On Tue, Feb 06, 2018 at 01:36:04AM -0600, Jeremy Finzel wrote: > Here is the basic structure - is the gist index significant?: > > CREATE UNLOGGED TABLE foo ( > as_of_date daterange NOT NULL, > customer_id integer, > bunch_of_fields_here); > > ALTER TABLE ONLY foo > ADD CONSTRAINT foo_as_of_date_excl EXCLUDE USING gist (customer_id WITH > =, as_of_date WITH &&); > > CREATE UNIQUE INDEX foo_idx1 ON foo USING btree (customer_id) WHERE > (upper(as_of_date) = 'infinity'::date); > > CREATE INDEX foo_idx2 ON foo USING btree (customer_id, lower(as_of_date)) > WHERE (upper(as_of_date) = 'infinity'::date); > > CREATE UNIQUE INDEX foo_idx3 ON foo USING btree (customer_id, > lower(as_of_date));
I am not sure, but I would think about something related to gist here when heavy insertions are done on it... I cannot put my finger on the thread though.
> This is all I see - please help me if there's a better command I can > run:
If the process is still running, can you attach gdb to it and then run the command bt? You may need to install debugging symbols to make the trace readable. -- Michael
I am trying a few other scenarios to see if I can reproduce. I was able to set to logged a copy of the table with no indexes. I am now attempting same with only the gist index. If I can reproduce it on a non production server I will try gdb.