Repost from @fjetter (#7725 (comment)):
I'm observing an artifact with the plots. If I am observing the dashboard while the computation is going on, the metrics are added to the chart as I'd expect.
Once the computation finishes, a couple of tasks are dropped from the dashboard. A couple seconds/a minute later some other tasks show up again.
If I refresh the page, I get accurate results for a short time but the same process of "drop a couple of tasks" happens again.
from dask.datasets import timeseries
from distributed import Client
client = Client()
ddf = timeseries("2000", "2020", partition_freq="1w")
ddf.groupby("id").x.mean().compute()
