novaclient can get hung when used for extended periods
Bug #1323862 reported by
Robert Collins
This bug affects 1 person
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
python-keystoneclient |
Fix Released
|
Medium
|
Ian Cordasco | ||
python-novaclient |
Fix Released
|
High
|
Ian Cordasco |
Bug Description
We see a regular situation with nodepool and the HP1 region of tripleo-test-cloud where nodepool has a novaclient session open but no requests get submitted. Debugging has uncovered that when this happens there is a hung tcp connection:
- novaclient has sent a request
- the packet with the request got ACKd
- no response has been received
- tcp keepalive is disabled
if the far end has reset its connection for some reason (e.g. it got a failed packet delivery onthe response, or driver issues, or $whatever) this will never recover. We have fixed it by manually triggering a RST to the stuck connection, but it would be a lot better if persistent connections used tcp keepalive.
Changed in python-keystoneclient: | |
status: | New → In Progress |
assignee: | nobody → Ian Cordasco (icordasc) |
Changed in python-novaclient: | |
importance: | Undecided → High |
Changed in python-keystoneclient: | |
importance: | Undecided → Medium |
Changed in python-keystoneclient: | |
milestone: | none → 1.1.0 |
Changed in python-keystoneclient: | |
status: | Fix Committed → Fix Released |
Changed in python-novaclient: | |
milestone: | none → 2.21.0 |
Changed in python-novaclient: | |
status: | Fix Committed → Fix Released |
To post a comment you must log in.
This sounds like it may be a bug in one of the libraries we are using, or perhaps how we are using the libraries. Perhaps related to the requests library.