Summary

timestamp_abort runs its workload in several iterations: It would run a workload in a child process, kill it later, recover, and then do it again. The trouble is, however, that each fork results in the child process starting with exactly the same random seed.

Suggested approach

We should re-initialize the two random states (one for data, one for extra) right either right before the fork or right after the fork, so that each iteration starts with a different state.

Ideally we would do this using testutil_random_init by using its n argument to get different starting positions from the same 3-byte seed, but the problem is that this does not work when the seed is already set, and moreover, the periodicity of n is 4, which is woefully low for our purposes. We would thus need to find a different solution.