Hbase REST API, Globbing Rows undocumented behaviour

[#1]

Hi,

In our cluster we would like to retrive rows from Hbase in the simpliest available way using a REST API.

To retrive multiple rows with a REST call (GET) we can use the "Globbing Rows" REST api documented here -> https://hbase.apache.org/1.2/book.html#_rest in this way:

http://hbase_url:hbase_port/namespace:table/keywithwildcard*/[...]

We discovered that there is an undocumented behaviour. We can call the globbing REST api using startrow,endrow parameter in this way:

http://hbase_url:hbase_port/namespace:table/startrow,endorow/[...]

In this case it works exaclty as a scan (and it breaks the rest server if you try to retreive too many rows).

We also searched in this hbase souce code mirror (https://github.com/apache/hbase/blob/master/hbase-rest/src/main/java/org/apache/hadoop/hbase/rest/Ro...) and noticed that the code is intended to work like this (function private int parseRowKeys(final String path, int i) line 65-110).

Why is this behaviour not documented?

Is it safe to use (with restriction) this REST call instead of use a scanner or the absence of documentation means that the function will change without any comunication (not even the "deprecated" stuff)?

[#2]

I would never rely on undocumented code. This is likely exposed because there is no functionality built into the REST server to safely handle it. If a client would request too much data than the REST server can fit into memory, it will cause an OutOfMemoryError and crash.

[#3 · Accepted solution]

We engaged vendor professional support and we discovered that: 1- There is an internal thread on HBase REST API documentation that needs enhancement 2- It is possible to use this endpoint (but is not safe if you not have the control of the client that call the service) and, because it can cause OOM, is better/mandatory to use kerberos auth for Hbase REST API