bitbake-worker: Try and avoid potential short write events issues

We're seeing occasional issues where builds fail as events were written from the
worker children in the form <event>partial data<event>full event</event>.

This causes failures as bitbake server can't parse that and exits. This could
be due to short writes to the worker event pipe which we weren't checking. Check
this and loop accordingly. Also add some asserts to detect other potential causes.

Thanks to Joshua Watt for help in spotting the issue.

[YOCTO #14181]

Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>

1 files changed, 6 insertions, 2 deletions

diff --git a/bin/bitbake-worker b/bin/bitbake-worker
index 9334f11fb..40da7a0c2 100755
--- a/bin/bitbake-worker
+++ b/bin/bitbake-worker
@@ -118,7 +118,9 @@ def worker_child_fire(event, d):
     data = b"<event>" + pickle.dumps(event) + b"</event>"
     try:
         worker_pipe_lock.acquire()
-        worker_pipe.write(data)
+        while(len(data)):
+            written = worker_pipe.write(data)
+            data = data[written:]
         worker_pipe_lock.release()
     except IOError:
         sigterm_handler(None, None)
@@ -321,7 +323,9 @@ class runQueueWorkerPipe():
         end = len(self.queue)
         index = self.queue.find(b"</event>")
         while index != -1:
-            worker_fire_prepickled(self.queue[:index+8])
+            msg = self.queue[:index+8]
+            assert msg.startswith(b"<event>") and msg.count(b"<event>") == 1
+            worker_fire_prepickled(msg)
             self.queue = self.queue[index+8:]
             index = self.queue.find(b"</event>")
         return (end > start)