Create separate queue for state reports
Bug #1496410 reported by
Eugene Nikanorov
This bug affects 2 people
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
networking-midonet |
Fix Released
|
High
|
YAMAMOTO Takashi | ||
neutron |
Fix Released
|
Medium
|
Eugene Nikanorov |
Bug Description
In big clusters having hundreds of nodes, neutron rpc workers could be consumed by rpc requests so much that they can't process state reports from agents on time.
That lead to a condition when agents begin to "flap", appear dead and alive. This in turn causes rescheduling which loads neutron-server even more, creating self-sustaining loop.
Changed in neutron: | |
importance: | Undecided → Medium |
Changed in neutron: | |
status: | New → In Progress |
Changed in networking-midonet: | |
importance: | Undecided → High |
assignee: | nobody → YAMAMOTO Takashi (yamamoto) |
assignee: | YAMAMOTO Takashi (yamamoto) → nobody |
status: | New → Fix Committed |
tags: |
added: kilo-backport-potential removed: neutron-core |
tags: |
added: neutron-core removed: kilo-backport-potential |
tags: | added: kilo-backport-potential |
Changed in networking-midonet: | |
milestone: | none → 1.0.0 |
assignee: | nobody → YAMAMOTO Takashi (yamamoto) |
status: | Fix Committed → Fix Released |
Changed in neutron: | |
status: | Fix Committed → Fix Released |
To post a comment you must log in.
https:/ /review. openstack. org/#/c/ 226362/