HltPackedDataWriter compressed buffer size fluctuates
The Size of compressed data counter from HltPackedDataWriter fluctuates slightly between runs.
| Counter | # | sum | mean/eff^* | rms/err^* | min | max |
#First run
| "Size of compressed data" | 421 | 3736579 | 8875.5 | 10531.0 | 550.00 | 59912.0 |
#Second run
| "Size of compressed data" | 421 | 3736634 | 8875.6 | 10531.0 | 550.00 | 59926.0 |
#Third run
| "Size of compressed data" | 421 | 3736653 | 8875.7 | 10531.0 | 550.00 | 59928.0 |
#Fourth run
| "Size of compressed data" | 421 | 3736534 | 8875.4 | 10530.0 | 550.00 | 59919.0 |This became apparent when creating references for the Moore test included by Moore!985 (merged). The numbers above were obtained running the test on 4 threads. Running it single-threaded also shows fluctuation:
#First Run
| "Size of compressed data" | 421 | 3760293 | 8931.8 | 10499.0 | 548.00 | 64510.0 |
#Second run
| "Size of compressed data" | 421 | 3760430 | 8932.1 | 10500.0 | 548.00 | 64516.0 |
#Third run
| "Size of compressed data" | 421 | 3760210 | 8931.6 | 10499.0 | 548.00 | 64511.0 |Edit: Enabling checksum in the HltPackedDataWriter shows that the input uncompressed data changes between runs. The logs for two runs are attached (log1.log, log2.log). The locations whose checksums change are
'/Event/HLT2/pPhys/Relations'
'/Event/HLT2/pPhys/Vertices'
'/Event/HLT2/pRec/Muon/CustomPIDs'
'/Event/HLT2/pRec/Muon/MuonPID'
'/Event/HLT2/pRec/ProtoP/Charged'
'/Event/HLT2/pRec/ProtoP/Custom'
'/Event/HLT2/pRec/ProtoP/Neutrals'
'/Event/HLT2/pRec/Rich/CustomPIDs'Edit 2: Applying Gerhard's changes from !3244 (closed) improves the situation, as most checksums stop fluctuating. Only the checksum associated with
'/Event/HLT2/pPhys/Relations'still fluctuates. The Size of compressed data counter still fluctuates (log1_gerhard_fix.log log2_gerhard_fix.log).
Edit 3: Once Moore!997 (merged) is merged, these checksums can be obtained running
Moore/run gaudirun.py Moore/Hlt/Hlt2Conf/tests/qmtest/test_hlt2_check_packed_data_checksums.qmt