)]}'
{"/PATCHSET_LEVEL":[{"author":{"_account_id":28619,"name":"Dmitriy Rabotyagov","email":"noonedeadpunk@gmail.com","username":"noonedeadpunk"},"change_message_id":"e64736b027aa67162be1180900e654b83c616ca2","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":1,"id":"78a2cdfe_749e756a","updated":"2023-07-18 07:54:34.000000000","message":"Ok, thanks for explanation, I indeed used RBD for testing. Now that makes way more sense to me.","commit_id":"3c014fb2b68db0fbcc4869370f75223f18cbc039"},{"author":{"_account_id":34150,"name":"Simon Hensel","email":"simon.hensel@inovex.de","username":"shensel"},"change_message_id":"f1584f3a3fbf9aa472d74b5826a12d0c05638da8","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":1,"id":"52bc1d99_a97d1c17","updated":"2023-07-10 11:44:20.000000000","message":"This change is a follow-up of the conversation at the cinder meeting:\nhttps://meetings.opendev.org/meetings/cinder/2023/cinder.2023-04-05-14.00.log.html#l-64\n\nAs mentioned in https://review.opendev.org/c/openstack/tripleo-common/+/845807, there are significant cuts in memory usage with both test and real deployments.\n\nThe performance improvements are most likely driver dependent and should only affect the `cinder-backup` and `cinder-volume` services.\nPerhaps @Gorka Eguileor could provide more insights on that topic.","commit_id":"3c014fb2b68db0fbcc4869370f75223f18cbc039"},{"author":{"_account_id":28619,"name":"Dmitriy Rabotyagov","email":"noonedeadpunk@gmail.com","username":"noonedeadpunk"},"change_message_id":"cb641b1b8f9344e23886b70556b5ad76bd12f19d","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":1,"id":"e87d1be2_509fd70a","updated":"2023-07-10 09:46:52.000000000","message":"We\u0027ve already tried out this approach like that: https://review.opendev.org/c/openstack/openstack-ansible/+/846228\n\nBut we failed to see good improvement in CI. Do you have any data you can share that will show resource consumption improvement?","commit_id":"3c014fb2b68db0fbcc4869370f75223f18cbc039"},{"author":{"_account_id":32755,"name":"Christian Rohmann","email":"christian.rohmann@inovex.de","username":"frittentheke"},"change_message_id":"74b3c9d37628a592ae4fc0ced54ef9be42b7fae0","unresolved":true,"context_lines":[],"source_content_type":"","patch_set":1,"id":"ead55e33_9450cbe1","in_reply_to":"0a8cea36_700f71fd","updated":"2023-07-11 14:28:03.000000000","message":"\u003e Well, I\u0027m pretty aware of the topic overall, it\u0027s just I failed to see difference when I tried to reproduce the issue on my own with OSA. So I am rather asking if that helped you indeed, rather it is claimed to help in devstack.\n\u003e \n\u003e As multiple things that were merged in devstack for reducing ram consumption had no effect in CI or in our POC deployments. Maybe service needs some time/load before change is visible, ofc, so I\u0027m more wondering if you\u0027ve witnessed any effect of that change on your own.\n\nAs Simon mentioned, I believe the issue depends on the (backup) driver used.\nIf RBD is used the backups are simply piped as RBD diffs and there is no volume data handled by Python directly. But drivers using the chunked driver as base (https://github.com/openstack/cinder/blob/master/cinder/backup/chunkeddriver.py), like all the object storage backends like S3, GCS, ... or I believe NFS even, will pump all the volume data as chunks through cinder backup and compress it and create checksums. This then indeed causes at least the cinder-backup memory footprint to grows massively.","commit_id":"3c014fb2b68db0fbcc4869370f75223f18cbc039"},{"author":{"_account_id":28619,"name":"Dmitriy Rabotyagov","email":"noonedeadpunk@gmail.com","username":"noonedeadpunk"},"change_message_id":"009f109731f166954551b8b8ec9c4e9cc0ee81a3","unresolved":true,"context_lines":[],"source_content_type":"","patch_set":1,"id":"0a8cea36_700f71fd","in_reply_to":"52bc1d99_a97d1c17","updated":"2023-07-11 12:16:22.000000000","message":"Well, I\u0027m pretty aware of the topic overall, it\u0027s just I failed to see difference when I tried to reproduce the issue on my own with OSA. So I am rather asking if that helped you indeed, rather it is claimed to help in devstack.\n\nAs multiple things that were merged in devstack for reducing ram consumption had no effect in CI or in our POC deployments. Maybe service needs some time/load before change is visible, ofc, so I\u0027m more wondering if you\u0027ve witnessed any effect of that change on your own.","commit_id":"3c014fb2b68db0fbcc4869370f75223f18cbc039"}]}