hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jim Hughes <jhug...@ccri.com>
Subject Best practices for stopping Coprocessor calls/scans in client code
Date Thu, 13 Feb 2020 16:53:35 GMT
Hi all,

I work on GeoMesa which uses HBase to store geospatial data.  One of our 
uses patterns to use GeoMesa+HBase client code in a web tier to respond 
to ad hoc user queries.  Those web requests are translated to HBase 
coprocessor calls or HBase Scans.

For this use case, we want to manage the number of threads and time that 
those threads can use for any given request. Concretely, we may want a 
particular deployment to 'time out' requests after 60 seconds.

With that in mind, I have two broad questions:

1.  Are there example open source code samples which show ways to the 
HBase client in this manner?  (For instance, has Apache Phoenix already 
solved this problem.  I kinda tried to look at that code base for ideas.)

2.  Is there an expected / documented pattern which would let me solve 
this problem well?

In terms of the second question, I have seen that the HTableInterface 
has a getTable(TableName tableName, ExecutorService pool) method.  One 
solution is to pass in a new ThreadPool/ExecutorService per GeoMesa 
request.  When the GeoMesa timeout hits, we can cool shutdownNow.

This solution is 'correct' in that it achieves the desired timeout 
behaviour.  The downside is that it is slower than our previous code 
which just fired off requests without a timeout capability.

Thanks in advance for any links, suggestions, positive feedback, etc!



View raw message