The following bug has been logged on the website:
Bug reference: 18474
Logged by: Clemens Eisserer
Email address: linuxhippy@gmail.com
PostgreSQL version: 16.3
Operating system: Linux
Description:
wal_writer_delay has a hard-coded limit of 10000ms, however the limit seems
artificial because depending on the type of workload 10s are either very
long of very short.
As far as I know wal_writer_delay is the only option for scanrios where the
durability of recently commited data is not that important, yet
database-corruption - as it can happen with fsync=off - should be avoided.
E.g. I have postgresql running on a raspberry pi storing sensor data, and
every 250ms an additional row with sensor data is written. Because of
fsync=on I have huge amounts of write-amplification (gigabytes), and it
would be great if it batching e.g. 10min of sensor-data would be possible.