geode-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Mario Ivanac <mario.iva...@est.tech>
Subject Re: Server recovery severely degrades client read traffic
Date Tue, 06 Aug 2019 07:54:44 GMT
Hi,


we see problem in BucketAdvisor.getPreferredNode, in random selection of member:


// Pick one at random.
int i = myRand.nextInt(locProfiles.length);
return locProfiles[i].peerMemberId;



BR


________________________________
Šalje: Anthony Baker <abaker@pivotal.io>
Poslano: 2. kolovoza 2019. 19:43:18
Prima: dev@geode.apache.org <dev@geode.apache.org>
Predmet: Re: Server recovery severely degrades client read traffic

Interesting find!  Can you share the code path you’re looking at?  I see one related to
putAll but not for get.  Thanks!

Anthony


> On Aug 1, 2019, at 11:01 PM, Mario Ivanac <mario.ivanac@est.tech> wrote:
>
> Hi,
>
> we are observing severe throttling from the cluster when getting data from a partitioned
region (no SH nor TX) while server hosting one of the redundant buckets is recovering (see
ticket https://issues.apache.org/jira/browse/GEODE-7039).
>
>
> Currently, Get operations that have not landed on a server hosting the bucket will be
proxied to other members that host bucket, in random fashion.
>
> Question is, can we use bucketProfile.isInitializing flag for member selection algorithm,
or do you have some better idea?


Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message