An alternative to "drop empty segment" would be to do a back-door
invocation of
"alter table tbl_test2 truncate partition maxvalue drop all storage;"
This (at least for 19.11) discards the segments but keeps the partition.
The equivalent for composite partitions would be messier, of course.
"Empty" is a bit tricky, of course, since there's a difference between
"empty never been used" and "empty because it's all deleted". For partition
maintenance Oracle already has a mechanism for knowing when a partition
generated by a split is going to be empty, so that's the one case where the
truncate option could be injected.
Regards
Jonathan Lewis
On Fri, 23 Jul 2021 at 13:04, Mark W. Farnham <mwf@xxxxxxxx> wrote:
It seems likely that “create” fixes in place the “deferred” mode at a
different point in the code than maintenance. That’s a complete guess since
I don’t get to see the code, but it is compatible with “create” correctly
not creating “deferred” segments in the cases where it might have to create
an out of line segment.
Possibly it would be easier to create a “drop empty segments” maintenance
that to insert the correction to deferred segments. Since at least
logically Oracle could know range partition key references refer to a
single partition by name instead of stats more easily, that might be better
for the optimizer using partition stats. And, of course, clearing up
clutter in the dictionary as well as the possibly designed large initial
segment sizes. Gargantuan segments for Monday through Friday and no
segments at all for Saturday and Sunday is still a common pattern, even in
the days of globalization.
So this message is mostly to the Oracle lurkers: Consider a drop empty
segments between <low partition key reference> and <high partition key
reference> as a fix up if the maintenance code is too different from the
create code to pattern match the “bug except for restriction documentation”
behavior.
Quite often such restrictions are because resources could not be
prioritized rather than it being arcane to implement without the
restriction.
Since this behavior has been around, documented, for a long time, if you
change it a lot of folks will be asking for the “cleanup” code anyway.
Good luck.
Thanks Gerald and JL for pointing out it has “always” been this way,
ending the mystery.
mwf
*From:* oracle-l-bounce@xxxxxxxxxxxxx [mailto:
oracle-l-bounce@xxxxxxxxxxxxx] *On Behalf Of *Jonathan Lewis
*Sent:* Friday, July 23, 2021 4:38 AM
*To:* list, oracle
*Subject:* Re: [External] : Re: small bug related to
deferred_segment_creation parameter
I know that Gerald Venzi quoted the 21c reference manual, and going so far
forward to "retro-document" a detail can look like a cop-out; but in this
case you can find exactly the same text in the 11gR2 reference manual -
search for "Deferred segment creation" at the URL:
https://docs.oracle.com/cd/E11882_01/server.112/e41084/statements_3001.htm#CJABFJEA
Regards
Jonathan Lewis
On Fri, 23 Jul 2021 at 07:45, ahmed.fikri@xxxxxxxxxxx <
ahmed.fikri@xxxxxxxxxxx> wrote:
Hi all,
My impression is that every time Oracle detects a bug it doesn't seem easy
to resolve, add two lines of documentation that says this is a restriction.
I've seen some examples of this on this forum. We even have conflicting
documentation, one saying this is a limitation, the other saying it's a
feature, and so on.
But that's just my impression, and doesn't mean it is true.
Many thanks to Jonathan and Andy.
Best regards
Ahmed
-----Original-Nachricht-----
Betreff: Re: [External] : Re: small bug related to
deferred_segment_creation parameter
Datum: 2021-07-23T02:44:53+0200
Von: "Gerald Venzl" <gerald.venzl@xxxxxxxxxx>
An: "ahmed.fikri@xxxxxxxxxxx" <ahmed.fikri@xxxxxxxxxxx>
Hi all,
It is reproducible not only in 19.11 but will also be in 21c as this is
not a bug but a documented restriction (see alter_table_partitioning
section)
<https://docs.oracle.com/en/database/oracle/oracle-database/21/sqlrf/ALTER-TABLE.html>:
- Deferred segment creation is not supported for partition maintenance
operations that create new segments on tables with LOB columns; segments
will always be created for the involved (sub)partitions.
Furthermore, you can see that this restriction holds true for all
partition maintenance operations and not only the SPLIT operation.
Nope this helps!
Thx,
---------------
Gerald Venzl | Distinguished Product Manager
Email: gerald.venzl@xxxxxxxxxx | Phone: +1.650.633.0085
Oracle ST & Database Development
400 Oracle Parkway | Redwood Shores | 94065 | USA
On Jul 22, 2021, at 15:14, Jonathan Lewis <jlewisoracle@xxxxxxxxx> wrote:
It looks like you missed the point in my second paragraph:
*>> However, it may be a little more interesting to point out that this
example does *
*>> the same thing in 19.11.0.0 - and still creates the segments even if
you include *
*>> the "segment creation deferred" option with each of the listed
partitions in the split. *
Regards
Jonathan Lewis
On Thu, 22 Jul 2021 at 20:23, ahmed.fikri@xxxxxxxxxxx <
ahmed.fikri@xxxxxxxxxxx> wrote:
in the current project I'm working on, 12.1 is still in use. Maybe someone
can test the script in 12.2 or 19.11.
In the meantime, I'll try to remove some dust from my private 12.2 and 19
instances and see if we have the same behavior.
Best regards
Ahmed
-----Original-Nachricht-----
Betreff: Re: small bug related to deferred_segment_creation parameter
Datum: 2021-07-22T20:14:53+0200
Von: "Jonathan Lewis" <jlewisoracle@xxxxxxxxx>
An: "ahmed.fikri@xxxxxxxxxxx" <ahmed.fikri@xxxxxxxxxxx>
I'm not surprised that you didn't get a response in less than 3 hours -
especially since you were talking about 11.2.0.4 and 12.1.0.2.
However, it may be a little more interesting to point out that this
example does the same thing in 19.11.0.0 - and still creates the segments
even if you include the "segment creation deferred" option with each of the
listed partitions in the split.
Regards
Jonathan Lewis
On Thu, 22 Jul 2021 at 12:53, ahmed.fikri@xxxxxxxxxxx <
ahmed.fikri@xxxxxxxxxxx> wrote:
Hi everyone,
it seems that the deferred_segment_creation parameter does not work
properly when splitting a table partition that contains a clob field.
If the partition is empty and has no segment, splitting it creates two
segments for the old and new partitions.
I tested this using this block in 11.2.0.4 and 12.1.0.2 (also attached and
run with a non dba user):
DECLARE
PROCEDURE ex(cmd VARCHAR2) IS BEGIN dbms_output.put_line(cmd); EXECUTE
IMMEDIATE cmd; END;
PROCEDURE ex_ignore_error(cmd VARCHAR2) IS BEGIN ex(cmd); EXCEPTION WHEN
OTHERS THEN NULL; END;
PROCEDURE split_part(tbl VARCHAR2,p_part_val VARCHAR2) IS BEGIN
ex('alter table '||tbl||' split partition maxvalue at
(to_date('''||p_part_val||''',''yyyymmdd''))
into (partition part_'||p_part_val||' tablespace users, partition
maxvalue tablespace users)') ;END;
BEGIN
ex('alter session set deferred_segment_creation = true');
ex_ignore_error('drop table tbl_test purge');
ex_ignore_error('drop table tbl_test2 purge');
/* create range-partitioned table without clob field */
ex('create table tbl_test(id number, part_key date) tablespace users
partition by range(part_key) (partition maxvalue values less than
(maxvalue))');
/* create range-partitoned table having a clob field */
ex('create table tbl_test2(id number, part_key date, clob_col clob)
tablespace users partition by range(part_key) (partition maxvalue values
less than (maxvalue))');
/* the table is empty. Splitting the maxvalue partition for the table
without clob */
split_part('tbl_test','20200101');
/* until here no segments is generated for tbl_test and tbl_test2 tables
spltting the maxvalue partition for the table having clob field */
split_part('tbl_test2','20200101');
-- two segments are created for the table having clob field
/* SELECT * FROM user_segments WHERE segment_name IN
('TBL_TEST','TBL_TEST2')
TBL_TEST2 PART_20200101 TABLE PARTITION ASSM USERS
TBL_TEST2 MAXVALUE TABLE PARTITION ASSM USERS
*/
END;
Is this already known? Any Idea?
For me, this behavior is annoying and is considered a bug.
Best regards
Ahmed
