inferschema error on parquet files, mergeSchema is time consuming

I stored the parequet files in ADLS Gen2. Most time I read the file with no problem, but some time the collected parquet files triggered the "can not inferSchema" errors. I googled the error, most likely it was caused by an empty parquet files. One suggested fix is to turn on the option "mergeSchema" , like:

parquetFile = spark.read.option("ignoreCorruptedFiles", "true").option("mergeSchema","true").parquet(*fileForModels)

The option fixed the error, but the runtime is substantially longer (tripled or more). My questions are:

1. how to find the problematic parquet files in the ADLS Gen2? I checked each directory, the only zero size file is "_SUCCESS", but it was present in each directory, and I don't think it is the problem
2. Is there any better fix than "mergeSchema" option?