Integration test that should benchmark how fast HBase can recover from failures. This test starts
different threads:
-
Load Test Tool.
This runs so that all RegionServers will have some load and WALs will be full.
-
Scan thread.
This thread runs a very short scan over and over again recording how log it takes to respond.
The longest response is assumed to be the time it took to recover.
-
Put thread.
This thread just like the scan thread except it does a very small put.
-
Admin thread.
This thread will continually go to the master to try and get the cluster status. Just like the
put and scan threads, the time to respond is recorded.
-
Chaos Monkey thread.
This thread runs a ChaosMonkey.Action.
The ChaosMonkey actions currently run are:
- Restart the RegionServer holding meta.
- Move the Regions of meta.
- Restart the RegionServer holding the table the scan and put threads are targeting.
- Move the Regions of the table used by the scan and put threads.
- Restart the master.
At the end of the test a log line is output on the INFO level containing the timing data that was
collected.