test: retry Dapr example port bind flake

Signed-off-by: Ghxst <200635707+GHX5T-SOL@users.noreply.github.com>

Author: GHX5T-SOL

tests/examples/conftest.py

@@ -12,6 +12,10 @@
 
 REPO_ROOT = Path(__file__).resolve().parent.parent.parent
 EXAMPLES_DIR = REPO_ROOT / 'examples'
+DAPR_PORT_BIND_FAILURE_MARKERS = (
+    'bind: address already in use',
+    'failed to start internal gRPC server: could not listen on any endpoint',
+)
 
 
 def pytest_configure(config: pytest.Config) -> None:
@@ -38,7 +42,18 @@ def _terminate(proc: subprocess.Popen[str]) -> None:
             terminate_process_group(proc, force=True)
             proc.wait()
 
-    def run(self, args: str, *, timeout: int = 30, until: list[str] | None = None) -> str:
+    @staticmethod
+    def _is_dapr_port_bind_failure(output: str) -> bool:
+        return all(marker in output for marker in DAPR_PORT_BIND_FAILURE_MARKERS)
+
+    def run(
+        self,
+        args: str,
+        *,
+        timeout: int = 30,
+        until: list[str] | None = None,
+        port_bind_retries: int = 1,
+    ) -> str:
         """Run a foreground command, block until it finishes, and return output.
 
         Use this for short-lived processes (e.g. a publisher that exits on its
@@ -49,7 +64,19 @@ def run(self, args: str, *, timeout: int = 30, until: list[str] | None = None) -
             timeout: Maximum seconds to wait before killing the process.
             until: If provided, the process is terminated as soon as every
                 string in this list has appeared in the accumulated output.
+            port_bind_retries: Retry count for Dapr sidecar startup failures
+                caused by a transient random-port collision.
         """
+        attempts = max(1, port_bind_retries + 1)
+        for attempt in range(attempts):
+            output = self._run_once(args, timeout=timeout, until=until)
+            if attempt < attempts - 1 and self._is_dapr_port_bind_failure(output):
+                continue
+            return output
+
+        return output
+
+    def _run_once(self, args: str, *, timeout: int, until: list[str] | None) -> str:
         proc = subprocess.Popen(
             args=('dapr', 'run', *shlex.split(args)),
             cwd=self._cwd,

tests/examples/test_dapr_runner.py

@@ -0,0 +1,55 @@
+import subprocess
+from pathlib import Path
+
+from tests.examples.conftest import DaprRunner
+
+
+class FakeProcess:
+    def __init__(self, output: str, returncode: int = 1) -> None:
+        self.stdout = iter(output.splitlines(keepends=True))
+        self.returncode = returncode
+
+    def poll(self) -> int:
+        return self.returncode
+
+    def wait(self, timeout: int | None = None) -> int:
+        return self.returncode
+
+
+def test_run_retries_transient_dapr_port_bind_failure(monkeypatch, tmp_path: Path) -> None:
+    outputs = [
+        (
+            'level=error msg="Failed to listen for gRPC server on TCP address :33223 '
+            'with error: listen tcp :33223: bind: address already in use"\n'
+            'level=fatal msg="Fatal error from runtime: failed to start internal gRPC '
+            'server: could not listen on any endpoint"\n'
+        ),
+        "{'secretKey': 'secretValue'}\n",
+    ]
+    popen_calls = []
+
+    def fake_popen(*args, **kwargs) -> FakeProcess:
+        popen_calls.append((args, kwargs))
+        return FakeProcess(outputs.pop(0))
+
+    monkeypatch.setattr(subprocess, 'Popen', fake_popen)
+
+    output = DaprRunner(tmp_path).run('--app-id=secretsapp -- python3 example.py', timeout=1)
+
+    assert output == "{'secretKey': 'secretValue'}\n"
+    assert len(popen_calls) == 2
+
+
+def test_run_does_not_retry_non_port_bind_failure(monkeypatch, tmp_path: Path) -> None:
+    popen_calls = []
+
+    def fake_popen(*args, **kwargs) -> FakeProcess:
+        popen_calls.append((args, kwargs))
+        return FakeProcess('application failed before printing expected output\n')
+
+    monkeypatch.setattr(subprocess, 'Popen', fake_popen)
+
+    output = DaprRunner(tmp_path).run('--app-id=secretsapp -- python3 example.py', timeout=1)
+
+    assert output == 'application failed before printing expected output\n'
+    assert len(popen_calls) == 1