We should probably store this data in a separate table and query all the rows that apply to a particular server from the DB when we need it, rather than combine them all into a single blob that then causes contention issues. (This might also solve various issues we have had with updating metadata while a resource is locked.)
In the short term though, Jan came up with this workaround which was fairly effective in spreading out the thundering herd:
We should probably store this data in a separate table and query all the rows that apply to a particular server from the DB when we need it, rather than combine them all into a single blob that then causes contention issues. (This might also solve various issues we have had with updating metadata while a resource is locked.)
In the short term though, Jan came up with this workaround which was fairly effective in spreading out the thundering herd:
delete_ delay_time: :RandomString _classes: :TestResource wait_secs: bastion_ node_cleanup
type: OS::Heat:
properties:
character
- min: 2
class: digits
length: 2
delete_delay:
type: OS::Heat:
properties:
action_
delete: {get_attr: [delete_delay_time, value]}
depends_on: deployment_