Schizophrenic sql or server?

Question

Thanks for taking my question!
We have 3 oracle servers (10.2.0.3). Each server has the same tables, data, and indexes but each can start acting different for no apparent reason. Sql that has run for months goes bad on our production and test server while the production's backup server works fine.
In this case, I can fix the problem (as described below) but don’t understand what went wrong or why my change fixed the problem. Each server collects stats nightly so indexes and tables are not stale.
Anyone have any ideas why sql is breaking?
Thank You! Kathie
BACKGOUND:
Sql works fine on 1 server and not on the other 2 servers (as I said before - same data and indexes on each server and sql has not changed on any of the servers).
TO FIX FULL TABLE SCAN:
MOVED: JOIN name_record n ON s.sr_key = n.nr_key AND n.nr_inactive_flag = 0
ADDED INDEX: (nr_key, nr_inactive_flag)
Already had an index with nr_key (which is unique) so it doesn't make since to add the new index but if I remove it the index the full table scan comes back.
*THIS IS THE ORIGINAL STATEMENT
SELECT ...sch.sch_name, sch.sch_state, v1.val_9pos AS sr_class_9,
v2.val_25pos AS sr_hsg_status_25, v3.val_25pos AS sr_hsg_list_25,
v4.val_9pos AS sr_selective_svc_9,
v5.val_madison AS sr_direct_deposit_10,
v6.val_3pos AS sr_for_lang_comp_3,
vtm1.vtm_13_display AS sr_matriculation_13,
vtm2.vtm_13_display AS last_ug_term,
vtm3.vtm_13_display AS last_g_term
FROM student_record s
JOIN user_names usr ON s.sr_public_w_check = usr.usr_key AND usr.usr_user_name = 'StudentUserID'
JOIN name_record n ON s.sr_key = n.nr_key AND n.nr_inactive_flag = 0JOIN value_terms vtm1 on s.sr_matriculation = vtm1.vtm_term
JOIN value_terms vtm2 ON s.sr_last_ug_term = vtm2.vtm_term
JOIN value_terms vtm3 ON s.sr_last_g_term = vtm3.vtm_term
JOIN value_master v1
ON v1.val_field = 205003 AND s.sr_class = v1.val_value_number
JOIN value_master v2
ON v2.val_field = 201204 AND s.sr_hsg_status = v2.val_value_number
JOIN value_master v3
ON v3.val_field = 201202 AND s.sr_hsg_list = v3.val_value_number
JOIN value_master v4
ON v4.val_field = 201173 AND s.sr_selective_svc = v4.val_value_number
JOIN value_master v5
ON v5.val_field = 201407 AND s.sr_direct_deposit = v5.val_value_number
JOIN value_master v6
ON v6.val_field = 201154 AND s.sr_for_lang_comp = v6.val_value_number
>>>> to here <<<<
LEFT JOIN school_records sch ON s.sr_high_school =sch.SCH_TYPE_AND_CODE

Answer 1

each can start acting different for no apparent reason. Sql that has run for
months goes bad on our production and test server while the production's
backup server works fine.What sort of "bad"? Please describe the behaviour in more detail.
Each server collects stats nightly so indexes and tables are not stale. The different usages of the servers might suggest they would have different rates of data change and hence might have different stats. In general collecting stats every night is usually bad practice: either the change in the stats is marginal - in which case why bother - or it can be severe, in whcih case it might have a deleterious effect on previously performant code (although I agree that is unlikely).
Cheers, APC
blog: http://radiofreetooting.blogspot.com