Order by two columns not using index in Oracle XE 11g - sql

Question

Why this simple query for retrieving first 100 rows, starting at given time, ordered by the time and primary key (client_time is not unique, that's why ordering by both), doesn't use the index?
SELECT *
FROM (SELECT *
FROM requests
WHERE client_time >= TO_TIMESTAMP('2017-07-01 10:00:00', 'YYYY-MM-DD HH24:MI:SS')
ORDER BY client_time ASC, transaction_id ASC
)
WHERE rownum <= 100;
client_time is TIMESTAMP WITH LOCAL TIME ZONE, transaction_id is VARCHAR2(255 CHAR).
Index I expect it to use is defined as
CREATE UNIQUE INDEX idx_time_id REQUESTS (client_time, transaction_id);
Query execution takes ca 2 seconds (6 milion rows in my system, will be much more in production) and yields following plan:
----------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes |TempSpc| Cost (%CPU)| Time |
----------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 100 | 110K| | 31237 (1)| 00:06:15 |
|* 1 | COUNT STOPKEY | | | | | | |
| 2 | VIEW | | 860K| 931M| | 31237 (1)| 00:06:15 |
|* 3 | SORT ORDER BY STOPKEY| | 860K| 65M| 86M| 31237 (1)| 00:06:15 |
|* 4 | TABLE ACCESS FULL | REQUESTS | 860K| 65M| | 15294 (1)| 00:03:04 |
----------------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
1 - filter(ROWNUM<=100)
3 - filter(ROWNUM<=100)
4 - filter("CLIENT_TIME">=TIMESTAMP' 2017-07-01 10:00:00,000000000')
When I remove the second part of my ORDER BY clause, this index actually is used and the query is executed in ca 1ms.
If I got this Use the index, Luke article right, shouldn't my query use this index too?
UPDATE:
The plan after removing my second order column looks like this:
--------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
--------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 100 | 65100 | 106 (0)| 00:00:02 |
|* 1 | COUNT STOPKEY | | | | | |
| 2 | VIEW | | 102 | 66402 | 106 (0)| 00:00:02 |
| 3 | TABLE ACCESS BY INDEX ROWID| TRX_REQUESTS_LTZ | 102 | 8160 | 106 (0)| 00:00:02 |
|* 4 | INDEX RANGE SCAN | IDX_TIME_ID | | | 3 (0)| 00:00:01 |
--------------------------------------------------------------------------------------------------
So I believe the WHERE clause is not the issue here. Also, nothing changes after rewriting the WHERE like this:
WHERE client_time >= TO_TIMESTAMP_TZ('2017-07-01 10:00:00 +10:00', 'YYYY-MM-DD HH24:MI:SS TZH:TZM')

Answer 1

I suspect the reason is the WHERE clause:
WHERE client_time >= TO_TIMESTAMP('2017-07-01 10:00:00', 'YYYY-MM-DD HH24:MI:SS')
You have specified that client_time is a TIMESTAMP WITH TIMEZONE. However, the constant is just a TIMESTAMP, without a timezone. This means that the types need to be converted -- and that often thwarts index usage.
You should try using TO_TIMESTAMP_TZ(), documented here.

Answer 2

Problem is that my index is built on VARCHAR2 column. Because of NLS, the index order can't be used just like that to sort the result set.
Changing transaction_id to NUMBER resolved the problem.