It seems that gin creation is triggering something nasty in the server that depends on previous history of the server.
If I vacuum full than drop the index and recreate it even with maintenance_work_mem='200MB' index creation may take forever. Stopping the execution may make vacuuming very slow or stopping the server very slow etc... I know that the server may be cleaning the new partially created index etc... but gin creation even with 200MB of maintenance_work_mem is having a too strange and to big hit on the server. I can *occasionally* succede to create a gin index in reasonable time if I shut down the server, vacuum full and drop and recreate the index separately from the transaction that load the data. gist creation is *predictably* much faster even inside the transaction that load the data and doesn't have side effect outside index creation. Even when gin creation succede it is definitively slower than 3x gist/gin index creation looks more cpu bounded than memory bounded. I'm checking if I made some mistake in other cfg parameters that may have some impact on index creation... Any further clue? -- Ivan Sergio Borgonovo http://www.webthatworks.it -- Sent via pgsql-general mailing list (pgsql-general@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-general
