Absurdly high glance haproxy timeout (6 hours)
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
kolla-ansible |
Triaged
|
Medium
|
Unassigned |
Bug Description
As in the summary: haproxy has both client and server timeout for glance default to 6 hours.
These timeouts are for inactivity periods, i.e. 6 hours of no traffic from either side.
It was originally proposed as a fix for glance timeouts on long transfers [1] but I could not confirm it happening: testing with timeouts as low as 1 second works just fine even if the upload takes an hour.
Such long timeouts are generally discouraged because they may create lots of unclosed or half-closed connections laying around. Also, many Glance actions are not lengthy in any sense - requests asking for image listing or image details laying around for 6 hours sound silly. :-)
I argue that the 6 hours timeout is not a fix but a bug now, i.e. it creates new issues and hides other issues - if you have uploads having 6 hours of *inactivity*, you are having issues with storage.
It's worth noting that there are undocumented flags that you can add to globals.yaml to control the timeout: glance_ api_client_ timeout: "1m" glance_ api_server_ timeout: "1m"
haproxy_
haproxy_