The demerits of Spark are:
Developers are always careful while running Spark applications.
There is a possibility of having a bottleneck in the cost-efficient big data process with the capability of Spark in-memory.
Spark uses more storage space when compared to MapReduce and Hadoop. It leads to certain problems.
The work should distribute over multiple clusters instead of running on a single node.
A huge amount of data can be consumed by Spark when compared to Hadoop.