Having done that, I was all set to trash the code when I wondered if it might be useful in its own right...
Here is a sample of the output after a pgbench run with shared_buffers=500:
bench=>dumpcache;
CACHE DUMP
-----------------------------------------------------------------
pg_type_oid_index 2
pg_proc 1
history 3
pg_class_relname_nsp_index 5
pg_statistic_relid_att_index 2
tellers_pkey 9
<invalid oid> 4
accounts 205
pg_am 1
pg_statistic 1
branches 2
accounts_pkey 215
pg_amproc_opc_procnum_index 2
branches_pkey 9
pg_operator_oid_index 3
tellers 4
pg_class_oid_index 2
pg_type 2
pg_index_indexrelid_index 2
pg_trigger_tgrelid_tgname_index 2
pg_amop_opr_opc_index 2
pg_index_indrelid_index 2
pg_class 2
pg_trigger 1
pg_proc_oid_index 1
pg_amop_opc_strategy_index 2
pg_attribute_relid_attnum_index 4
pg_amop 1
pg_amproc 1
pg_index 2
pg_operator 3
pg_attribute 3
(32 rows)bench=>
Does this seem like a useful thing to be able to display ?
I implemented this by adding a command (as I wanted to know how this was done), but I suspect it would make more sense to use a function a bit like the pg_stat* collection.
I have included (most of) the code I used, so that interested parties can show me what I have done wrong :-)
(Some things I wondered about were : should I be locking buffers before peeking at the corresponding descriptors?, does it make sense to call RelationIdGetRelation on a relNode?.... )
regards
Mark
dumpcache.tar.gz
Description: application/macbinary
---------------------------(end of broadcast)--------------------------- TIP 7: don't forget to increase your free space map settings
