Robustness to BiocParallel
Closed this issue · 4 comments
drisso commented
- Handling an error from a normalization method (not propagated)
- Handling error in the projection
drisso commented
@HenrikBengtsson if you want to help, this is probably the place where we need it the most!
HenrikBengtsson commented
Related to what we discussed over lunch:
You're looking for what I would call "early propagation of errors" or possible shorter "early stopping". It can only occur when the calling R processes checks in with the background R processes ("pools the workers for errors"), that is, if a background R process detects an error, none of the asynchronous backends I know of in R can push this to the main process such that the error occurs immediately.
drisso commented
We should have a look at this:
https://www.bioconductor.org/packages/release/bioc/vignettes/BiocParallel/inst/doc/Errors_Logs_And_Debugging.pdf
drisso commented
Useful param values for debugging:
register(SerialParam(stop.on.error=TRUE, log=TRUE, threshold = "TRACE"))