Difference between revisions of "Swift/ideas"
< Swift
| Line 13: | Line 13: | ||
* in the situation where a proxy needs to talk to a storage node on the same server, call the storage node method(s) directly instead of putting bytes on the network (somewhat in progress) | * in the situation where a proxy needs to talk to a storage node on the same server, call the storage node method(s) directly instead of putting bytes on the network (somewhat in progress) | ||
* skeleton middleware, well-commented | * skeleton middleware, well-commented | ||
| − | * fsync directories (talk to portante or notmyname in IRC about this) | + | * fsync directories (talk to portante or notmyname in IRC about this) - https://review.openstack.org/#/c/126923/ |
* DiskFile entry point | * DiskFile entry point | ||
* policy descriptions, exactly like how https://review.openstack.org/#/c/120424/ is not implemented (see the review comments) | * policy descriptions, exactly like how https://review.openstack.org/#/c/120424/ is not implemented (see the review comments) | ||
Revision as of 16:23, 8 October 2014
Ideas for OpenStack Swift
- use keystone api in swift client, if possible
- change default ports - https://review.openstack.org/#/c/118200/
- reverse listings - https://review.openstack.org/#/c/120709/
- stdin streaming to swiftclient
- ring validator (in swift-recon): eg test that something is running on the ports in the ring -- not just "something", must verify correct type, because deployers sometimes flip a/c/o and then inexplicable 400 happens -- https://wiki.openstack.org/wiki/OutreachProgramForWomen/Ideas#Swift_-_storage_server_OPTIONS_support_and_checker_tool
- deep health check to test all the way to drives (? or at least storage servers)
- better ring deployment inside of swift itself
- need internal net sec enforcement - https://blueprints.launchpad.net/swift/+spec/secure-internal-network-requests
- change the ssync (object-replicator) design to be run as another process for concurrency
- durability simulator based on swift specification (e.g. # of replicas, # of devices and network speed to replicate) (inprogress by Christian)
- in the situation where a proxy needs to talk to a storage node on the same server, call the storage node method(s) directly instead of putting bytes on the network (somewhat in progress)
- skeleton middleware, well-commented
- fsync directories (talk to portante or notmyname in IRC about this) - https://review.openstack.org/#/c/126923/
- DiskFile entry point
- policy descriptions, exactly like how https://review.openstack.org/#/c/120424/ is not implemented (see the review comments)
