ReflexiveMoveout configuration parameter
Hello,
I would like to know what is the usage of this ReflexiveMoveout parameter? We have 12 nodes version 7.2.3-10 vertica and one of the nodes's WOS keeps getting full (16GB) and starts doing WOS_SPILL which looks like it impacts whole cluster's performance. Even though i decreased both MoveOutInterval (200) and MoveOutMaxAgeTime(900) , we're still seeing one node doing too many WOS_SPILL . For permenant solution we may need to check segmentation of projections but for short term solution , I'm trying to optimize mouveou and mergeout intervals . I was looking for detailed ReflexiveMoveout explanation but could not find on documentation.Could you please give examples of use case of this parameter please ?
Thanks
0
Comments
Are you doing trickle loading? Do you use DIRECT keyword when loading data?
Hello ,
We're using both trickle and direct loading , we found our problem is TM moveout is delayed because some open sessions(SELECT 1) coming from our report servers holding Global Catalog X lock . I would still like to learn use cases of enabling ReflexiveMoveout parameter, I could not find any explanation for how does that parameter work.
ReflexiveMoveout works with MoveOutSizePct, and what it does is when you commit vertica sees if the wos meets the MoveOutSizePct, if it so, it force a moveout.
If the ReflexiveMoveout is off data is moved from the WOS when the TM runs on committed data.
Hope this helps.
Eugenia
I also recomend you check following knowledge base document on tuple mover. It does not talk about reflexiveMoveout but has useful information to solve your problem.
https://community.dev.hpe.com/t5/Vertica-Knowledge-Base/Tuple-Mover-Best-Practices/ta-p/235261