fix: use status code for PERMISSION_DENIED, remove timeout floor, add tests

- Check e.code() == PERMISSION_DENIED instead of fragile string matching
  on e.details(), preventing false positives/negatives
- Remove 0.5s floor on channel_ready_timeout to prevent overshooting
  dial_timeout when remaining time is small
- Add test verifying channel_ready_timeout is bounded by remaining deadline
- Add tests for PERMISSION_DENIED with custom details and UNAUTHENTICATED
  with "permission denied" text

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: Ambient Code Bot

python/packages/jumpstarter/jumpstarter/client/lease.py

@@ -315,12 +315,14 @@ def __contextmanager__(self) -> Generator[Self]:
 
     # DEADLINE_EXCEEDED and CANCELLED are excluded: they indicate client-side
     # timeout or cancellation, not server/network transients worth retrying.
-    _TRANSIENT_GRPC_CODES = frozenset({
-        grpc.StatusCode.UNAVAILABLE,
-        grpc.StatusCode.RESOURCE_EXHAUSTED,
-        grpc.StatusCode.ABORTED,
-        grpc.StatusCode.INTERNAL,
-    })
+    _TRANSIENT_GRPC_CODES = frozenset(
+        {
+            grpc.StatusCode.UNAVAILABLE,
+            grpc.StatusCode.RESOURCE_EXHAUSTED,
+            grpc.StatusCode.ABORTED,
+            grpc.StatusCode.INTERNAL,
+        }
+    )
 
     # UNKNOWN error messages that indicate transient tunnel teardowns.
     # We don't blanket-retry all UNKNOWN errors (they could be permanent
@@ -333,9 +335,7 @@ def _retry_delay(attempt: int, remaining: float, base: float = 0.3, cap: float =
         """Compute exponential-backoff delay, capped by *cap* and *remaining* time."""
         return min(base * (2**attempt), cap, remaining)
 
-    async def _dial_and_connect(
-        self, stream: SocketStream, channel_ready_timeout: float = 10.0
-    ) -> None:
+    async def _dial_and_connect(self, stream: SocketStream, channel_ready_timeout: float = 10.0) -> None:
         """Single attempt; raises on failure for caller-driven retry."""
         response = await self.controller.Dial(jumpstarter_pb2.DialRequest(lease_name=self.name))
         async with connect_router_stream(
@@ -360,7 +360,7 @@ async def handle_async(self, stream: SocketStream) -> None:
         attempt = 0
         while True:
             remaining = deadline - time.monotonic()
-            channel_ready_timeout = max(min(10.0, remaining), 0.5)
+            channel_ready_timeout = max(min(10.0, remaining), 0)
             try:
                 await self._dial_and_connect(stream, channel_ready_timeout=channel_ready_timeout)
                 return
@@ -409,7 +409,7 @@ async def handle_async(self, stream: SocketStream) -> None:
                     await sleep(delay)
                     attempt += 1
                     continue
-                if "permission denied" in str(e.details()).lower():
+                if e.code() == grpc.StatusCode.PERMISSION_DENIED:
                     self.lease_transferred = True
                     logger.warning(
                         "Lease %s has been transferred to another client. Your session is no longer valid.",

python/packages/jumpstarter/jumpstarter/client/lease_test.py

@@ -678,9 +678,7 @@ async def dial_side_effect(req):
             nonlocal call_count
             call_count += 1
             if call_count == 1:
-                raise _make_aio_rpc_error(
-                    grpc.StatusCode.UNKNOWN, "watch channel closed"
-                )
+                raise _make_aio_rpc_error(grpc.StatusCode.UNKNOWN, "watch channel closed")
             return dial_response
 
         lease.controller.Dial = AsyncMock(side_effect=dial_side_effect)
@@ -703,9 +701,7 @@ async def test_unknown_without_known_message_not_retried(self):
         lease = _make_lease_for_handle()
 
         lease.controller.Dial = AsyncMock(
-            side_effect=_make_aio_rpc_error(
-                grpc.StatusCode.UNKNOWN, "some unexpected server bug"
-            ),
+            side_effect=_make_aio_rpc_error(grpc.StatusCode.UNKNOWN, "some unexpected server bug"),
         )
 
         with patch("jumpstarter.client.lease.sleep", new_callable=AsyncMock):
@@ -874,9 +870,7 @@ async def dial_side_effect(req):
             nonlocal call_count
             call_count += 1
             if call_count <= total_failures:
-                raise _make_aio_rpc_error(
-                    grpc.StatusCode.UNAVAILABLE, "tunnel dropped"
-                )
+                raise _make_aio_rpc_error(grpc.StatusCode.UNAVAILABLE, "tunnel dropped")
             return dial_response
 
         lease.controller.Dial = AsyncMock(side_effect=dial_side_effect)
@@ -904,10 +898,7 @@ async def fake_router(*args, **kwargs):
         actual_delays = [call.args[0] for call in mock_sleep.call_args_list]
         assert len(actual_delays) == len(expected_delays)
         for actual, expected in zip(actual_delays, expected_delays, strict=True):
-            assert actual == pytest.approx(expected), (
-                f"Expected delay {expected}, got {actual}"
-            )
-
+            assert actual == pytest.approx(expected), f"Expected delay {expected}, got {actual}"
 
     @pytest.mark.anyio
     async def test_failed_precondition_not_ready_retries_then_succeeds(self):
@@ -920,9 +911,7 @@ async def dial_side_effect(req):
             nonlocal call_count
             call_count += 1
             if call_count == 1:
-                raise _make_aio_rpc_error(
-                    grpc.StatusCode.FAILED_PRECONDITION, "exporter not ready"
-                )
+                raise _make_aio_rpc_error(grpc.StatusCode.FAILED_PRECONDITION, "exporter not ready")
             return dial_response
 
         lease.controller.Dial = AsyncMock(side_effect=dial_side_effect)
@@ -946,9 +935,7 @@ async def test_failed_precondition_returns_after_timeout(self):
         lease.dial_timeout = 0.0  # already expired
 
         lease.controller.Dial = AsyncMock(
-            side_effect=_make_aio_rpc_error(
-                grpc.StatusCode.FAILED_PRECONDITION, "exporter not ready"
-            ),
+            side_effect=_make_aio_rpc_error(grpc.StatusCode.FAILED_PRECONDITION, "exporter not ready"),
         )
 
         with patch("jumpstarter.client.lease.sleep", new_callable=AsyncMock):
@@ -963,16 +950,79 @@ async def test_permission_denied_sets_lease_transferred(self):
         lease = _make_lease_for_handle()
         assert lease.lease_transferred is False
 
+        lease.controller.Dial = AsyncMock(
+            side_effect=_make_aio_rpc_error(grpc.StatusCode.PERMISSION_DENIED, "permission denied"),
+        )
+
+        with patch("jumpstarter.client.lease.sleep", new_callable=AsyncMock):
+            await lease.handle_async(Mock())
+
+        assert lease.lease_transferred is True
+
+    @pytest.mark.anyio
+    async def test_permission_denied_with_custom_details_still_detected(self):
+        """PERMISSION_DENIED with non-standard detail text should still set lease_transferred."""
+        lease = _make_lease_for_handle()
+        assert lease.lease_transferred is False
+
+        lease.controller.Dial = AsyncMock(
+            side_effect=_make_aio_rpc_error(grpc.StatusCode.PERMISSION_DENIED, "lease reassigned to another client"),
+        )
+
+        with patch("jumpstarter.client.lease.sleep", new_callable=AsyncMock):
+            await lease.handle_async(Mock())
+
+        assert lease.lease_transferred is True
+
+    @pytest.mark.anyio
+    async def test_unauthenticated_with_permission_text_does_not_set_transferred(self):
+        """UNAUTHENTICATED with 'permission denied' in details should NOT set lease_transferred."""
+        lease = _make_lease_for_handle()
+        assert lease.lease_transferred is False
+
         lease.controller.Dial = AsyncMock(
             side_effect=_make_aio_rpc_error(
-                grpc.StatusCode.PERMISSION_DENIED, "permission denied"
+                grpc.StatusCode.UNAUTHENTICATED,
+                "permission denied: token expired",
             ),
         )
 
         with patch("jumpstarter.client.lease.sleep", new_callable=AsyncMock):
             await lease.handle_async(Mock())
 
-        assert lease.lease_transferred is True
+        assert lease.lease_transferred is False
+
+    @pytest.mark.anyio
+    async def test_channel_ready_timeout_bounded_by_remaining(self):
+        """channel_ready_timeout should decrease as the dial deadline approaches."""
+        lease = _make_lease_for_handle()
+        lease.dial_timeout = 3.0
+
+        call_count = 0
+        captured_timeouts = []
+
+        async def tracking_dial_and_connect(self_inner, stream, channel_ready_timeout=10.0):
+            nonlocal call_count
+            call_count += 1
+            captured_timeouts.append(channel_ready_timeout)
+            if call_count <= 3:
+                raise _make_aio_rpc_error(grpc.StatusCode.FAILED_PRECONDITION, "exporter not ready")
+            # Succeed on 4th attempt (won't normally reach here with 3s timeout)
+
+        with (
+            patch.object(type(lease), "_dial_and_connect", tracking_dial_and_connect),
+            patch("jumpstarter.client.lease.sleep", new_callable=AsyncMock),
+        ):
+            await lease.handle_async(Mock())
+
+        # With a 3s dial_timeout, the first call should have channel_ready_timeout <= 3.0
+        # and subsequent calls should have progressively smaller values
+        assert len(captured_timeouts) >= 2
+        assert all(t <= 10.0 for t in captured_timeouts), f"All timeouts should be <= 10.0, got {captured_timeouts}"
+        # The first timeout should be bounded by remaining (~3.0), not the default 10.0
+        assert captured_timeouts[0] <= 3.1, (
+            f"First timeout should be bounded by dial_timeout (~3.0), got {captured_timeouts[0]}"
+        )
 
 
 class TestRetryDelay: