dc85b2f21e
There have been several issues with the way in which Trove does its monkey patching including the fact that oslo.messaging gets rather annoyed if eventlet is incorrectly patched. So, this change follows the recommendations in (note multi-line URL) http://specs.openstack.org/openstack/\ openstack-specs/specs/eventlet-best-practices.html Other things to keep in mind are that the old way in which Trove was doing monkey patching was flawed. What Trove did was to first call eventlet.monkey_patch(all=True, thread=False) with the fond hope that this meant monkey patching of "all but thread" would be done. However, the documentation and the code for the function are pretty clear that this is not the case. See: http://eventlet.net/doc/basic_usage.html "If all is True, then all modules are patched regardless of the other arguments. If it’s False, then the rest of the keyword arguments control patching of specific subsections of the standard library." This also explains why debugger didn't quite work well with Trove, the whole purpose of the shenanigans that Trove was attempting to do. Then there's also the wonderful fact that you can't reliably check (with is_monkey_patched()) whether the 'thread' module is in fact monkey patched. See (note the multi-line URL) http://stackoverflow.com/questions/32452110/\ does-eventlet-do-monkey-patch-for-threading-module Therefore the code actually attempts to issue a meaningful warning if debugging is attempted with eventlet monkey patched. This is related to a number of other bugs I've found that indict the way in which projects have monkey_patched eventlet. See, for example, https://review.openstack.org/#/c/156942/, https://review.openstack.org/#/c/153699/, and https://bugs.launchpad.net/neutron/+bug/1417386. Change-Id: I92ab95f9113eae862e0cd56e7d65cdd48944615f Closes-Bug: 1365736 Closes-Bug: 1510542 |
||
|---|---|---|
| .. | ||
| __init__.py | ||
| api.py | ||
| common.py | ||
| conductor.py | ||
| fakemode.py | ||
| guest.py | ||
| manage.py | ||
| taskmanager.py | ||