pg_upgradecluster: Port old postgres.conf file before starting data migration

[Tim Watt]

This is more an enhancement request than a bug report.

Background: I was upgrading from postgres 8.4 (Ubuntu 11.04) to 9.1 (Ubuntu 11.10) using the steps in README.Debian and noticed it taking unusually long. I checked the log for the new version, and I noticed that it was checkpointing quite often (once every few seconds). I checked /etc/postgresql/9.1/main/postgresql.conf and noticed it did not have any of my customizations from 8.4 (work_mem, fsync, etc.).

Example logs: "checkpoints are occurring too frequently (10 seconds apart)" / 'Consider increasing the configuration parameter "checkpoint_segments"'

I worked around this by adding 'system("/bin/bash");' to pg_upgradecluster around line 275 (right before it started the target postmaster). This shell gave me time to edit /etc/postgresql/9.1/main/postgresql.conf before the data transfer commenced.

I'm not sure what a general approach would be, but I figure logging this here will at least help guide someone who notices it in the future.

If it's too risky to port the whole config before data transfer, perhaps pg_upgradecluster could transfer performance-critical settings from the old configuration.

[Martin Pitt (pitti)]

I guess you can break upgrades either way, but I'll think about porting the config before starting the upgrade, and then just (temporarily) resetting some particular settings (port, etc.).