instances with hugepages intermittently fail to start

[James Page]

Ubuntu 18.04 LTS + OpenStack Queens

Example flavour:

+----------------------------+--------------------------------------+
| Field | Value |
+----------------------------+--------------------------------------+
| OS-FLV-DISABLED:disabled | False |
| OS-FLV-EXT-DATA:ephemeral | 0 |
| access_project_ids | None |
| disk | 0 |
| id | 9f3946cc-df0c-468a-80a9-808702d9a620 |
| name | 4c-8192m |
| os-flavor-access:is_public | True |
| properties | hw:mem_page_size='large' |
| ram | 8192 |
| rxtx_factor | 1.0 |
| swap | |
| vcpus | 4 |
+----------------------------+--------------------------------------+

*some* instance creation requests fail to start instances due to lack of available hugepages on the numa node being targeted for the instance (log data to follow). Workaround to alleviate this symptom has been to split the flavor across to numa nodes using hw:numa_nodes=2 as an additional property, which means the cores for the instance can be split across the numa nodes on the hypervisor. This does have a performance overhead.

[James Page (james-page)]

This may be related to bug 1734204 but right now I can't actually reproduce that bug on a Queens deployment

[Andre Ruiz (andre-ruiz)]

Have been trying to reproduce it, nothing so far:

https://pastebin.canonical.com/p/Wq7bnBS8cN/

There must be some other variable involved.

[Launchpad Janitor (janitor)]

[Expired for nova (Ubuntu) because there has been no activity for 60 days.]