---Multiraft-problem has been found Shall we consider having ozone recommend leader for Ratis group and we have some kind of load balance algorithm for Ratis data pipeline? There was JIRA for this: https://issues.apache.org/jira/browse/HDDS-2922
-Li On 2020/5/12, 7:48 PM, "Elek, Marton" <[email protected]> wrote: * Jitendra: Multiraft-problem has been found --> In case of multi-raft, multiple leader on the some Datanode can cause performance issues --> Is randomization enough after some problem, during re-election? --> Is it possible to write via a follower? --> Multiple SSD can be used for Raft logs (can be useful for for Multi-raft) --> Shashi: we see performance problems when tests run with only a few clients (under investigation) --> flush is one problem (we have stronger guarantees) --> We need to check if sync is required for Raft log. --> Can we do it with increase the leader election timeout on all datanodes which already have too many leader roles? * Uma: ViewFS status --> First patch is committed, on the trunk, can be tested * Marton: Move to TLP? --> Discussion about moving Ozone to a top level Apache project --> Process: --> Discuss on *[email protected] (the fact to move) --> Write proposal (with full details, list of committers, PMC, etc.) --> Check naming / trademark issues with LEGAL --> Discuss the proposal on *[email protected] --> Vote --> Submit it to the Board --> As with submarine: Opt-in committership to any Hadoop committers --> There were no arguments against to start the discussion --> We agreed that diversity in contributors is one important aspect which can/should be improved --> Next steps: start discussion on Hadoop mailing lists Marton --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
