element.py: Handle log flushing at termination more carefully

If the child process is in the middle of using the I/O stack
in python while trying to flush the python file handle, this is non
reentrant and will raise an exception.

Handle this exception and simply flush the underlying file
descriptor instead.

1 files changed, 11 insertions, 3 deletions

diff --git a/buildstream/element.py b/buildstream/element.py
index 332cb0fc5..46422b44c 100644
--- a/buildstream/element.py
+++ b/buildstream/element.py
@@ -749,9 +749,17 @@ class Element(Plugin):
 
             # Write one last line to the log and flush it to disk
             def flush_log():
-                logfile.write('\n\nAction {} for element {} forcefully terminated\n'
-                              .format(action_name, self.name))
-                logfile.flush()
+
+                # If the process currently had something happening in the I/O stack
+                # then trying to reenter the I/O stack will fire a runtime error.
+                #
+                # So just try to flush as well as we can at SIGTERM time
+                try:
+                    logfile.write('\n\nAction {} for element {} forcefully terminated\n'
+                                  .format(action_name, self.name))
+                    logfile.flush()
+                except RuntimeError:
+                    os.fsync(logfile.fileno())
 
             self._set_log_handle(logfile)
             with _signals.terminator(flush_log):