Skip to content

[ATR-20466] Improve performance of trigger runtime validation.

Tim Martin requested to merge tamartin/athena:runtimeValidationMK2 into master
  • Keep track of nodes which have already been fully explored from by the recursive graph walk-and-validate function. Use this to avoid walking the branches more than once.
  • Only back-validate one step. As we back-validate from each HypoAlg, we only really need to validate until we reach back to the HypoAlg from the previous Step.
  • Add node-name validation
  • Add more generic link-name validation, also validate InputMaker and L1Decoder required links
  • Make methods static, such that they may be used externally

Fix minor bug this locate in PEBInfoWriterAlg.cxx which wasn't naming its HypoAlg node hypoAlgNodeName().

@wprzygod, you should be able to use this to validate your graph too. You would give it the HLTPassRaw node, with onlyValidateOneStep=false (to validate the whole graph in one go, rather than just the previous Step as will be done when using this in the trigger)

And, with runTwoConversion=true. At the moment this just lets the validation know that it should not expect to find any Filter nodes, these are only needed online so are skipped in your implementation. If there turn out to be other differences, we can included them off of this flag too.

Merge request reports