This is an attempt to observe the behavior Accumulo displays when compacting and reading cells. There are two components to this package: org.apache.accumulo.test.stress.random.Write and org.apache.accumulo.test.stress.random.Scan.
Since this test is meant to stress the system, there is no pre-defined test period and no success criteria that we can programmatically check. During a successful test period, the Accumulo cluster should remain stable and functional. Possible failure conditions include component failures, such as tablet servers running out of memory.
The org.apache.accumulo.test.stress.random.Write utility provides facilities for writing random sized cells. Users can configure minimum and maximum sized portions of a cell. The portions users can configure are the row, column family, column qualifier and value. Note that the sizes are uniformly distributed between the minimum and maximum values.
The org.apache.accumulo.test.stress.random.Scan utility provides users with the ability to query tables generated by the Write. It will pick a tablet at random and scan the entire range. The amount of times this process is performed is user configurable. By default, it happens 1,024 times. Users can also specify whether or not the scan should be isolated.
There is no shared state intended by either of these services. This allows multiple clients to be run in parallel, either on the same host or distributed across hosts.
The utilities are run on a normal, initialized Accumulo cluster. By default, they only work with a table named “stress_test”.
The start and stop scripts rely on pssh. Before running them, you will need to install pssh on the machines that will be controlled by them.
Copy stress-env.sh.example to stress-env.sh and edit it, supplying at least correct values for the USERPASS and INSTANCE variables. The remainder of this file describes and provides defaults for many of the configuration options for the stress utilities.
Before starting a fresh write test, clear out the test table using the Accumulo shell.
> deletetable -f stress_test > createtable stress_test
To run a writer:
$ ./writer.sh
The writer will begin logging into a “logs” subdirectory, using timestamped filenames. You can stop the writer by killing it or using Control-C.
To run a reader:
$ ./reader.sh
The reader logs like the writer. By default, the reader reads forever, but it can be configured for a fixed number of scan iterations.
To run writers across multiple machines, create a writers file and list each host that will run a writer, one per line.
Finally, to start and stop the writers:
$ ./start-writers.sh $ ./stop-writers.sh
For readers, make a similar readers file, and use start-readers.sh and stop-readers.sh.
The start scripts copy out stress-env.sh to each of the writers and readers, overwriting what is there, so you only need to edit the file on the host from which you run those scripts.
Note that in versions of Accumulo greater than 1.6.0, users should verify that the cell sizes they supply meet the criteria set forth by the key constraints on their tables. By default, tables are configured to reject keys that are greater than 1MB.