The code here describes a rounding issue that we have worked around on persistence of the scan info, by subtracting a millisecond. This code seems to indicate that blobs affected by this won't be picked up until the next restart. For a continuous WebJob (or Dedicated Function App) that might be a long time (days/weeks).

Because only the persisted value is decremented, the operating value being used is still susceptible to the issue. E.g. if a blob is added 10ms after the LastSweepCycleLatestModified value has been set, on the next scan it won't be picked up because the check done here is strictly greater than. This blob won't be picked up until after the next host restart.

I think we need to modify the in memory value of LastScan, not just the persisted value.