Description of problem: When requesting the cinder-volume and the cinder-api services to stop they will not do a graceful shutdown as expected, instead they will be abruptly terminated. Scheduler and backup services do not have this issue. Version-Release number of selected component (if applicable): This affects all versions since oslo service implemented graceful shutdown on v1.3.0. How reproducible: It happens every time. Steps to Reproduce: - For the API: 1. Slow down the attach process on the cinder-volume service (introducing a delay, this is only to be able to simulate the problem) 2. Send the attach request 3. systemctl stop openstack-cinder-api - For the volume: 1. Slow down the create process on the cinder-volume service (introducing a delay, this is only to be able to simulate the problem) 2. Send the volume creation request 3. systemctl stop openstack-cinder-volume Actual results: Status of the resource is attaching or creating, depending on the operation. Expected results: Status should be in-use or available, depending on the operation. Additional info: The problem is that systemd.kill is sending the SIGTERM signal to all processes instead of just sending it to the main process and letting oslo.service propagate this signal to child processes. Since scheduler and backup don't have child processes they are safe.
Proposed fix: http://review.rdoproject.org/r/3486
patch is landed on rpm-master and newton-rdo, patch for mitaka not yet landed that I could tell.
Patches landed, closing.