Error in race_state$recover() : argument "scenario" is missing, with no default

[felixvuo]

Hi

I've been trying out irace 4.0 on our university's cluster. After 12 hours of running, irace hadn't quite finished, so I tried to re-start the process by using the --recover-file option. Unfortunately, I get this error:

$ srun -n 1 -c 42 -t 300 irace --recovery-file "./bak.irace.Rdata" --scenario "./scenario.txt" &
...
Error in race_state$recover() : 
  argument "scenario" is missing, with no default
Calls: irace_cmdline ... eval.parent -> eval -> eval -> <Anonymous> -> <Anonymous>
Execution halted

I get the same error with and without the --scenario parameter.

The file bak.irace.Rdata is a copy of the irace.Rdata at the end of the timed out run - unfortunately that run was killed by the slurm scheduler after 12 hours - is it possible that things were therefore left in an unusable state?

The scenario.txt file lives in the working directory and is fairly brief:

$ cat scenario.txt 
parameterFile = "./params-kissat.txt"
configurationsFile = "./params-default.txt"
logFile = "./irace.Rdata"
trainInstancesFile = "trn-instances.txt"
testInstancesFile = "tst-instances.txt"
capping = 1
deterministic = 1
#maxExperiments = 1024
maxTime = 360000
debugLevel = 0
seed = 2024
boundMax = 600

The version of irace is:

$ irace --version
#------------------------------------------------------------------------------
# irace: An implementation in R of (Elitist) Iterated Racing
# Version: 4.0.886dd4c

The R version:

$ R
R version 4.4.1 (2024-06-14) -- "Race for Your Life"
Copyright (C) 2024 The R Foundation for Statistical Computing
Platform: x86_64-pc-linux-gnu

Attached are the log and the rdata file.
rdata-and-log.zip

Any pointers would be appreciated. Thank you.

[MLopez-Ibanez]

Hi Felix,

This seems to be a bug in irace 4.0. I can reproduce it with the files that you have provided but I cannot test the fix without additional files. Could you share "./params-kissat.txt", "./params-default.txt" and "trn-instances.txt"?

Thanks!

[MLopez-Ibanez]

Hi Felix,

This bug will be tricky to fix and it will take me some time, so I'd recommend to relaunch irace instead of waiting for a fix.

[felixvuo]

This seems to be a bug in irace 4.0. I can reproduce it with the files that you have provided but I cannot test the fix without additional files. Could you share "./params-kissat.txt", "./params-default.txt" and "trn-instances.txt"?

Hi Manuel, thank you for looking into this. Here are the 3 files.
params-and-instances.zip

[MLopez-Ibanez]

Hi, this bug is fixed in github (which will become irace 4.2) but it won't be able to recover the files that you provided because the recovery has been reimplemented from scratch.

[felixvuo]

Hi Manuel,
Thank you for fixing this. It's ok not to recover past runs. It's nice to know that from now if I get my cluster timing estimates wrong it may be possible to resume a long tuning run which is killed by slurm.