Skip to content

New TLA Workflow and Alignment

Marco Montella requested to merge mmontell/athena:TLA-Align-TJ into master

This MR features a major reworking of the TLA workflow following the discussion in ATR-24393, building upon the line of action suggested by TJ.

The following areas of the TLA workflow are targeted by the present MR:

TLA Step Implementation

TLA Step(s) were previously defined and added within the relevant signature code, as part of the chainPart configuration (see example ). This paradigm, when applied to multi-signature TLA chains, would cause serious issues in the menu alignment phase.

The operation of adding a TLA step is now moved outside the signature code, following the example of the existing EventBuilding step implementation.

Furthermore, while previously one TLA step would be added per chain signature, now only one TLA step per chain is added. The TLA step in combined TLA chains will feature multiple parallel sequences (see here

Object-copying operation

The central TLA task of copying the relevant HLT-level objects to be written to the TLA stream is moved from a reco-style algorithm to the HypoAlg. This change was spurred by the observation that, under the existing implementation, view-based low-pt objects failing the chain threshold would still be written to the TLA stream if belonging to the same EventView of an object passing the chain.

While this change of paradigm is essential for view-based objects such as Photons and Muons, the relevant changes were extended to the Jet TLA HypoAlg as well.

Alignment

The alignment of TLA steps is now rendered trivial by virtue of there only being at most a TLA step per chain, occurring at the end (with only the PEB step coming after) of the chain by design. The alignment is performed in the same way and menu configuration stage as the existing EventBuilding alignment.

Chain Count changes

chainComp INFO     Found 2 chains with count differences:
chainComp INFO       HLT_g35_loose_PhysicsTLA_L1eEM22M:
chainComp INFO         eventCount: 0 -> 2
chainComp INFO         stepCounts:
chainComp INFO           4: 0 -> 2
chainComp INFO           5: 0 -> 2
chainComp INFO         stepFeatures:
chainComp INFO           4: 0 -> 2
chainComp INFO           5: 0 -> 2

While unforeseen, we believe this change in the counts passing Step4 of the chain (the TLA copying step) actually fixes a previously undiscovered bug in the L1eEM22M TLA chain. We infer this since 2 events pass the last non-TLA step (Step3 in the chain), and the intended behaviour of the TLA and PEB steps (respectively 4 and 5) is such that no alteration in the number of counts is expected.

Tagging @tamartin for review

Thank you for your assistance,

The Run 3 TLA Developers Team

[@cantel , @boveia , @tberry , @doglioni , @mmontell , @fpastore]

Edited by Marco Montella

Merge request reports