Hi,
Perhaps this post is better suited to another PG mailing list? If this
is the cast then apologies up front :/
I've been playing around with improving some of the GiST code in PG
8.1 to better take advantage of domain knowledge where I am applying it.
I am now getting "ERROR: too many LWLocks taken" when I create index on
a largish table (about 50Mb index should be generated).
I only get this when shuffling around some index tuples inside
picksplit(). The pointers and left/right splits all seem to be in order
when I return and the smaller indexes I've created using this code are
all fine. My GiST is based on an existing implementation from PG 8.1 so
the allocation of the pages etc should all be fine.
Any hints as to what could be eating up LWLocks in this case?
Thanks.
