The consistency level for write requests determines the number of replicas which must respond to the coordinator with an acknowledgment that the mutation was written to (a) the commitlog and (b) the memtable so "enforce" is not quite the right word for it.

The driver will report an exception depending on the error that occurred. Using the Java driver for illustration purposes, the following exceptions are reported by the driver in the application logs:

Unavailable

In this situation the request reached the coordinator but not enough replicas were available to satisfy the the consistency level requested.

An example error message looks like this:

UnavailableException: Not enough replicas available for query at consistency TWO (2 required but only 0 alive)

Timeout error

In this situation the request reached the coordinator but one or more replicas were too slow to respond within write_request_timeout_in_ms.

An example error message looks like this:

WriteTimeoutException: Cassandra timeout during SIMPLE write query at consistency LOCAL_QUORUM (2 replica were required but only 1 acknowledged the write)"

Failure error

This is a non-timeout error and occurs when some replicas replied with an error. The kind of errors that could trigger this could be one of the replicas had a running out of disk space or a replica experiencing a hardware failure.

An example error message looks like this:

WriteFailureException: Cassandra failure during write query at consistency QUORUM (4 responses were required but only 3 replica responded, 1 failed)

For more info, see How write requests are accomplished in Cassandra. There's explanation of the server-side exceptions which result in driver errors in the Java driver Retries page. For other drivers, see the respective driver's documentation in the DataStax drivers matrix page. Cheers!