Maybe improve the cube such that more calculation is done offline instead of scanning many segments and data online?
On Tue, Apr 17, 2018 at 10:09 AM, 小村庄 <[email protected]> wrote: > hi all: > Some SQL will take up too much memory in the application, the > service is not stable, after analysis because the SQL query need to scan > the relevant segemnt corresponding hbase table, is through the thread > asynchronous request related to SQL cuboid corresponding region. After each > segment makes a request, wait for the data to process the hbase callbacks. > But sometimes SQL involves more segments, and before all the segments make > requests, there is a large amount of data returned, leading to memory alarm > and service instability. >
