Slai Agent Runtime：Improve Linux PTY output draining for large command logs

launch/core/platforms/linux.py

@@ -95,22 +95,35 @@ def _read_raw_output(self, timeout:int=30) -> tuple[str, Optional[CmdOutputMetad
         '''
         timeout: in seconds
         '''
-
-        accumulated_output = ""
+
+        # Do not repeatedly normalize and regex-scan the complete PTY transcript
+        # after every chunk. Large build/test commands can emit tens of megabytes;
+        # rescanning the growing string turns one command read into quadratic host
+        # work after the command itself has already completed.
+        output_chunks: list[str] = []
+        prompt_probe = ""
+        prompt_end = CMD_OUTPUT_PS1_END.strip()
         start_time = time.time()
 
         while time.time() - start_time < timeout:
             try:
                 chunk = self.output_queue.get(timeout=0.1)
-                accumulated_output += chunk.decode("utf-8", errors="ignore")
-                # PSReadLine injects ANSI + cursor control; normalize before matching
-                accumulated_clean = ANSI_ESCAPE.sub("", accumulated_output).replace("\r", "")
-                ps1_matches = CmdOutputMetadata.matches_ps1_metadata(accumulated_clean)
-                if ps1_matches:
+                normalized_chunk = ANSI_ESCAPE.sub(
+                    "", chunk.decode("utf-8", errors="ignore")
+                ).replace("\r", "")
+                output_chunks.append(normalized_chunk)
+
+                # The delimiter can straddle two socket reads, so retain only the
+                # short suffix needed to recognize it incrementally. Parse the
+                # complete metadata exactly once after the delimiter is observed.
+                prompt_probe += normalized_chunk
+                if prompt_end in prompt_probe:
                     break
+                prompt_probe = prompt_probe[-(len(prompt_end) - 1):]
             except queue.Empty:
                 continue
-        accumulated_output = ANSI_ESCAPE.sub("", accumulated_output).replace("\r", "")
+
+        accumulated_output = "".join(output_chunks)
         ps1_matches = CmdOutputMetadata.matches_ps1_metadata(accumulated_output)
         metadata = (
             CmdOutputMetadata.from_ps1_match(ps1_matches[-1]) if ps1_matches else None

tests/linux_output_drain_test.py

@@ -0,0 +1,52 @@
+import queue
+import sys
+import unittest
+from pathlib import Path
+from unittest.mock import patch
+
+
+ROOT = Path(__file__).resolve().parents[1]
+sys.path.insert(0, str(ROOT))
+
+from launch.core.platforms.base import CmdOutputMetadata
+from launch.core.platforms.linux import LinuxRuntime
+
+
+class LinuxOutputDrainTests(unittest.TestCase):
+    def test_large_output_scans_for_prompt_once_and_accepts_split_delimiter(self):
+        runtime = object.__new__(LinuxRuntime)
+        runtime.stopped = True
+        runtime.output_queue = queue.Queue()
+        for _ in range(256):
+            runtime.output_queue.put(b"x" * 4096)
+        runtime.output_queue.put(b'\n###PS1JSON###\n{"exit_code": "0", "username": "root"')
+        runtime.output_queue.put(
+            b', "hostname": "test", "working_dir": "/testbed", '
+            b'"py_interpreter_path": ""}\n###PS1END###\n'
+        )
+
+        original_matches = CmdOutputMetadata.matches_ps1_metadata
+        calls = 0
+
+        def counted_matches(output_text):
+            nonlocal calls
+            calls += 1
+            return original_matches(output_text)
+
+        with patch.object(
+            CmdOutputMetadata,
+            "matches_ps1_metadata",
+            new=counted_matches,
+        ):
+            output, metadata = runtime._read_raw_output(timeout=1)
+
+        self.assertEqual(calls, 1)
+        self.assertIsNotNone(metadata)
+        self.assertEqual(metadata.exit_code, 0)
+        self.assertEqual(metadata.working_dir, "/testbed")
+        self.assertTrue(output.startswith("x" * 4096))
+        self.assertEqual(output.count("x"), 256 * 4096)
+
+
+if __name__ == "__main__":
+    unittest.main()