Hi @Gonzalo Herreros, You are right, there is a deadlock between threads 139795204400896 and 139794642351872 in your dump. One of the locks is the Ambari-specific lock added as an attempt to workaround threading issues in the subprocess module that we have run into. The other one is required for imports. It turns out that os.fork tries to acquire the import lock. Thread 139795204400896 calls subprocess.Popen from a module being imported. Lock order is: import, subprocess. Thread 139794642351872 simply calls subprocess.Popen, which calls os.fork. Lock order is: subprocess, import. We are now dropping the custom lock in favor of the subprocess32 module, which is the backport of Python 3.2 subprocess implementation, and is recommended even in the Python 2 docs. Thanks for reporting this, and sorry for taking so long to answer.
... View more
