BARS_TIMEOUT
+
+    table = pa.table({"t": [1_700_000_000_000], "c": [1.5]})
+    session = FakeSession(
+        [response(body=b"not arrow"), response(body=b"t,c\n1700000000000,1.5\n")]
+    )
+    client = Client(api_key="secret", session=session, max_retries=0)
+
+    client.get_bars("GC.FUT", "30s", 1, 2)
+
+    assert len(session.calls) == 2
+    assert session.calls[1][2]["timeout"] == BARS_TIMEOUT
+    assert session.calls[1][2]["params"]["format"] == "csv"
+
+
 def test_rate_limit_exposes_retry_after():
     session = FakeSession(
         [response(429, b'{"detail":"rate limit"}', {"Retry-After": "7"})]

quantpad_data-0.2.0/.gitignore

@@ -1,11 +0,0 @@
-__pycache__/
-*.py[cod]
-*$py.class
-.env
-.env.local
-.venv/
-venv/
-*.egg-info/
-.mypy_cache/
-.ruff_cache/
-.pytest_cache/