Adrian Klaver <adrian.klaver@xxxxxxxxxxx> writes: > On 2/27/23 08:36, Tom Lane wrote: >> If it was based on something like "du", perhaps the measurement >> was fooled by the fact that most of the data files will be hard-linked >> between the old and new clusters. > Does that happen without the --link option? No, but the OP didn't mention whether he used that, and even if he didn't say it explicitly the pg_upgradecluster wrapper might've supplied it. regards, tom lane