| opendevreview | Pranali Deore proposed openstack/glance master: Add Location Import task flow https://review.opendev.org/c/openstack/glance/+/886749 | 09:29 |
|---|---|---|
| opendevreview | Pranali Deore proposed openstack/glance master: Add new add location api https://review.opendev.org/c/openstack/glance/+/881940 | 09:29 |
| opendevreview | Pranali Deore proposed openstack/glance master: Add functional tests for new add-location API https://review.opendev.org/c/openstack/glance/+/886947 | 09:29 |
| opendevreview | Pranali Deore proposed openstack/glance master: Add new get location api https://review.opendev.org/c/openstack/glance/+/882498 | 09:29 |
| opendevreview | Pranali Deore proposed openstack/glance master: Add api-ref for new location apis https://review.opendev.org/c/openstack/glance/+/890508 | 09:29 |
| opendevreview | Pranali Deore proposed openstack/glance master: Bump Images API version to 2.17 https://review.opendev.org/c/openstack/glance/+/890509 | 09:29 |
| opendevreview | Pranali Deore proposed openstack/python-glanceclient master: WIP Add support for new location APIs https://review.opendev.org/c/openstack/python-glanceclient/+/890634 | 12:33 |
| pdeore | dansmith, I've updated the client patch for new add location api & works for nova & cinder, could you please have a look https://review.opendev.org/c/openstack/python-glanceclient/+/890634 | 13:40 |
| dansmith | pdeore: I don't see a recheck on my nova test patch, do you have one of your own I should be looking at? | 13:44 |
| dansmith | or did you mean you just tested locally? | 13:44 |
| pdeore | dansmith, yeah i tested locally, but i will add recheck on your nova patch | 13:45 |
| dansmith | I can do it | 13:45 |
| pdeore | ack | 13:46 |
| mnaser | dansmith: i know you were playing around with this for a while.. but how hard would it be to make changes to interop image import so that it uses a glance store rather than local disk? | 17:26 |
| mnaser | like in a case where you have a giant ceph cluster connected to glance.. and you have glance-api running on a system with not a lot of resources.. | 17:26 |
| dansmith | mnaser: not possible | 18:14 |
| dansmith | mnaser: import requires running commands against the file in staging | 18:15 |
| dansmith | it's the fundamental flaw in the design, IMHO, and why I made things like my format inspector work on the stream, so it works for upload and import | 18:15 |
| mnaser | dansmith: blargh, well i think i was approaching it from a rbd pov where it could be possible to do a bunch of things like qemu-img convert directly rbd..rbd | 18:16 |
| dansmith | mnaser: if you do a local host mount then it's the same of course | 18:16 |
| dansmith | mnaser: qemu-img does not, TMK, support working against a remote target like an rbd url or anything | 18:17 |
| mnaser | dansmith: it does, for rbd at least :D https://docs.ceph.com/en/reef/rbd/qemu-rbd/ | 18:17 |
| dansmith | oh okay well, that's news to me then | 18:18 |
| dansmith | mnaser: that example shows converting a file format from disk to rbd:// not rbd:// to rbd:// so I dunno if that would also work or not | 18:19 |
| dansmith | (or be terrible performance-wise) | 18:19 |
| mnaser | dansmith: i've done it in the past personally, it will have to 'passthrough' the qemu-img process of course | 18:20 |
| mnaser | so you've got to do read and writes | 18:20 |
| mnaser | but then i guess the argument could be that we mount a big volume from ceph for every glance-api and use that instead | 18:20 |
| dansmith | right, but for the same reason that using qcow2 on rbd is slow, I would expect that conversion to also be slow | 18:20 |
| dansmith | yeah, so I guess if qemu-img can do this then we could special-case it for rbd, but I think there'd need to be substantial evidence that it's worth that trouble over just doing a host mount | 18:21 |
| mnaser | yeah im just thinking that will increase traffic a lot more cause we have to do a write of the whole thing the first time its streamed, then another full read and full write on the convert | 18:22 |
| mnaser | but i guess if its a host mount then yeah its also gonna be a full read and write over the network if you dont have a lot of scratch disk space | 18:22 |
| mnaser | its an annoying problem to solve :< | 18:22 |
| dansmith | well, that's how staging is supposed to work, so it'd really be a pretty big change if we somehow didn't do that | 18:24 |
| dansmith | unless you had a direct rbd import mechanism that took an rbd url or uuid and did the conversion from there to the destination, | 18:24 |
| dansmith | but that's a lot of new stuff and rules that aren't there currently | 18:25 |
| dansmith | even web-download grabs it locally and then pushes it to the destination store | 18:25 |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!