Why architecture matters here

Read amp failures show up as slow reads + timeouts. Architecture matters because bloom + cache + compaction all decide.

Advertisement

The architecture: every piece explained

The top strip is the path. Read request. Block cache. Bloom filter. HFile fetch.

The middle row is cost. Tombstones. MemStore. Read amplification. Compaction strategy.

The lower rows are ops. Metrics. Row bloom vs prefix bloom. Ops — cache + compaction.

HBase read amp — bloom + block cache + tombstones + compactionkeep reads cheapRead requestGet / ScanBlock cachehot dataBloom filterskip HFilesHFile fetchactual dataTombstoneshurt scansMemStorerecent writesRead amplification# HFiles touchedCompaction strategyminimize ampMetricsblock hit + tombstonesRow bloom vs prefix bloomoptionsOps — cache sizing + compaction cadence + tombstone TTLcheckrecentmeasureplanwatchchoosechooseoperateoperate
HBase read path amplification factors.
Advertisement

End-to-end flow

End-to-end: read for row K. MemStore + block cache checked. Bloom filter skips 8 of 10 HFiles. 2 HFiles read. Block cache hits some. Total IO minimal.