Mark Dilger <hornschnorter@gmail.com> writes:
> While working on the problem of XID wraparound within the LISTEN/NOTIFY
> system, I tried to increment XIDs by more than one per transaction.
> This leads to a number of test failures, many which look like:
IIRC, the XID-creation logic is designed to initialize the next clog
page whenever it allocates an exact-multiple-of-BLCKSZ*4 transaction
number. Skipping over such numbers would create trouble.
> First, I'd like a good method of burning through transaction ids in
> tests designed to check for problems in XID wrap-around.
Don't "burn through them". Stop the cluster and use pg_resetwal to
set the XID counter wherever you want it. (You might need to set it
just before a page or segment boundary; I'm not sure if pg_resetwal
has any logic of its own to initialize a new CLOG page/file when you
move the counter this way. Perhaps it's worth improving that.)
> Second, I'd like to add Asserts where appropriate regarding this
> assumption.
I'm not excited about that, and it's *certainly* not a problem that
justifies additional configure infrastructure.
regards, tom lane