DFTracer trace files not generated and missing hydra_log folder with profiling and logging enabled

I attempted to enable DFTracer profiling and hydra logging in the mlpstorage training workflow, but I was unable to locate the expected output trace/log files.

Steps Taken:

Installed DFTracer:
pip install pydftracer

Exported environment variable:
export DFTRACER_ENABLE=1

Ran the training command with DFTracer:

mlpstorage training run \
  --hosts <ip> --num-client-hosts 1 \
  --client-host-memory-in-gb 256 \
  --num-accelerators 4 \
  --accelerator-type h100 \
  --model unet3d \
  --data-dir /mnt/mlperf/june25/data/ \
  --results-dir /mnt/mlperf/june25/result_july7/run9 \
  --param dataset.num_files_train=500 \
           dataset.num_subfolders_train=70 \
           reader.read_threads=8 \
           reader.odirect=True \
           reader.prefetch_size=0 \
           workflow.profiling=True \
           workflow.profiling=dftracer \
  --checkpoint-folder /mnt/chkpt

Despite the setup, I was not able to find the DFTracer trace files in the results directory or elsewhere.

Tried alternate logging settings (iostat + hydra):

mlpstorage training run \
  --hosts <ip> --num-client-hosts 1 \
  --client-host-memory-in-gb 256 \
  --num-accelerators 4 \
  --accelerator-type h100 \
  --model unet3d \
  --data-dir /mnt/mlperf/june25/data/ \
  --results-dir /mnt/mlperf/june25/result_july8/run1 \
  --param dataset.num_files_train=500 \
           dataset.num_subfolders_train=70 \
           reader.read_threads=8 \
           reader.odirect=True \
           reader.prefetch_size=0 \
           workflow.profiling=True \
           workflow.profiling=iostat \
           workflow.hydra_logging=enabled \
           workflow.job_logging=enabled \
  --checkpoint-folder /mnt/chkpt

However, I couldn't find the expected hydra_log folder or logs in the results directory either.

Questions:

• Is there any additional step required to make sure DFTracer generates trace files?
• Where should the hydra_log directory be located when logging is enabled?
• Are there known issues with workflow.profiling=dftracer or workflow.hydra_logging=enabled flags?

Please let me know if further logs or config dumps would be helpful.

Start at the mlpstorage training run entry point and reproduce the two commands from the issue with workflow.profiling=dftracer and the Hydra logging flags. Inspect the configured results directories and runtime configuration to determine where trace files and the hydra_log folder are expected. Done means identifying whether the flags work, documenting the output locations, or isolating a reproducible configuration problem.