I ran some of Aphyr's Jepsen tests against HBase. The conclusion is that HBase performed well.
The test I used is at https://github.com/rayokota/jepsen/blob/old/src/jepsen/hbase.clj. I used CDH 5.1.2, which bundles hbase-0.98.1+cdh5.1.2+70. The first test (hbase-app) is to simply create unique key-value pairs while a network partition occurs. 0 unrecoverable timeouts Collecting results. Writes completed in 200.047 seconds 2000 total 2000 acknowledged 2000 survivors All 2000 writes succeeded. :-D The second test (hbase-append-app) is to modify the same cell repeatedly using checkAndPut (by appending to a list stored as a blob) while a network partition occurs. As expected, most writes did not succeed due to the checkAndPut call failing because another client concurrently modified the row. However, no data loss occurred. 0 unrecoverable timeouts Collecting results. Writes completed in 200.05 seconds 2000 total 282 acknowledged 282 survivors all 282 acked writes out of 2000 succeeded. :-) The third test (hbase-isolation-app) is to modify two cells in the same row while a network partition occurs, to test if row updates are atomic. This is the same test as the one for Cassandra that demonstrated that updates to multiple keys in the same row are not atomic in Cassandra ( http://aphyr.com/posts/294-call-me-maybe-cassandra/). 0 unrecoverable timeouts Collecting results. () Writes completed in 200.043 seconds 2000 total 2000 acknowledged 2000 survivors All 2000 writes succeeded. :-D
