"Device or resource busy" with Toil and Slurm

To make better use of cluster resources I'm trying to run LINC with Toil, but running into a little snag. I have it running for the most part, submitting jobs to a Slurm queue, running the first calibrate step etc. but somehow it crashes when it does the first applycal of the PA solutions. The error I get is

WARNING: SINGULARITY_BINDPATH and APPTAINER_BINDPATH have different values, using the latter
WARNING: SINGULARITYENV_TMPDIR and APPTAINERENV_TMPDIR have different values, using the latter
WARNING: Skipping environment variable [SINGULARITYENV_TMPDIR=/tmp], TMPDIR is already overridden with different value [/cosma8/data/do011/dc-swei1/tmp.8wJZyv9ssH/tmpdir_LINC_calibrator/]
terminate called after throwing an instance of 'casacore::AipsError'
  what():  RegularFile::move error on /MVptMR/out_L726576_SB002_uv.MS/table.dat_tmp to /MVptMR/out_L726576_SB002_uv.MS/table.dat: Device or resource busy

with the accompanying DP3 output of

WARNING - software has been build with -march=znver2 but current machine reports -march=sapphirerapids.
If you encounter strange behaviour or Illegal instruction warnings, consider building a container with the appropriate architecture set.
WARNING - software has been build with -mtune=znver2 but current machine -mtune=sapphirerapids.
If you encounter strange behaviour or Illegal instruction warnings, consider building a container with the appropriate architecture set.
MSReader
  input MS:       /MVptMR/out_L726576_SB002_uv.MS
  band            0
  startchan:      0  (0)
  nchan:          4  (0)
  ncorrelations:  4
  nbaselines:     2775
  first time:     2019/07/12/14:00:02
  last time:      2019/07/12/14:09:59
  ntimes:         150
  time interval:  4.00556
  DATA column:    DATA
  WEIGHT column:  WEIGHT_SPECTRUM
  FLAG column:    FLAG
  autoweight:     false
ApplyCal applycal.
  H5Parm:         /var/lib/cwl/stge175b851-7861-4d3d-a674-f8b1b494d07b/cal_solutions.h5
    SolSet:       calibrator
    SolTab:       polalign
  Direction:      0
  Interpolation:  nearest
  Missing antennas: error
  Correction:       diagonal phase
  Update weights:   false
  Invert:           true
    SigmaMMSE:    0
  TimeSlotsPerParmUpdate: 200
Counter count.
MSUpdater msout.
  MS:             /MVptMR/out_L726576_SB002_uv.MS
  datacolumn:     CORRECTED_DATA  (has been added to the MS)
  flagcolumn:     FLAG
  weightcolumn    WEIGHT_SPECTRUM
  writing:        data flags
  Compressed:     no

  flush:          0

*** WARNING: the following parset keywords were not used ***
             maybe they are misspelled
    msout.storagemanager.databitrate
Processing 150 time slots ...

0%....10....20....30....40....50....60....70....80....90....100%
Finishing processing ...

NaN/infinite data flagged in reader
===================================

Percentage of flagged visibilities detected per correlation:
  [753201,0,0,0] out of 1665000 visibilities   [45%, 0%, 0%, 0%]
0 missing time slots were inserted

Flags set by OneApplyCal applycal.
<snip>

Has anyone managed to run LINC successfully with Toil yet? I'm posting here as I'm not entirely sure why it's specifically this step that crashes here, if it's my environment/container/setup or the workflow, and if some other eyes maybe have an idea of where to look further for debugging.

The way I set things up, if useful information, can be found here: https://github.com/tikk3r/flocs/blob/toil-runners/runners/run_LINC_calibrator_HBA_toil.sh

I'm using toil 6.0.0