guest-diag: read the trace snapshot non-blocking (no stuck tail processes)

1 files changed, 33 insertions, 2 deletions

diff --git a/tools/guest-diag/guest_pin_stream.py b/tools/guest-diag/guest_pin_stream.py
index c0cf485..97c68f2 100644
--- a/tools/guest-diag/guest_pin_stream.py
+++ b/tools/guest-diag/guest_pin_stream.py
@@ -807,6 +807,38 @@ def resolve_channel_pins(kc, ch, out):
 
 
 
+def _read_trace_snapshot(max_bytes=1 << 20, budget=0.5):
+    """Best-effort non-blocking read of the trace ring file.
+
+    Spawning `tail` here left uninterruptible D-state processes when the
+    seq_file read blocked under load (observed r127: dozens of stuck tails,
+    load 26); a direct os.read with O_NONBLOCK returns whatever is ready and
+    never accumulates processes.
+    """
+    try:
+        fd = os.open(TRACE, os.O_RDONLY | os.O_NONBLOCK)
+    except OSError:
+        return b""
+    chunks = []
+    got = 0
+    t0 = time.time()
+    try:
+        while got < (8 << 20) and time.time() - t0 < budget:
+            try:
+                d = os.read(fd, 262144)
+            except BlockingIOError:
+                break
+            except OSError:
+                break
+            if not d:
+                break
+            chunks.append(d)
+            got += len(d)
+    finally:
+        os.close(fd)
+    return b"".join(chunks)[-max_bytes:]
+
+
 def seed_events_from_trace(seen_sems, seen_pbs, seen_chs):
     """One-shot tail scan of the trace ring after a pid switch / start.
 
@@ -821,8 +853,7 @@ def seed_events_from_trace(seen_sems, seen_pbs, seen_chs):
     (the mem candidates are scored and deferred).
     """
     try:
-        out = subprocess.run(["tail", "-c", "1048576", TRACE],
-                             capture_output=True, timeout=10).stdout
+        out = _read_trace_snapshot()
     except Exception:
         return 0
     n = 0