Oracle Scratchpad

April 25, 2014

Modify PK – 2

Filed under: Indexing,Oracle — Jonathan Lewis @ 5:48 pm BST Apr 25,2014

In an earlier posting I described how we can play games with primary key indexes in 12c because you can create multiple indexes on a table for the same (ordered) column list provided they have some differences in attributes and only one of them is visible at a time. But how, if you’re not on 12c, can you change a primary key index from unique to non-unique (or vice versa, as this question on OTN wants) without any down-time ?

Of course you can’t “change” the uniqueness of an index – that attribute embedded in the way that Oracle stores rowids for the index – but you can create a new index for the constraint; and you can’t avoid the little bit of “not quite down”-time that it takes to start and finish an online rebuild. But how do you get around the limitation of Oracle error: ORA-01408: such column list already indexed

Easy – though it requires you to do some work you would prefer to avoid – just work through an intermediate step using an index with an extra column. Here’s an example where I start with a non-unique index supporting the PK and end up with a unique index.

create table t1 as select * from all_objects where rownum <= 10000;

create index t1_pk on t1(object_id);

alter table t1 add constraint t1_pk primary key(object_id);

create index t1_i1 on t1(object_id, 0) online;

alter table t1 modify primary key using index t1_i1;

drop index t1_pk;

create unique index t1_pk on t1(object_id) online;

alter table t1 modify primary key using index t1_pk;

drop index t1_i1;

April 18, 2014

Bitmap loading

Filed under: bitmaps,Indexing,Oracle — Jonathan Lewis @ 12:43 pm BST Apr 18,2014

Everyone “knows” that bitmap indexes are a disaster (compared to B-tree indexes) when it comes to DML. But at an event I spoke at recently someone made the point that they had observed that their data loading operations were faster when the table being loaded had bitmap indexes on it than when it had the equivalent B-tree indexes in place.

There’s a good reason why this can be the case.  No prizes for working out what it is – and I’ll supply an answer in a couple of days time.  (Hint – it may also be the reason why Oracle doesn’t use bitmap indexes to avoid the “foreign key locking” problem).


As Martin (comment 3) points out, there’s a lot of interesting information in the statistics once you start doing the experiment. So here’s some demonstration code, first we create a table with one of two possible indexes:

create table t1
with generator as (
	select	--+ materialize
		rownum id
	from dual
	connect by
		level <= 1e4
	rownum			id,
	mod(rownum,1000)	btree_col,
	mod(rownum,1000)	bitmap_col,
	rpad('x',100)		padding
	generator	v1,
	generator	v2
	rownum <= 1e6

		ownname		 => user,
		tabname		 =>'T1',
		method_opt	 => 'for all columns size 1'

create        index t1_btree on t1(btree_col) nologging;
-- create bitmap index t1_bitmap on t1(bitmap_col) nologging;

You’ll note that the two columns I’m going to build indexes on hold the same data in the same order – and it’s an order with maximum scatter because of the mod() function I’ve used to create it. It’s also very repetitive data, having 1000 distinct values over 1,000,0000 rows. With the data and (one of) the indexes in place I’m going to insert another 10,000 rows:

execute snap_my_stats.start_snap

insert /* append */ into t1
with generator as (
	select	--+ materialize
		rownum id
	from dual
	connect by
		level <= 1e4
	1e6 + rownum		id,
	mod(rownum,1000)	btree_col,
	mod(rownum,1000)	bitmap_col,
	rpad('x',100)		padding

execute snap_my_stats.end_snap

You’ll note that I’ve got an incomplete append hint in the code – I’ve tested the mechanism about eight different ways, and left the append in as a convenience, but the results I want to talk about (first) are with the hint disabled so that the insert is a standard insert. The snap_my_stats calls are my standard mechanism to capture deltas of my session statistics (v$mystat) – one day I’ll probably get around to using Tanel’s snapper routine everywhere – and here are some of the key results produced in the two tests: with btree
Name                                                                     Value
----                                                                     -----
session logical reads                                                   31,403
DB time                                                                     64
db block gets                                                           31,195
consistent gets                                                            208
db block changes                                                        21,511
redo entries                                                            10,873
redo size                                                            3,591,820
undo change vector size                                                897,608
sorts (memory)                                                               2
sorts (rows)                                                                 1 with bitmap
Name                                                                     Value
----                                                                     -----
session logical reads                                                   13,204
DB time                                                                     42
db block gets                                                            8,001
consistent gets                                                          5,203
db block changes                                                         5,911
redo entries                                                             2,880
redo size                                                            4,955,896
undo change vector size                                              3,269,932
sorts (memory)                                                               3
sorts (rows)                                                            10,001

As Martin has pointed out, there are a number of statistics that show large differences between the B-tree and bitmap approaches, but the one he didn’t mention was the key: sorts (rows). What is this telling us, and why could it matter so much ? If the B-tree index exists when the insert takes place Oracle locates the correct place for the new index entry as each row is inserted which is why you end up with so many redo entries, block gets and block changes; if the bitmap index exists, Oracle postpones index maintenance until the table insert is complete, but accumulates the keys and rowids as it goes then sorts them to optimize the rowid to bitmap conversion and walks the index in order updating each modified key just once.

The performance consequences of the two different strategies depends on the number of indexes affected, the number of rows modified, the typical number of rows per key value, and the ordering of the new data as it arrives; but it’s possible that the most significant impact could come from ordering.  As each row arrives, the relevant B-tree indexes are modified – but if you’re unlucky, or have too many indexes on the table, then each index maintenance operation could result in a random disk I/O to read the necessary block (how many times have you seen complaints like: “we’re only inserting 2M rows but it’s taking 45 minutes and we’re always waiting on db file sequential reads”). If Oracle sorts the index entries before doing the updates it minimises the random I/O because it need only update each index leaf block once and doesn’t run the risk of re-reading many leaf blocks many times for a big insert.

Further Observations

The delayed maintenance for bitmap indexes (probably) explains why they aren’t used to avoid the foreign key locking problem.  On a large insert, the table data will be arriving, the b-tree indexes will be maintained in real time, but a new child row of some parent won’t appear in the bitmap index until the entire insert is complete – so another session could delete the parent of a row that exists, is not yet committed, but is not yet visible. Try working out a generic strategy to deal with that type of problem.

It’s worth noting, of course, that when you add the /*+ append */ hint to the insert then Oracle uses exactly the same optimization strategy for B-trees as it does for bitmaps – i.e. postpone the index maintenance, remember all the keys and rowids, then sort and bulk insert them.  And when you’ve remembered that, you may also remember that the hint is (has to be) ignored if there are any enabled foreign key constraints on the table. The argument for why the hint has to be ignored and why bitmap indexes don’t avoid the locking problem is (probably) the same argument.

You may also recall, by the way, that when you have B-tree indexes on a table you can choose the optimal update or delete strategy by selecting a tablescan or index range scan as the execution path.  If you update or delete through an index range scan the same “delayed maintenance” trick is used to optimize the index updates … except for any indexes being used to support foreign key constraints, and they are maintained row by row.

In passing, while checking the results for this note I re-ran some tests that I had originally done in 2006 and added one more test that I hadn’t considered at the time; as a result I can also point out that index will see delayed maintenance if you drive the update or delete with an index() hint, but not if you drive it with an index_desc() hint.


April 2, 2014

Easy – Oops.

Filed under: Bugs,Function based indexes,Indexing,Oracle — Jonathan Lewis @ 7:47 pm BST Apr 2,2014

A question came up on OTN today asking for suggestions on how to enforce uniqueness on a pair of columns only when the second column was not null. There’s an easy and obvious solution – but I decided to clone the OP’s example and check that I’d typed my definition up before posting it; and the result came as a bit of a surprise. Here’s a demo script (not using the OP’s table):

create table t1  
	col1	int not null,
	col2	varchar2(1)

create unique index t1_i1 on t1( 
--	case col2 when null then cast(null as int) else col1 end,
--	case when col2 is null then cast(null as int) else col1 end,
	case when col2 is not null then col1 end,

insert into t1 values(1,null);
insert into t1 values(1,null);
insert into t1 values(1,'x');
insert into t1 values(1,'y');
insert into t1 values(1,'y');


column ind1_is   format a5
column ind1_when format 9999

set null N/A

	case when col2 is null then cast (null as int) else col1 end	ind1_is,
	case col2 when null then cast (null as int)  else col1 end	ind1_when

The strategy is simple, you create a unique function-based index with two columns; the first column of the index id defined to show the first column of the table if the second column of the table is not null, the second column of the index is simply the second column of the table. So if the second column of the table is null, both columns in the index are null and there is no entry in the index; but if the second column of the table is not null then the index copies both columns from the table and a uniqueness test applies.

Based on the requirement and definition you would expect the first 4 of my insert statements to succeed and the last one to fail. The index will then have two entries, corresponding to my 3rd and 4th insertions.

I’ve actually shown three ways to use the case statement to produce the first column of the index. The last version is the cleanest, but the first option is the one I first thought of – it’s virtually a literal translation the original requirement. The trouble is, with my first definition the index acquired an entry it should not have got, and the second insert raised a “duplicate key” error; the error didn’t appear when I switched the syntax of the case statement to the second version.

That’s why the closing query of the demo is there – when you run it the two values reported should be the same as each other for all four rows in the table – but they’re not. This is what I got on

------- ---------
N/A             1
N/A             1
      1         1
      1         1

I’m glad I did a quick test before suggesting my original answer.

Anyone who has other versions of Oracle available is welcome to repeat the test and report back which versions they finding working correctly (or not).


It’s not a bug (see note 2 below from Jason Bucata), it’s expected behaviour.



Filed under: Function based indexes,Indexing,Oracle — Jonathan Lewis @ 6:24 pm BST Apr 2,2014

The following question came up on OTN recently:

Which one gives better performance? Could please explain.

1) nvl( my_column, ‘N’) <> ‘Y’

2) nvl( my_column, ‘N’) = ‘N’

It’s a question that can lead to a good 20 minute discussion – if you were in some sort of development environment and had a fairly free hand to do whatever you wanted.

The most direct answer is that you could expect the performance to be the same whichever option you chose – but the results might be different, of course, unless you had a constraint on my_column that ensured that it would hold only null, ‘Y’, or ‘N’.  (Reminder:  the constraint check (my_column in (‘Y’,’N’) will allow nulls in the column).

On the other hand, if you create a function-based index on nvl(my_column,’N’) then the second option would give the optimizer the opportunity to use the index – which could make it the more efficient option if a sufficiently small quantity of the data from a large table matched the predicate. Of course in this case you would need a histogram on the hidden column definition supporting the index so that the optimizer could detect the data skew.

But if a function-based index is a step in the right direction it’s worth thinking about how to pick the best function-based index.

create index t1_i1 on t1(case my_column when 'Y' then null else 'N' end);

execute dbms_stats.gather_table_stats(user,'t1',method_opt=&gt;'for all columns size 1');

select * from t1 where case my_column when 'Y' then null else 'N' end = 'N';

Provided you can change the queries that use the original predicate, you can create a function-based index that is the smallest possible index to assist your query, and you won’t need a histogram to allow the optimizer to get the correct cardinality since there will be just one distinct value in the index (and the underlying column).

It’s possible to go a little further – what if the typical query that started this discussion was really something like:

select  *
from    t1
where   nvl(my_column, 'N') = 'N'
and     other_column = {some value}

If the nvl() predicate always returned a small fraction of the data we might engineer an index to handle the requirement of the entire where clause:

create index t1_i2 on t1(case my_column when 'Y' then null else other_column end);
execute dbms_stats.gather_table_stats(user,'t1',method_opt=&gt;'for all hidden columns size 254');

In this case you might need a histogram on the hidden column definition supporting the index, which is why I’ve changed my method_opt. We’ve constructed an index that is the smallest possible index that will satisfy the query as precisely as possible while giving the optimizer the best chance of producing accurate cardinality estimates. [Update: until the appearance of Kirill’s comment (#1) I forgot to point out that once again you have to change the original SQL from using the nvl() expression to using a case expression that matches the index.]


It would be easy to fall into the trap of playing around with this type of tweaking every time you hit a performance problem. Bear in mind, though, that even indexes which are as small and accurately targeted as possible can introduce various types of overhead (contention, cost of gathering stats, instability of histogram collection); so always try to think through the potential costs as well as the benefits of such an approach – is it really worth the effort.

It’s also worth noting that things don’t always work the way you expect.

Finally, of course, you might choose to create “proper” virtual columns in 11g so that you can refer to them by name, rather than handling them indirectly as you have to when they are “undocumented” hidden columns generated by function-based indexes.


March 24, 2014


Filed under: Function based indexes,Indexing,Oracle — Jonathan Lewis @ 11:11 pm BST Mar 24,2014

One of my most-repeated observations about trouble-shooting Oracle is that things break when you start combining features. Here’s an example that demonstrates the point.

It’s possible to create “descending” indexes – or indexes with descending columns, as I prefer to call them, and there’s a special “min/max range scan” optimizer operation for a particular kind of index usage – demonstrated in the following code fragment (running under, and reporting the rowsource execution statistics):

March 13, 2014

Shrink Space

Filed under: Bugs,deadlocks,fragmentation,Index Rebuilds,Indexing,Locks,Oracle — Jonathan Lewis @ 7:08 am BST Mar 13,2014

Here’s an example of a nasty accident that can be seen in a slightly unusual output from v$lock (on

March 2, 2014

Auto Sample Size

Filed under: Function based indexes,Indexing,Infrastructure,IOT,LOBs,Oracle,Statistics — Jonathan Lewis @ 6:38 pm BST Mar 2,2014

In the past I have enthused mightily about the benefits of the approximate NDV mechanism and the benefit of using auto_sample_size to collect statistics in 11g; however, as so often happens with Oracle features, there’s a down-side or boundary condition, or edge case. I’ve already picked this up once as an addendum to an earlier blog note on virtual stats, which linked to an article on OTN describing how the time taken to collect stats on a table increased dramatically after the addition of an index – where the index had this definition:

create bitmap index i_s_rmp_eval_csc_msg_actions on
    s_rmp_evaluation_csc_message (
        decode(instr(xml_message_text,' '),0,0,1)

As you might guess from the column name, this is an index based on an XML column, which is stored as a CLOB.

In a similar vein, I showed you a few days ago an old example I had of indexing a CLOB column with a call to dbms_lob.getlength(). Both index examples suffer from the same problem – to support the index Oracle creates a hidden (virtual) column on the table that can be used to hold statistics about the values of the function; actual calculated values for the function call are stored in the index but not on the table itself – but it’s important that the optimizer has the statistics about the non-existent column values.


February 25, 2014

FBI Skip Scan

Filed under: Bugs,Function based indexes,Indexing,Oracle — Jonathan Lewis @ 6:45 pm BST Feb 25,2014

A recent posting on the OTN database forum highlighted a bug (or defect, or limitation) in the way that the optimizer handles index skip scans with “function-based” indexes – it doesn’t do them. The defect has probably been around for a long time and demonstrates a common problem with testing Oracle – it’s very easy for errors in the slightly unusual cases to be missed; it also demonstrates a general principle that it can take some time for a (small) new feature to be applied consistently across the board.

The index definitions in the original posting included expressions like substr(nls_lower(colX), 1, 25), and it’s possible for all sorts of unexpected effects to appear when your code starts running into NLS  settings, so I’ve created a much simpler example. Here’s my table definition, with three index definitions:


February 21, 2014

Indexing LOBs

Filed under: Function based indexes,Indexing,Infrastructure,LOBs,Oracle — Jonathan Lewis @ 6:42 pm BST Feb 21,2014

Many years ago, possibly when most sites were still using Oracle 8i, a possible solution to a particular customer problem was to create a function-based index on a CLOB column using the dbms_lob.getlength() function call. I can’t find the notes explaining why this was necessary (I usually have some sort of clue – such as the client name – in the script, but in this case all I had was a comment that “the manuals say you can’t do this, but it works provided you wrap the dbms_lob call inside a deterministic function”).

I never worked out why the dbms_lob.getlength() function wasn’t declared as deterministic – especially since it came complete with a most restrictive restricts_references pragma – so I had just assumed there was probably some good reason based on strange side effects when national language charactersets came into play. But here’s a little detail I noticed recently about the dbms_lob.getlength() function: it became deterministic in 11g, so if the client decided to implement my suggestion (which included the usual sorts of warnings) it’s now legal !

Footnote – the length() function has been deterministic and usable with LOBs for a long time, certainly since late 9i, but in 8i length(lob_col) will produce Oracle error “ORA-00932: inconsistent datatypes”

Index Compression – aargh

Filed under: Bugs,compression,Indexing,Infrastructure,Oracle — Jonathan Lewis @ 7:57 am BST Feb 21,2014

The problem with telling people that some feature of Oracle is a “good thing” is that some of those people will go ahead and use it; and if enough people use it some of them will discover a hitherto undiscovered defect. Almost inevitably the bug will turn out to be one of those “combinations” bugs that leaves you thinking: “Why the {insert preferred expression of disbelief here} should {feature X} have anything to do with {feature Y}”.

Here – based on index compression, as you may have guessed from the title – is one such bug. I got it first on, but it’s still there on and


February 4, 2014


Filed under: Indexing,Oracle — Jonathan Lewis @ 1:33 am BST Feb 4,2014

To create an index on a table (with no existing indexes) Oracle has to start by doing a tablescan.

What’s the difference between the tablescan it uses for a B-tree index and the tablescan it uses for a bitmap index ? Why ?


I was going to give a hint that if you answered the “why” first that might lead you to the right idea and a test for the “what”, but we already have an answer, with a sample of proof.

February 1, 2014

Modify PK

Filed under: Indexing,Oracle — Jonathan Lewis @ 11:00 am BST Feb 1,2014

Sitting in the lounge waiting to be called for my flight I was musing on the 12c feature of having multiple indexes defined on the same ordered column set  when a thought crossed my mind and I decided to run a little test that looked like this:

create table t1 as select * from all_objects where rownum <= 10000;

create unique index t1_pk on t1(object_id);

alter table t1 add constraint t1_pk primary key(object_id);

create index t1_i1 on t1(object_id, object_name);

drop index t1_pk;

    expect ORA-02429: cannot drop index used for enforcement of unique/primary key

alter table t1 modify primary key using index t1_i1;

drop index t1_pk;

For years I’ve been assuming that you really have to mess around with the PK (and any related FKs) if you want to change the index supporting the primary key – but this code demonstrates that you can add a new index to a table and “move” the primary key to it before dropping the original index.

The worrying thing about this (for me, at any rate) is that it isn’t a new feature – after testing it on I started working backwards, and it works down to (the earlist 9i I have access to). It doesn’t work on, and the version behaves slightly differently from later versions because the original PK index disappears as the constraint is moved.

As I’ve often said about trust – keep an eye on the date and version of any article you read, it may no longer be true.

January 26, 2014


Filed under: Bugs,Indexing,Oracle,Troubleshooting,Tuning — Jonathan Lewis @ 12:08 pm BST Jan 26,2014

I was involved in a thread on Oracle-L recently started with the question: “How many LIOs is too many LIOs”. Rather than rewrite the whole story, I’ve supplied a list of links to the contributions I made, in order – the final “answer” is actually the answer to a different question – but travels an interesting path to get there.#

I’ve got a script to emulate the requirement so that people can see for themselves the bug that I mention in post 15; I’ll try to add a couple of notes to it and publish it some time, but for the moment I’ll just remind myself that it’s called (slightly counter-intuitively: no_sort_problem.sql)

January 17, 2014

Bitmap question

Filed under: bitmaps,Indexing,Oracle — Jonathan Lewis @ 7:06 pm BST Jan 17,2014

If you know anything about bitmap indexes you probably know that a single entry in a bitmap index takes the form (key_value, starting rowid, ending rowid, BBC compressed bit string). So an entry covers a single value for a column over a range of rowids  in the table, and the string of bits for that (notional) range is reduce to a minimum by a compression mechanism that eliminate repeated zeros in multiples of 8.

So here’s a question – to which I don’t know the answer, although you may be surprised when you try to find it:

If you have a very large table and in one of its columns the first row and the last row (and no others) hold the value 0 (say) and you create a bitmap index on this column, what’s the largest number of rows you could have in the table before Oracle would HAVE to create two index entries in order to cover both rows ?

Follow-up question – once you start getting close to working out the answer, can you think of a way to provide an example without actually creating a table with that many rows in it ?


January 3, 2014

Index Hash

Filed under: Bugs,CBO,Hints,Ignoring Hints,Index Joins,Indexing,Oracle — Jonathan Lewis @ 6:56 pm BST Jan 3,2014

I’m afraid this is one of my bad puns again – an example of the optimizer  making a real hash of the index hash join. I’m going to create a table with several indexes (some of them rather similar to each other) and execute a query that should do an index join between the obvious two indexes. To show how obvious the join should be I’m going to start with a couple of queries that show the cost of simple index fast full scans.

Here’s the data generating code:


« Previous PageNext Page »

Blog at