From 991fede562aaef383ffe2f1fe0514aa93e63c74b Mon Sep 17 00:00:00 2001
From: Rich Chiodo false <rchiodo@microsoft.com>
Date: Tue, 9 Dec 2025 14:36:15 -0800
Subject: [PATCH] Try fixing gw worker failures

---
 pytest.ini               |  2 +-
 tests/debug/session.py   | 20 ++++++++++++++++++--
 tests/net.py             | 28 ++++++++++++++++++++++++++--
 tests/pytest_fixtures.py | 14 +++++++++++---
 4 files changed, 56 insertions(+), 8 deletions(-)

diff --git a/pytest.ini b/pytest.ini
index 7f3d7d6b..3439ac33 100644
--- a/pytest.ini
+++ b/pytest.ini
@@ -1,5 +1,5 @@
 [pytest]
 testpaths=tests
-timeout=60
+timeout=120
 timeout_method=thread
 addopts=-n8
diff --git a/tests/debug/session.py b/tests/debug/session.py
index d43ee0c7..0b2bea26 100644
--- a/tests/debug/session.py
+++ b/tests/debug/session.py
@@ -281,7 +281,11 @@ class Session(object):
 
         if self.adapter_endpoints is not None and self.expected_exit_code is not None:
             log.info("Waiting for {0} to close listener ports ...", self.adapter_id)
+            timeout_start = time.time()
             while self.adapter_endpoints.check():
+                if time.time() - timeout_start > 10:
+                    log.warning("{0} listener ports did not close within 10 seconds", self.adapter_id)
+                    break
                 time.sleep(0.1)
 
         if self.adapter is not None:
@@ -290,8 +294,20 @@ class Session(object):
                 self.adapter_id,
                 self.adapter.pid,
             )
-            self.adapter.wait()
-            watchdog.unregister_spawn(self.adapter.pid, self.adapter_id)
+            try:
+                self.adapter.wait(timeout=10)
+            except Exception:
+                log.warning("{0} did not exit gracefully within 10 seconds, force-killing", self.adapter_id)
+                try:
+                    self.adapter.kill()
+                    self.adapter.wait(timeout=5)
+                except Exception as e:
+                    log.error("Failed to force-kill {0}: {1}", self.adapter_id, e)
+            
+            try:
+                watchdog.unregister_spawn(self.adapter.pid, self.adapter_id)
+            except Exception as e:
+                log.warning("Failed to unregister adapter spawn: {0}", e)
             self.adapter = None
 
         if self.backchannel is not None:
diff --git a/tests/net.py b/tests/net.py
index f248c1a4..2f468050 100644
--- a/tests/net.py
+++ b/tests/net.py
@@ -17,7 +17,7 @@ from tests.patterns import some
 
 used_ports = set()
     
-def get_test_server_port():
+def get_test_server_port(max_retries=10):
     """Returns a server port number that can be safely used for listening without
     clashing with another test worker process, when running with pytest-xdist.
 
@@ -27,6 +27,9 @@ def get_test_server_port():
 
     Note that if multiple test workers invoke this function with different ranges
     that overlap, conflicts are possible!
+    
+    Args:
+        max_retries: Number of times to retry finding an available port
     """
 
     try:
@@ -39,11 +42,32 @@ def get_test_server_port():
         ), "Unrecognized PYTEST_XDIST_WORKER format"
         n = int(worker_id[2:])
 
+    # Try multiple times to find an available port, with retry logic
+    for attempt in range(max_retries):
+        port = 5678 + (n * 300) + attempt
+        while port in used_ports:
+            port += 1
+        
+        # Verify the port is actually available by trying to bind to it
+        sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
+        sock.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
+        try:
+            sock.bind(("127.0.0.1", port))
+            sock.close()
+            used_ports.add(port)
+            log.info("Allocated port {0} for worker {1}", port, n)
+            return port
+        except OSError as e:
+            log.warning("Port {0} unavailable (attempt {1}/{2}): {3}", port, attempt + 1, max_retries, e)
+            sock.close()
+            time.sleep(0.1 * (attempt + 1))  # Exponential backoff
+    
+    # Fall back to original behavior if all retries fail
     port = 5678 + (n * 300)
     while port in used_ports:
         port += 1
     used_ports.add(port)
-
+    log.warning("Using fallback port {0} after {1} retries", port, max_retries)
     return port
 
 
diff --git a/tests/pytest_fixtures.py b/tests/pytest_fixtures.py
index 27d21f1a..6c99b760 100644
--- a/tests/pytest_fixtures.py
+++ b/tests/pytest_fixtures.py
@@ -46,19 +46,27 @@ def test_wrapper(request, long_tmpdir):
 
     session.Session.reset_counter()
 
-    session.Session.tmpdir = long_tmpdir
+    # Add worker-specific isolation for tmpdir and log directory
+    try:
+        worker_id = os.environ.get("PYTEST_XDIST_WORKER", "gw0")
+        worker_suffix = f"_{worker_id}"
+    except Exception:
+        worker_suffix = ""
+
+    session.Session.tmpdir = long_tmpdir / f"session{worker_suffix}"
+    session.Session.tmpdir.ensure(dir=True)
     original_log_dir = log.log_dir
 
     failed = True
     try:
         if log.log_dir is None:
-            log.log_dir = (long_tmpdir / "debugpy_logs").strpath
+            log.log_dir = (long_tmpdir / f"debugpy_logs{worker_suffix}").strpath
         else:
             log_subdir = request.node.nodeid
             log_subdir = log_subdir.replace("::", "/")
             for ch in r":?*|<>":
                 log_subdir = log_subdir.replace(ch, f"&#{ord(ch)};")
-            log.log_dir += "/" + log_subdir
+            log.log_dir += "/" + log_subdir + worker_suffix
 
         try:
             py.path.local(log.log_dir).remove()