Test RASBASE durability

[Dimitar Misev]

Motivated by the database corruption issues reported in #1014, I'm proposing to develop a RASBASE stress test.

This is a rough sketch:

• let W be a set of representative insert and update queries
• let R be a set of representative select queries R
• test execution:
  1. a random query from W is executed, with random queries from R in parallel
  2. rasserver is killed with SIGKILL at a random point of time (perhaps other signals as well)
  3. rasdaman is restarted, and checked for consistency
     • save logs and query details in case of corruption
  4. repeat

[Peter Baumann]

that looks like an important test class indeed, and could be broadened in future.

[George Merticariu]

Preferably, we should use the intefaces used in rasdaman in a UnitTest like setup so we can reproduce.

We already have the UnitTest framework in rasnet, so we can use that one.

[Dimitar Misev]

The test would be completely randomized, repeating a unit test would not reproduce the condition of the previous run - that's why we need to save all information needed so we can reproduce a single run:

• save logs and query details in case of corruption

[Dimitar Misev]

This has been implemented for single-thread testing only, should be further adapted for parallel queries.

[Alex Dumitru]

Can we open a new ticket for parallel queries? The initial feature request seems completed.