The following documentation comment has been logged on the website:
Page: https://www.postgresql.org/docs/13/sql-refreshmaterializedview.html
Description:
Hi,
it may seem obvious, but it took us TB of WALs and some time to identify the
problem, why a bigger materialized view created much more WALs than a normal
refresh.
A concurrent refresh needs at least one unique index
and no column of these indexes should be null: unique columns with null
cannot identified as unchanged and will be replaced.
I am not sure about the wording, but there should be a hint that this can be
a problem.
This might be ok for only a few rows but we had >14.000.000 rows and only a
few thousand had all values set for the unique index. Hourly refresh, TB of
WAL/day for just 4GB of database :-(
This is at least valid for 11 - 13
Thanks :)
Stefan
