Project

General

Profile

Rbd - cloud management platform features » History » Version 2

Jessica Mack, 06/22/2015 12:24 AM

1 1 Jessica Mack
h1. Rbd - cloud management platform features
2
3
h3. Summary
4
5
This blueprint is a placeholder for updating various cloud management platforms (OpenStack, Cloudstack, etc.) for better ceph integration.
6
7
h3. Owners
8
9
* Name (Affiliation)
10
* Name (Affiliation)
11
* Name
12
13
h3. Interested Parties
14
15
* Daniele Stroppa (ZHAW)
16
* Sébastien Han (eNovance)
17
* Josh Durgin
18
* Name
19
20
h3. Current Status
21
22
There has been a large amount of work integrating ceph (and rbd in particular) with various cloud management platforms. What features are in some but not others? What features are missing from all of them, or need better documentation or ease of use?       
23
24
h3. Detailed Description
25
26
h4. All
27
28
* flatten automatically when clone chains are too long
29
* flatten children automatically when deleting a parent snapshot with fewer than N children (otherwise flatten and delete the parent snapshot (and image if it's supposedly deleted) when fewer than N children remain).
30
* taking advantage of new ceph features
31 2 Jessica Mack
** namespace support (see [[rbd - namespace support]])
32 1 Jessica Mack
*** create rados client for each tenant and put new volumes in namespace named after tenant
33
*** pass tenant id and key to compute node
34
*** maybe store tenant's keys in separate service (Barbican for OpenStack?)
35
** configuring localized or balanced reads
36
** configuring local shared cache per compute node (maybe this is just running a daemon, with configuration for guests to use it)
37
** adjusting write-back cache parameters (size etc.)
38
** configuring copy-on-read
39
40
h4. OpenStack
41
42
* once auto-flattening is in place, clone_volume in cinder can be implemented with an rbd clone instead of the plain copy it's currently using
43
44
h4. CloudStack
45
46
* snapshot and create_from_snapshot (clone) support
47
* backing up to rbd (using diffs between snapshots as a bonus)
48
* configuring rbd caching
49
50
h3. Work items
51
52
h4. Coding tasks
53
54
# OpenStack: implement direct url for RBD in nova while booting (ie: https://review.openstack.org/#/c/37817/) (leseb)
55
# OpenStack: implement RBD stripe_unit, stripe_count and rbd_chunk_size with Cinder (I already coded this, I just need some tests then I'll push it) (leseb)
56
# OpenStack: Ability to store snapshots and images to different places (ie: https://blueprints.launchpad.net/glance/+spec/ability-to-separate-snapshots-and-images); Glance's folks where quite entousiastic about this, but weren't sure if it should be either in Glance or Nova. (leseb) It seems like Glance is missing a level of indirection between users and backend stores (joshd)
57
# OpenStack: Join the LFS initiative to build a transparent fronted API for Object Stores (leseb) => rgw see http://www.mail-archive.com/ceph-devel@vger.kernel.org/msg15479.html
58
# OpenStack: hopefully this code https://review.openstack.org/#/c/36042/ will land into Havana, then as soon as it's implement. The depreciation work must start by joining the 'brick' implementation. (leseb)
59
# OpenStack: enhance user experience by allowing to set the body of the RBD image. Not sure if makes sense right now, we better discuss that. It's definitely end-user oriented (ex: cinder create --metadata stripe_count=16 stripe_unit=65536 --display-name=tiny-ios 10).(leseb)
60
61
h4. Build / release tasks
62
63
# Task 1
64
# Task 2
65
# Task 3
66
67
h4. Documentation tasks
68
69
# OpenStack: the Havana cycle brought some changes and made configuration easier, this has to be documented. (leseb)
70
# Task 2
71
# Task 3
72 2 Jessica Mack
73 1 Jessica Mack
h4. Deprecation tasks
74
75
# Task 1
76
# Task 2
77
# Task 3