get_machine_ips took too long to complete

Bug #1996966 reported by norman shen
This bug affects 1 person
Affects Status Importance Assigned to Milestone
OpenStack Compute (nova)

Bug Description


I found that get_machine_ips could took too long before returning IP addresses. There are
around 160 instances with about 200 nics which results in around 1000 network adapters on the host.
calling netifaces.ifaddresses approximately took around 0.2 ~ 0.5 seconds.

Steps to reproduce

1. use an arm64 host or host load is high
2. booting 200 instances with neutron hybird SG driver enabled
3. reboot nova-compute

Expected result

get_machine_ips should take no more than 2 seconds to return

Actual result

Took around 500 seconds

1. phyitum arm64 as compute node

Revision history for this message
Artom Lifshitz (notartom) wrote :

This ran a bell for me, so I went digging, and found [1]. I'm not sure we can make get_machine_ips() take less time with a reasonable amount of error (some sort of parallelism or scatter/gather type thing seems like overkill here), but we can probably make the _check_my_ip() check at libvirt startup optional and configurable, since it only prints a warning.


Changed in nova:
importance: Undecided → Wishlist
status: New → Confirmed
To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.