High Availability is most often defined as a system that is free of
Single Points Of Failure (often shortened to SPOFs). Thus, any component failing will result in system slowdown, but not in full loss of service. In practice, if an entire node in a High Availability
cluster fails, there will be a brief loss of service (e.g, active
TCP connections will be aborted), and a slowdown while the
caches are being
warmed.
In older texts, especially marketing material, you will see Fault Tolerant and Highly Available used interchangably. Nowadays, Fault Tolerant is most commonly used to describe a system where the hardware let any component fail without having any impact on the software.
Note that a highly available system may be more attractive than a fault tolerant system, as the former is usually also resistant to many forms of software fault.