Difference between revisions of "Meetings/Swift"
< Meetings
Line 25: | Line 25: | ||
*** https://bugs.launchpad.net/swift/+bug/1277223 | *** https://bugs.launchpad.net/swift/+bug/1277223 | ||
*** https://bugs.launchpad.net/swift/+bug/1068426 | *** https://bugs.launchpad.net/swift/+bug/1068426 | ||
+ | ** Suggested patch breakdown | ||
+ | *** patch 1: split ContainerSync to two classes : (1) manager class that iterates over dbs and assigns each db to a different process (2) a sync class that does the sync work on over a db | ||
+ | *** patch 2: Introduce the process pool, and the SyncStats class that collects stats from all processes allowing the existing one aggregated report. | ||
+ | *** patch 3: Add thread pool to the db sync process | ||
+ | *** patch 4: Add the head command to the phase where the pointers are not equal, greatly mitigating the problem where each object is sent 4 times. | ||
---- | ---- | ||
Revision as of 14:59, 15 July 2015
Meeting Time: Every Wednesday at 21:00 UTC in #openstack-meeting Meeting Logs: http://eavesdrop.openstack.org/meetings/swift/2015/
Agenda: July 15, 2015 2100 UTC
- Hackathon reminder (https://www.eventbrite.com/e/swift-hackathon-tickets-17308818141)
- Python3 work: https://review.openstack.org/#/q/status:open+project:openstack/swift+branch:master+topic:py3,n,z
- Option 1: Do what we're currently doing. big patches, hard to review, little immediate payoff
- Option 2: Patch one module at a time and blacklist non-ported modules in the gate jobs. smaller pieces of work, measurable progress, still will take a while
- Option 3: Freeze the world, port everything, done. painful up front, but done all at once
- updates from ongoing work
- encryption, EC work, hummingbird?
- notification middleware [cschwede]
- Hot Bugs [notmyname]
- EC bugs
- https://bugs.launchpad.net/swift/+bugs?field.tag=ec (needs triage)
- swiftclient bugs
- https://goo.gl/uO4b7l ("new" bugs that need to be looked at)
- EC bugs
- Specs that are open and should land? [notmyname]
- Container Sync performance. [eranrom]
- short deck on the what and why: http://www.slideshare.net/DmitrySotnikov1/container-sync-performanceimprovement
- code reference illustrating the approach (should be broken to smaller patches based on master, imho, more readable this way...): https://github.com/eranrom/swift-thoughts/blob/master/container-sync/sync.py
- Addressing existing bugs:
- Suggested patch breakdown
- patch 1: split ContainerSync to two classes : (1) manager class that iterates over dbs and assigns each db to a different process (2) a sync class that does the sync work on over a db
- patch 2: Introduce the process pool, and the SyncStats class that collects stats from all processes allowing the existing one aggregated report.
- patch 3: Add thread pool to the db sync process
- patch 4: Add the head command to the phase where the pointers are not equal, greatly mitigating the problem where each object is sent 4 times.
When adding an item, please include your IRC nickname with it.