Bugfix/dont crash on invalid homogeneous batch dataset

[mgovers]

The issue

Non-caught exceptions caused by real-world update scenarios thrown by a worker thread in parallel batch calculations cause fatal crashes.

Considerations

After investigation related to exceptions during thread creation/execution, the following findings were made:

• It is possible that the main thread throws an error when creating a thread
  • If this happens, stack unwinding is done
  • then a joinable std::thread will be destructed
  • this causes std::terminate()
  • This is OK because it is of the highest severity and the PowerGridModel instance cannot continue with its current tasks
    • Decision: intended behavior. the users' program crashes and we do not catch this to throw a PowerGridError
• It is possible that a worker thread raises a exception
  • Option 1: in a scenario
    • this should be handled by the scenario runner and instead a PowerGridBatchError should be raised
    • This is intended behavior
  • Option 2: during creation of the thread
    • Option 2a: during copy of MainModel, e.g.
      • This points at a severe issue like running out of memory
      • This causes std::terminate()
      • This is intended behavior (see above)
    • Option 2b: during other caching mechanisms
      • This may be caused by a real-life scenario (see below)
      • This causes std::terminate() (fatal crash).
      • This is a bug: instead, it should cause an actual PowerGridError
      • This is solved by this PR

Preconditions

The above-mentioned bug solved in this PR arises when all of the following conditions are satisfied:

1. There is a batch calculation
2. Running in parallel with at least 2 threads (threading=0 or threading=x where x >= 2)
3. With at least 2 scenarios
4. With homogeneous update data (all scenarios touch the same IDs)
5. And at least 1 ID in the update data is not present in the input data

These conditions trigger the following contributing factors:

• The 5th condition triggers the exception
• The 4th condition triggers throwing the exception in the setup part instead of in a scenario
• The 1st, 2nd and 3rd condition triggers throwing the exception in a worker thread

The solution

The caching in which the exception was moved from the worker-thread setup part to all-scenario combined setup. This prevents non-handled exceptions from being raised in the worker thread

[sonarqubecloud]

Quality Gate passed

Issues
1 New issue
0 Accepted issues

Measures
0 Security Hotspots
100.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarCloud