apache kafka - MirrorMaker Cross DC Deployment Setup -
i have question regarding mirrormaker. see in official documentation (see https://cwiki.apache.org/confluence/pages/viewpage.action?pageid=27846330), mirrormaker placed on destination data center. there reason why can't placed on source data center instead?
i'm asking because in our current setup, mirrormaker placed on destination data center , seeing owning partitions on startup or rebalance takes long time.
our setup so:
- 6 total kafka clusters across 4 data centers
- each data centers has local kafka cluster
- 2 aggregate kafka clusters in 2 of data centers
- 400+ topics in each local kafka cluster
- 2 groups of mirrormakers on same data center respective destination aggregate kafka clusters consumes 4 local kafka clusters
- mirrormaker instances configured pull 400+ topics with catch regex 4 local kafka clusters aggregate clusters
- our kafka versions 0.8.2.1 offsets stored in zookeeper
our issue on mirrormaker restart / consumer-rebalance partition ownership takes on 30+ mins when consuming local kafkas in different data centers.
my guess partition ownership process has go across data centers source kafka zookeepers, , cross dc connections slowing down mirrormaker partition ownership process.
i thinking of moving mirrormaker source data center instead, speed partition ownership on startup / rebalance. wondering if has tried setup , if there issues approach?
the mirror maker used high level consumer consume source cluster. default mirror maker takes time while re-balancing. not own problem. see uber struck mirror maker , how overcome ureplicator (https://eng.uber.com/ureplicator/). hope helps
Comments
Post a Comment