The Memory usage metric is measured at the VM level and does not include the OS cache. It's equivalent to "MemTotal - MemAvailable" from the VM's /proc/meminfo.
The Total memory usage metric is measured at the database container level. It includes memory used by the database container itself, and the OS cache.
The metric names might suggest that Total memory usage is always higher or equal to Memory usage. This is not true because those metrics are measured at different levels. (VM versus Docker). In other words, it's not Memory usage and Total memory usage for the same thing, but rather Memory usage for one thing and Total memory usage for another thing.
For example:
Memory usage can be higher than Total memory usage because it's measured at the entire VM level, so it includes memory usage by components other than the database container. If the container and cache usage is low (e.g. because the database is small, idle, and doesn't use much OS cache), VM usage can potentially be higher than container usage. There are also differences in how memory allocation is measured in Docker versus the OS, which can contribute to the discrepancy.
Total memory usage can be, and often is, higher than Memory usage because it includes OS cache. Busy, warm databases can use large amounts of OS cache, which inflates Total memory usage.