Turn the GUI memory bar for a worker to orange when it passes the target threshold, and to red when it pauses or enters its graceful retirement stage.
The cluster bar changes accordingly to orange if there's at least one worker spilling and to red if there's at least one worker paused or retiring.

In main: individual worker bars change to orange when above 50% and to red when above 100%.
The cluster bar changes to orange when the total cluster memory is above 50% and to red when above 100%.

Demo

import dask
import distributed
import dask.array as da

dask.config.set(
    {
        "distributed.worker.memory.max_spill": "250 MiB",
        "distributed.worker.memory.terminate": False,
    }
)
c = distributed.Client(
    threads_per_worker=4, n_workers=4, memory_limit="4 GiB"
)

rng = da.random.RandomState(0)
a = rng.random((16_000, 16_000), chunks=(4000, 4000))
b = (a @ a.T).sum().round(3)
b.compute()