[HBASE-21642] CopyTable by reading snapshot and bulkloading will save a lot of time.

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 3.0.0-alpha-1, 2.2.0
• Hadoop Flags: Reviewed
• Assignee: Zheng Hu
• Reporter: Zheng Hu
• Created: 26/Dec/18 06:51
• Updated: 17/Jan/19 05:26
• Resolved: 27/Dec/18 11:58

Description

In our HBase clusters, some users has the need to merge two diff table's data into one. Currently , the CopyTable will scan the source table , and put mutations into destination table.
Although CopyTable with bulkload can speed a lot (compared to CopyTable with scan and put), it still take lots of time to scan the source table. and the worst thing is: CopyTable with scan table will impact the cluster's availablity, it cost lots of resource in RS to scanning, the cpu, memory, gc stw, rs handlers, disk io, network io ... etc. All those things will affect the availablity.

So in our clusters, we tried to do all scanning job by using scan snapshot instead of scan table. it at least isolate the cpu & memory & gc resource between the online RS and scanning job. What's more, the snapshot scanning is much faster than scaning RS, and it's more stable.

So, here, I'll make the copy table tool support snapshot scanning.

Issue Links

• relates to: HBASE-21675 Port HBASE-21642 (CopyTable by reading snapshot and bulkloading will save a lot of time) to branch-1
• relates to: HBASE-21514 Refactor CacheConfig
• links to: Review board