From 94952e6471f92c2100284d85ba259cafb408bab3 Mon Sep 17 00:00:00 2001
From: Daniel McCoy Stephenson <dmccoystephenson@gmail.com>
Date: Wed, 30 Sep 2026 03:00:46 -0600
Subject: [PATCH] Keep the sender thread alive if _send ever raises

_drain now wraps each _send call in a try that logs at DEBUG and moves on,
so the thread's survival no longer depends on every future edit to _send
keeping all raising code inside its own try. No behavior change today.

Closes #5

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 tests/test_trace_client.py   | 24 ++++++++++++++++++++++++
 trace_client/trace_client.py |  5 ++++-
 2 files changed, 28 insertions(+), 1 deletion(-)

diff --git a/tests/test_trace_client.py b/tests/test_trace_client.py
index 7d5d59f..e1150ea 100644
--- a/tests/test_trace_client.py
+++ b/tests/test_trace_client.py
@@ -247,6 +247,30 @@ def test_a_base_url_without_a_scheme_is_logged_not_a_dead_thread(self):
                          % [r.getMessage() for r in self.log])
         self.assertTrue(all(r.levelno == logging.DEBUG for r in self.log))
 
+    def test_the_sender_thread_survives_an_unexpected_error_from_send(self):
+        crashes = []
+        real_send = TraceClient._send
+        calls = []
+
+        def send_that_fails_once(client, body):
+            calls.append(body)
+            if len(calls) == 1:
+                raise RuntimeError("boom")
+            real_send(client, body)
+
+        with mock.patch.object(threading, "excepthook", crashes.append), \
+                mock.patch.object(TraceClient, "_send", send_that_fails_once):
+            client = TraceClient(self.base_url, "MyGame", "1.2.3", key="k")
+            client.report("first")
+            client.report("second")
+            self.assertTrue(self.capture.arrived.wait(5), "a later report is still delivered")
+            client.close()
+        self.assertEqual([], crashes, "the sender thread must not die with a traceback on stderr")
+        self.assertEqual(["second"], [json.loads(r["body"])["name"] for r in self.capture.requests])
+        self.assertTrue(any("sender failed" in r.getMessage() and "boom" in r.getMessage() for r in self.log),
+                        [r.getMessage() for r in self.log])
+        self.assertTrue(all(r.levelno == logging.DEBUG for r in self.log))
+
     def test_constructor_rejects_a_missing_base_url_or_application(self):
         for base_url, application in ((None, "MyGame"), (" ", "MyGame"), ("http://x", None), ("http://x", "")):
             with self.assertRaises(ValueError):
diff --git a/trace_client/trace_client.py b/trace_client/trace_client.py
index 0954f67..2c3797f 100644
--- a/trace_client/trace_client.py
+++ b/trace_client/trace_client.py
@@ -198,7 +198,10 @@ def _drain(self) -> None:
             body = q.get()
             if body is None:
                 return
-            self._send(body)
+            try:
+                self._send(body)
+            except Exception as failure:  # noqa: BLE001 - a dead sender would leave the queue filling forever
+                _LOG.debug("[trace] sender failed on %s: %s", body, failure)
 
     def _send(self, body: bytes) -> None:
         try: