In our production swift cluster, we have container sync middleware enabled in the pipeline. But swift container sync is not configured.
We noticed 100% CPU utilization and a lot of read/write i/os from the swift-container-sync process on our meta data nodes recently after we have seen increasing volumes of POST/DELETE requests. From the code it makes sense since container_sync() calls broker.get_info() which will merge the pending requests to the container db before it gets the metadata from the container_stat table. I have two questions: 1. Is it safe just to stop the swift-container-sync process on all of our metadata nodes(have all the account/container daemons running)? Any side effect if we do so? 2. Since container_updater is doing the same broker.get_info() in the container_sweep() code path, and the concurrency is 4, why don’t I see swift-container-updater sitting on the top CPU consumer list? My assumption is due to each swift-container-updater child process is forked to handle only one partition per device. So it is a short lived process which won’t be captured by atop or top. Is that true? Thanks. Alan Jiang __________________________________________________________________________ OpenStack Development Mailing List (not for usage questions) Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
