bit vector + backoff timer + simpler implementation for faster slab reassignment#542
bit vector + backoff timer + simpler implementation for faster slab reassignment#542byrnedj wants to merge 3 commits intomemcached:nextfrom
Conversation
dormando
left a comment
dormando
left a comment
There was a problem hiding this comment.
so much simpler and covers so many more improvements :)
|
Thanks! Great timing, I was just about to sit down and give this a whack myself. How did this look in your tests vs your previous patch? |
|
It reassigns slabs at nearly the same speed. Which for all practical purposes is fast enough on our end (a few hundred GET requests for a few hundred slabs). Ran mc-crusher overnight, no issues so far. |
|
hmm I really want to include this in the next release but I haven't spent much time with it. It doesn't seem like anyone else has looked at it or tried it on a shadow/production workload either. This might miss the next release, but they're coming pretty frequently right now. |
|
Sounds good, let me know if there are any other issues. |
|
Fixed a couple minor nits and wrote a real commit message. It's now merged for the next release. Think the 'next' branch will end up baking for a few days or a week, just doing a merge-down while I have a chance. |
|
Out in 1.5.20 |
2 participants
This PR addresses the implementation details for faster slab reassignment and the suggestions made by dormando. See #524 for more details.