How many records are participating in match and merge? It could be that the process is creating large tmp files under /hub/cleanse/tmp (on all nodes, if this is a clustered env). Even though you remove them, it still creates those files when you rerun the job and that may be eating up disk space.

Setting this to DELETE might work:

cmx.server.datalayer.cleanse.working_files=DELETE

Also, you might want to monitor total disk space utilization when job is running and after it has failed.

This property controls whether tmp files should be deleted or retained.

These tmp files can be considered as logs; temporary working files that MDM processes use: typically stage, tokenization, and match. MDM creates these intermediate temporary files for internal processing before moving/inserting records to database. These files can grow in size depending the amount of data processed by these jobs. The property lets you keep these files for debugging/troubleshooting purposes.

This could be one of the reasons why all files are not deleted:

KB 515965

Are you still running into the match error?

How how many records you are running for M&M.

Please share the complete stack trace if you are still facing the issue.

We need to see complete log to get the exact root cause.

You can try to limit the number of records in the match cycle using "Number of rows per match job batch cycle" to reduce the size of the files as a workaround.