Troubleshooting
Find the error message you got, then the fix.
Before the run starts
The following invalid input values have been detected
A parameter or combination of parameters failed validation, and nothing was submitted. The text in brackets after each line says what to change. For example:
| Message | Fix |
|---|---|
--gpuqueue (gpu-normal): Expected any of [gpu, cub22-inference, tiger22-inference] | Use one of the listed queues |
Missing required parameter(s): costcode (...required for cub and tiger queues) | Add --costcode |
--costcode (s1): Expected null (...only used with cub and tiger queues) | Remove --costcode, or add --gpuqueue cub22-inference |
--estimator (mean): Expected null (...only available for CellBender 0.3) | Remove it, or use --version 0.3 |
--exclude_features (Peaks): Expected All (...) | CellBender 0.2 only takes All |
Object does not have some of the required properties [droplets] (...) | With CellBender 0.2, give --cells and --droplets together |
--samples (samples.csv): Validation of file failed: ... duplicate entries | Make every id in the table unique |
The extra line Value matches against schema from 'if' but does not match against schema from 'then' is printed under every failed rule; ignore it. Parameters lists every rule.
Only a short usage message is printed
--samples is missing. If you used --sample_table, it was renamed to --samples, and --output_dir to --outdir.
WARN: The following invalid input values have been detected: * --foo: 1
An unknown parameter is only a warning, and the run goes ahead without it. Check its spelling and, after an upgrade, the changelog for renamed parameters.
nextflow run -preview never finishes
Known Nextflow behaviour with workflow outputs: when validation passes, -preview waits forever for outputs that are never produced. Use it only to check that parameters are rejected, or test with -stub instead.
Downloading from iRODS
CAT_NO_ACCESS_PERMISSION
ERROR: getUtil: get error for multiplex-flex/README.md status = -818000 CAT_NO_ACCESS_PERMISSION
Your iRODS account can list the collection but not read its files. Retries will not help: stop the run, get read access as described in Data on iRODS, then rerun with -resume.
CAT_INVALID_USER from ichmod
The user was given without its zone. On /archive paths write ab76#Sanger1 for a user, or cellgeni#archive for the team group.
CAT_INVALID_AUTHENTICATION
The account running the iRODS command is not logged in to iRODS. Run iinit and try again; for a service account, use its iRODS password.
While the run is going
Disk quota exceeded
On /lustre/scratch124, files owned by your primary Unix group (for example team278) count against that group’s quota, which can be as small as 1 MB. Files created under a folder owned by the cellgeni group normally inherit cellgeni, but a tool that writes in /tmp and then copies the result into the work folder brings the primary group with it.
Check:
lfs quota -h -g team278 /lustre/scratch124
find nf-work -group team278 | head
Start the run from a folder in the cellgeni group. The MultiQC module already keeps its temporary files inside the work folder for this reason; if another tool hits it, tell the developers which one.
CellBender keeps failing on one sample
Look at the task’s .command.log in nf-work/ (the run log prints the folder). Common causes:
Error: Input directory does not contain expected structure:pathis not one of the layouts in Cell and droplet counts.ERROR: Mapper preset is not available for ... input type: CellBender 0.2, or 0.3 with--mapper_preset, on a bare.h5or.mtxinput. Give--cellsand--droplets.- Out of GPU or host memory: memory grows with each retry. If it fails on every attempt, raise it with your own config (see GPU queues).
To let the other samples finish while you look into it, rerun with --ignore_failed -resume.
In the QC report
The soup fraction is NA
The sample ran with CellBender 0.2, whose output does not store the soup fraction, and its input matrix could not be read at QC time. See QC report.
The cell calling plot has no UMI curve, or the gene removal plot is not available
Both need the CellBender input, which the QC step reads only in --qc_mode 3 and only while it is still readable at the path in the CellBender log. See QC report.
A sample is missing from the report
Its CellBender task failed and was ignored (--ignore_failed), or its folder holds no cellbender.log. The QC log in nf-work/ names every skipped folder.