Increasing number of tokens will make repairs worse not better. You can just split the sub ranges into smaller chunks, you dont need to use vnodes to do that. Simple approach is to iterate through each host token range and split by N and repair them (ie https://github.com/onzra/cassandra_range_repair) To be more efficient you can grab ranges and split based on number of partitions in the range (ie fetch system.size_estimates and walk that) so you dont split empty or small ranges a ton unnecessarily, and because not all tables have some fixed N that is efficient.
Using TLPs reaper https://github.com/thelastpickle/cassandra-reaper or DataStax OpsCenter's repair service is easiest solution without a lot of effort. Repairs are hard. Chris On Wed, Oct 4, 2017 at 11:48 AM, Jeff Jirsa <jji...@gmail.com> wrote: > You don't need to change the number of vnodes, you can manually select > CONTAINED token subranges and pass in -st and -et (just try to pick a > number > 2^20 that is fully contained by at least one vnode). > > > > > On Wed, Oct 4, 2017 at 9:46 AM, Mohapatra, Kishore < > kishore.mohapa...@nuance.com> wrote: > >> Hi, >> >> We are having a lot of problems in repair process. We use sub >> range repair. But most of the time, some ranges fails with streaming error >> or some other kind of error. >> >> So wondering if it will help if we increase the no. of VNodes from 256 >> (default) to 512. But increasing the VNodes will be a lot of efforts, as it >> involves wiping out the data and bootstrapping. >> >> So is there any other way of splitting the range into small ranges ? >> >> >> >> We are using version 2.1.15.4 at the moment. >> >> >> >> Thanks >> >> >> >> *Kishore Mohapatra* >> >> Principal Operations DBA >> >> Seattle, WA >> >> Email : kishore.mohapa...@nuance.com >> >> >> >> >> > >
