Why architecture matters here

Splitting failures produce hot spots + region hell. Architecture matters because split policy + schema decide balance.

Advertisement

The architecture: every piece explained

The top strip is mechanics. Region. Auto-split policy. Split point selection. Split process.

The middle row is management. Presplitting. Hot spots. Salting / bit-reverse. Region balancer.

The lower rows are ops. Metrics. Anti-split hints. Ops — schema + monitoring + capacity.

HBase region splitting — policies + presplit + skew + hot spotshow regions grow + how to keep them balancedRegionrow range on RSAuto-split policysize / age / customSplit point selectionmedian row keySplit processseal + create childrenPresplittingmanual on createHot spotsmonotonic keySalting / bit-reversespreadRegion balancermove regionsMetricsregion count + size + hotAnti-split hintskeep togetherOps — schema + monitoring + capacitypreemptdetectmitigatebalancewatchhinthintoperateoperate
HBase region splitting lifecycle.
Advertisement

End-to-end flow

End-to-end: presplit 24 regions at table create using key hash prefixes. Auto-split policy: 10GB per region. Balancer keeps 4-6 regions per RS. Metrics show even distribution. If monotonic key hot spot detected, add salt prefix.