Improve view performance - sql

Question

I need to improve my view performance, right now the SQL that makes the view is:
select tr.account_number , tr.actual_collection_trx_date ,s.customer_key
from fct_collections_trx tr,
stg_scd_customers_key s
where tr.account_number = s.account_number
and trunc(tr.actual_collection_trx_date) between s.start_date and s.end_date;
Table fct_collections_trx has 170k+-(changes every day) records.
Table stg_scd_customers_key has 430mil records.
Table fct_collections_trx have indexes as following:
(SINGLE INDEX OF ALL OF THEM) (ACCOUNT_NUMBER, SUB_ACCOUNT_NUMBER, ACTUAL_COLLECTION_TRX_DATE, COLLECTION_TRX_DATE, COLLECTION_ACTION_CODE)(UNIQUE) and ENTRY_SCHEMA_DATE(NORMAL). DDL:
alter table stg_admin.FCT_COLLECTIONS_TRX
add primary key (ACCOUNT_NUMBER, SUB_ACCOUNT_NUMBER, ACTUAL_COLLECTION_TRX_DATE, COLLECTION_TRX_DATE, COLLECTION_ACTION_CODE)
using index
tablespace STG_COLLECTION_DATA
pctfree 10
initrans 2
maxtrans 255
storage
(
initial 80K
next 1M
minextents 1
maxextents unlimited
);
Table structure:
create table stg_admin.FCT_COLLECTIONS_TRX
(
account_number NUMBER(10) not null,
sub_account_number NUMBER(5) not null,
actual_collection_trx_date DATE not null,
customer_key NUMBER(10),
sub_account_key NUMBER(10),
schema_key VARCHAR2(10) not null,
collection_group_code CHAR(3),
collection_action_code CHAR(3) not null,
action_order NUMBER,
bucket NUMBER(5),
collection_trx_date DATE not null,
days_into_cycle NUMBER(5),
logical_delete_date DATE,
balance NUMBER(10,2),
abbrev CHAR(8),
customer_status CHAR(2),
sub_account_status CHAR(2),
entry_schema_date DATE,
next_collection_action_code CHAR(3),
next_collectin_trx_date DATE,
reject_key NUMBER(10) not null,
dwh_update_date DATE,
delta_type VARCHAR2(1)
)
Table stg_scd_customers_key have indexes : (SINGLE INDEX OF ALL OF THEM)
(ACCOUNT_NUMBER, START_DATE, END_DATE). DDL :
create unique index stg_admin.STG_SCD_CUST_KEY_PKP on stg_admin.STG_SCD_CUSTOMERS_KEY (ACCOUNT_NUMBER, START_DATE, END_DATE);
This table is also partitioned:
partition by range (END_DATE)
(
partition SCD_CUSTOMERS_20081103 values less than (TO_DATE(' 2008-11-04 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIAN'))
tablespace FCT_CUSTOMER_SERVICES_DATA
pctfree 10
initrans 1
maxtrans 255
storage
(
initial 8M
next 1M
minextents 1
maxextents unlimited
)
Table structure:
create table stg_admin.STG_SCD_CUSTOMERS_KEY
(
customer_key NUMBER(18) not null,
account_number NUMBER(10) not null,
start_date DATE not null,
end_date DATE not null,
curr_ind NUMBER(1) not null
)
I Can't add filter on the big table(need all range of dates) and i can't use materialized view. This query runs for about 20-40 minutes, I have to make it faster..
I've already tried to drop the trunc, makes no different.
Any suggestions?
Explain plan:

Answer 1

First, write the query using explicit join syntax:
select tr.account_number , tr.actual_collection_trx_date ,s.customer_key
from fct_collections_trx tr join
stg_scd_customers_key s
on tr.account_number = s.account_number and
trunc(tr.actual_collection_trx_date) between s.start_date and s.end_date;
You already have appropriate indexes for the customers table. You can try an index on fct_collections_trx(account_number, trunc(actual_collection_trx_date), actual_collection_trx_date). Oracle might find this useful for the join.
However, if you are looking for a single match, then I wonder if there is another approach that might work. How does the performance of the following query work:
select tr.account_number , tr.actual_collection_trx_date,
(select min(s.customer_key) keep (dense_rank first order by s.start_date desc)
from stg_scd_customers_key s
where tr.account_number = s.account_number and
tr.actual_collection_trx_date >= s.start_date
) as customer_key
from fct_collections_trx tr ;
This query is not exactly the same as the original query, because it is not doing any filtering -- and it is not checking the end date. Sometimes, though, this phrasing can be more efficient.
Also, I think the trunc() is unnecessary in this case, so an index on stg_scd_customers_key(account_number, start_date, customer_key) is optimal.
The expression min(x) keep (dense_rank first order by) essentially does first() -- it gets the first element in a list. Note that the min() isn't important; max() works just as well. So, this expression is getting the first customer key that meets the conditions in the where clause. I have observed that this function is quite fast in Oracle, and often faster than other methods.

Answer 2

If the start and end dates have no time elements (ie. they both default to midnight), then you could do:
select tr.account_number , tr.actual_collection_trx_date ,s.customer_key
from fct_collections_trx tr,
stg_scd_customers_key s
where tr.account_number = s.account_number
and tr.actual_collection_trx_date >= s.start_date
and tr.actual_collection_trx_date < s.end_date + 1;
On top of that, you could add an index to each table, containing the following columns:
for fct_collections_trx: (account_number, actual_collection_trx_date)
for stg_scd_customers_key: (account_number, start_date, end_date, customer_key)
That way, the query should be able to use the indexes rather than having to go to the table as well.

Answer 3

I suggest you an index based on the most selective field in your case
START_DATE, END_DATE
try reverting the (or adding a proper) index
START_DATE, END_DATE, ACCOUNT_NUMBER
in table stg_scd_customers_key