Skip to content
This repository has been archived by the owner on Feb 21, 2024. It is now read-only.

MaxShard value incorrect on cluster startup #916

Closed
raskle opened this issue Oct 31, 2017 · 7 comments
Closed

MaxShard value incorrect on cluster startup #916

raskle opened this issue Oct 31, 2017 · 7 comments
Assignees
Labels

Comments

@raskle
Copy link
Contributor

raskle commented Oct 31, 2017

Description

I'm proposing that we store max slice and inverse max slice with the Index .meta file.
Currently thecluster wide maxslice value is determined via gossip. During startup each node's default state is the max slice it contains, and as we grow to more slices there is larger dispersion of slices per node. Over several gossip iterations the nodes will coalesce on the correct value, but the time to achieve this grows with the number nodes, number of slices, and the memberlistConfig.PushPullInterval. I would like to see the cluster at startup achieve agreement on this value sooner.

Success criteria (What criteria will consider this ticket closeable?)

Max Slice is stored and retrievable from the index .meta file. Cluster starts up immediately with correct max slice value.

@codysoyland codysoyland changed the title Store Max Slice on disk MaxShard value incorrect on cluster startup Aug 9, 2018
@codysoyland
Copy link
Contributor

Still running into this problem. I clarified the title and tagged as a bug which I think is more accurate. We should get this fixed soon.

@travisturner
Copy link
Member

@codysoyland I believe the work currently being done in #593 will address this, but we should confirm once that's in review.

@jaffee
Copy link
Member

jaffee commented Aug 9, 2018

@travisturner how does that fit together with #1008 - are these three tickets basically all the same?

@travisturner
Copy link
Member

@jaffee I believe they're all basically the same thing. #1008 seems to be a summary of the issue and represents the current work better than just #593. This issue (#916) is a specific part of the implementation.

@codysoyland
Copy link
Contributor

Fixed in #1600

@travisturner
Copy link
Member

@codysoyland I left this open because I wasn't 100% sure that #1600 addressed this. Can you make sure that there's a test covering this case?

@travisturner travisturner reopened this Sep 10, 2018
@travisturner
Copy link
Member

fixed by #1663

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
Projects
None yet
Development

No branches or pull requests

5 participants