A monkey test tool framework that can be used to stress test POSIX filesystems. There is a library of common filesystem operations (open,mmap, etc). And a small framework library that does some housekeeping (like tracking open FDs). Some example tools are provided.
Reading the monkey test overview, would help you understand the goals of this project.
Develop a new monkey test tool that will meet these objectives:
- Simplify adding new tests to the operational pool.
- Increase code sharing of operation pools between different test tools.
- Every test tool should use the same execution framework.
- Every test tool should have the same command line API.
- Every test tool should have a clearly defined test strategy.
- Develop methods to reduce the effects of the Pesticide Paradox
- Develop methods to reduce the amount of code and the execution time of the monkey test to reproduce the kernel bug.
- Develop methods that take advantage of clusters.
- The operation pool is maintained as a library and can be shared between multiple test tools with different test strategies.
- Instead of altering an existing operation -- fork the operation and create a new one.
- Use the krt framework for all test tools.
- All tools need to have the same command line API to allow script reuse.
- All test tools use the krt framework and common operation libraries.
- Each test tool needs to have a well defined test strategy.
- Instead of changing a tool's strategy, fork the tool and create a new one.
- Make it easy to add new operations to the pool.
- Make it easy to prototype a new test strategy.
Split the monkey tool into two parts: one tool uses entropy to generate a file system operation list and another tool that replays this pre-generated operation list.
Using pre-generated operation lists is a win due to the future improvements this technique enables:
- competitive algorithms can be used to create operation lists that maximize bug count
- corpus distillation techniques can be used to create operation lists that maximize code coverage
- regression test suites can be created using a set of minimized operation lists that have caused past regressions
- improvements can be made to the front end without effecting existing regression tests
- automated techniques can be developed to reduce a operation list to the minimum required to reproduce the failure in the software under test