I am examining various global variables in ShmemVariableCache, pg_control and pg_controldata. To force and debug xid wraparound, I've implemented a function, that allows to set nextXid value in control file manually, but it doesn't work as expected.
Why not use pg_resetxlog -x ?
After an investigation I found that, while loading global variables in StartupXlog() we read checkPoint location from ControlFile, and then read all the information from the checkPoint record. And never ever use nextXid stored in ControlFile.
What is the purpose of having CheckPoint stored in control file then? I thought, that it allows us to start up server after correct shutdown without any xlog, as said in the comment in pg_control.h
* Body of CheckPoint XLOG records. This is declared here because we keep * a copy of the latest one in pg_control for possible disaster recovery. * Changing this struct requires a PG_CONTROL_VERSION bump.
But as far as I see, we never use this data.
Could you please explain, why don't we use information from control file in StartapXlog()?
I believe pg_resetxlog uses the data from the control file to create an artificial checkpoint record, and injects that record into an artificial xlog file, which will allow you start the system. So it is a small use, but a sometimes very important one.