horizon image upload failing (multible issues)
Bug #1971179 reported by
Alexander Binzxxxxxx
This bug affects 1 person
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
OpenStack-Ansible |
Fix Released
|
High
|
Unassigned |
Bug Description
Using xena with haproxy.
Image upload via horizon fails due to multiple issues:
1) horizon upload mode direct (default with the playbook):
* CSP headers do not allow upload (only allowing from local/over haproxy) by default
* something with CORS is also wrong
2) horizon upload mode legacy:
* fails with the issue here: https:/
simple workagound for legacy mode is haproxy "mode tcp" for glance.
If you try fixing http proxy mode also have a look here for some buffer size issue: https:/
Also note, that CSP headers should include a hostname/fqdn and not just IP.
tags: | added: glance haproxy horizon |
description: | updated |
description: | updated |
Changed in openstack-ansible: | |
status: | New → Confirmed |
importance: | Undecided → High |
Changed in openstack-ansible: | |
status: | Confirmed → Fix Released |
To post a comment you must log in.
For legacy mode and https:/ /bugs.launchpad .net/glance/ +bug/1916482 - disabling uwsgi for glance should work nicely. For that you can simply set `glance_use_uwsgi: false` in user_variables.
Out of the tests we had, we spotted issue with python- glanceclient, that does not handle chunking. Issue doesn't exist with openstacksdk/ python- openstackclient . They have quite different implementations and way of image upload. I can imagine, that glanceclient just missing chunking implementation, that causes the issue.
Regarding CSP/CORS I guess there's a valid points for fixing that. Eventually we suppose that outside of AIO, external_ lb_vip_ address as FQDN if you intend to access cluster through domain name and not IP address. Then CORS would be set correctly as well as all public endpoints would be created with same fqdn in them. Otherwise clients would still access endpoints by IP and not by domain.