centive-sdk 2.2.0.dev4__tar.gz → 2.2.0.dev5__tar.gz

{centive_sdk-2.2.0.dev4 → centive_sdk-2.2.0.dev5}/CHANGELOG.md

@@ -46,6 +46,14 @@ Centive batcher and the host callback).
   `ServerError`) for every request, since retrying the same body cannot help.
 - Log redaction now masks `page_path`, `page_title`, `path`, `title` and
   `referrer_path`.
+- **A 404 is no longer retried.** It is not a transient failure: the route is
+  absent on that deployment. This is the shape of a Centive backend older than
+  the SDK — `/anam/events` does not exist before page tracking — and of a
+  `trigger-session` for a user with no session in the organization. Previously
+  both were retried three times and then opened a circuit breaker to reach the
+  same answer. Telemetry now drops the batch, says once that the deployment has
+  no product-events endpoint, and keeps everything else working; a rolling
+  upgrade that brings the endpoint up is picked up by the next batch.
 
 ## v2.1.0
 

{centive_sdk-2.2.0.dev4 → centive_sdk-2.2.0.dev5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: centive-sdk
-Version: 2.2.0.dev4
+Version: 2.2.0.dev5
 Summary: Python SDK for Centive backend integration
 Project-URL: Homepage, https://github.com/TheAgenticAI/centive-backend-sdk
 Project-URL: Repository, https://github.com/TheAgenticAI/centive-backend-sdk

{centive_sdk-2.2.0.dev4 → centive_sdk-2.2.0.dev5}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "centive-sdk"
-version = "2.2.0.dev4"
+version = "2.2.0.dev5"
 description = "Python SDK for Centive backend integration"
 readme = "README.md"
 license = "MIT"

{centive_sdk-2.2.0.dev4 → centive_sdk-2.2.0.dev5}/src/centive_sdk/resources/async_sessions.py

@@ -416,6 +416,19 @@ class AsyncSessions:
                     error_detail = self._extract_error_message(response)
                     raise AuthError(f"Authentication failed: {error_detail}", request_id=request_id)
 
+                elif response.status_code == 404:
+                    # Not transient, so it must not be retried: the route is
+                    # absent on this deployment. That is the shape of a Centive
+                    # backend older than the SDK — /anam/events does not exist
+                    # before page tracking — and of a trigger-session for a user
+                    # with no session in this org. Retrying either one three
+                    # times, then opening a circuit breaker, spends requests and
+                    # fills logs to reach the same answer.
+                    error_detail = self._extract_error_message(response)
+                    raise ValidationError(
+                        f"Not found (404): {error_detail}", request_id=request_id
+                    )
+
                 elif response.status_code in [400, 413, 422]:
                     # 413 is "payload over the API's size caps": retrying the
                     # same body cannot succeed, so it is a validation failure.

{centive_sdk-2.2.0.dev4 → centive_sdk-2.2.0.dev5}/src/centive_sdk/resources/telemetry_batcher.py

@@ -95,6 +95,7 @@ class TelemetryBatcher:
         self._wake = asyncio.Event()
         self._running = False
         self._auth_error_logged = False
+        self._missing_endpoint_logged = False
         self.on_context: Optional[ContextCallable] = None
 
     # ------------------------------------------------------------------ state
@@ -246,15 +247,33 @@ class TelemetryBatcher:
                 safe_log(self._logger, "error", f"telemetry rejected: invalid API key ({e})")
             return False
         except ValidationError as e:
-            # 400 / 413 / 422: our payload was wrong; retrying cannot help.
+            # 400 / 404 / 413 / 422: the payload or the route was wrong, and
+            # retrying cannot change either.
             self._stats.dropped_non_retryable += len(batch)
             self._stats.last_error = "ValidationError"
-            safe_log(
-                self._logger,
-                "warning",
-                f"telemetry batch rejected by API, dropped: {e}",
-                {"events": len(batch)},
-            )
+            if "(404)" in str(e):
+                # A Centive deployment older than this SDK: page tracking has
+                # no endpoint to reach there. Said once, plainly, because the
+                # fix is a backend upgrade and not something a retry reaches.
+                # Not fatal: everything else the SDK does still works, and a
+                # rolling upgrade can bring the endpoint up under us, so later
+                # batches keep trying quietly.
+                if not self._missing_endpoint_logged:
+                    self._missing_endpoint_logged = True
+                    safe_log(
+                        self._logger,
+                        "warning",
+                        "this Centive deployment has no product-events endpoint; page "
+                        "events are being discarded. Everything else is unaffected. "
+                        "Set telemetry_enabled=False to stop sending them.",
+                    )
+            else:
+                safe_log(
+                    self._logger,
+                    "warning",
+                    f"telemetry batch rejected by API, dropped: {e}",
+                    {"events": len(batch)},
+                )
             return True
         except Exception as e:  # unknown: treat as retryable but bounded by the breaker
             self._breaker.record_failure()

{centive_sdk-2.2.0.dev4 → centive_sdk-2.2.0.dev5}/tests/test_telemetry_batcher.py

@@ -265,3 +265,56 @@ async def test_context_relay_failure_does_not_break_flush():
     assert batcher.queue_depth == 0
     assert batcher.stats()["accepted"] == 1
     assert batcher.circuit_breaker.state == "closed"
+
+
+@pytest.mark.asyncio
+async def test_missing_endpoint_drops_without_retrying_or_tripping_the_breaker():
+    """A Centive deployment older than this SDK has no /anam/events.
+
+    This is the rollout shape that matters: the SDK ships ahead of the backend
+    for a couple of weeks. A 404 must cost the page events and nothing else —
+    no retry storm, no circuit breaker, no effect on transcript saves.
+    """
+    logged: list[str] = []
+
+    class Recorder:
+        def warning(self, message, *args, **kwargs):
+            logged.append(str(message))
+
+        def __getattr__(self, _name):
+            return lambda *a, **k: None
+
+    sender = RecordingSender([ValidationError("Not found (404): Not Found")])
+    batcher = TelemetryBatcher(sender, make_config(telemetry_flush_max_events=10), logger=Recorder())
+    batcher.enqueue("alice", None, [make_event(), make_event()])
+
+    await batcher.flush()
+
+    assert batcher.queue_depth == 0  # dropped, not requeued
+    assert batcher.circuit_breaker.state == "closed"
+    assert batcher.stats()["dropped_non_retryable"] == 2
+    assert len(sender.requests) == 1  # not retried
+    assert any("no product-events endpoint" in line for line in logged)
+
+
+@pytest.mark.asyncio
+async def test_the_missing_endpoint_is_reported_once_not_per_batch():
+    logged: list[str] = []
+
+    class Recorder:
+        def warning(self, message, *args, **kwargs):
+            logged.append(str(message))
+
+        def __getattr__(self, _name):
+            return lambda *a, **k: None
+
+    sender = RecordingSender(
+        [ValidationError("Not found (404): Not Found"), ValidationError("Not found (404): Not Found")]
+    )
+    batcher = TelemetryBatcher(sender, make_config(telemetry_flush_max_events=1), logger=Recorder())
+    batcher.enqueue("alice", None, [make_event(), make_event()])
+
+    await batcher.flush()
+
+    assert len(sender.requests) == 2
+    assert sum("no product-events endpoint" in line for line in logged) == 1