[
https://issues.apache.org/jira/browse/DERBY-1213?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Dag H. Wanvik updated DERBY-1213:
---------------------------------
Issue Type: Improvement (was: Test)
> create new space reclamation test based on functionality removed from
> access.sql test
> -------------------------------------------------------------------------------------
>
> Key: DERBY-1213
> URL: https://issues.apache.org/jira/browse/DERBY-1213
> Project: Derby
> Issue Type: Improvement
> Components: Store, Test
> Affects Versions: 10.2.1.6
> Reporter: Mike Matrigali
> Priority: Minor
>
> In order to get rid of the inconsistent regression test results reported in
> DERBY-990 and DERBY-1210, removed the following
> from the access.sql script. This methodology has been problematic and
> multiple "hacks" have been put in to try and get the
> results reproducible. The results are very timing dependent on background
> threads. The test probably should be rewritten
> as a .java program and at the very least report the differences in space
> management in the error (rather than the lazy approach
> used because logic was not possible in a .sql script).
> Here is the original scripting:
> --- buddy test case for reclaiming deleted rows during split.
> --- test space reclamation using the SpaceTable VTI.
> --- o insert bunch of rows with sequential keys.
> --- o create non-unique index.
> --- o check space usage for index using SpaceTable VTI
> --- o delete every other one less than 1000 - this will make normal post
> --- commit not fire as no page should have all deleted rows (< 1000
> avoids
> --- possible case where last page only has 1 row).
> --- o commit
> --- o now reinsert rows into the "holes" which before the fix would cause
> --- splits, but now will force reclaim space and reuse existing space in
> --- btree.
> --- o redo delete, commit and insert several times over
> --- o check space usage for index using SpaceTable VTI is same as above
> --- set page size to default.
> -call SYSCS_UTIL.SYSCS_SET_DATABASE_PROPERTY('derby.storage.pageSize',
> '4096');
> -call
> SYSCS_UTIL.SYSCS_SET_DATABASE_PROPERTY('derby.storage.minimumRecordSize',
> '1');
> -
> -autocommit off;
> -commit;
> -drop table foo;
> -drop table foo2;
> -
> --- create and load a table with values from 1024 down to 1,
> -create table foo (a int, b char(200), c int);
> -
> -insert into foo values (1024, 'even', 0);
> -insert into foo (select foo.a - 1, 'odd' , 1 from foo);
> -
> -insert into foo (select foo.a - 2, foo.b, foo.c from foo);
> -insert into foo (select foo.a - 4, foo.b, foo.c from foo);
> -insert into foo (select foo.a - 8, foo.b, foo.c from foo);
> -insert into foo (select foo.a - 16, foo.b, foo.c from foo);
> -insert into foo (select foo.a - 32, foo.b, foo.c from foo);
> -insert into foo (select foo.a - 64, foo.b, foo.c from foo);
> -insert into foo (select foo.a - 128, foo.b, foo.c from foo);
> -insert into foo (select foo.a - 256, foo.b, foo.c from foo);
> -insert into foo (select foo.a - 512, foo.b, foo.c from foo);
> -
> --- insert into the "holes", but different keys (even2 instead of even)
> -create table foo2 (a int, b char(200), c int);
> -insert into foo2 (select * from foo);
> -delete from foo2 where foo2.c = 1;
> -delete from foo2 where foo2.a >= 1000;
> -
> --- create "packed" index.
> -create index a on foo (a, b);
> -commit;
> -
> -create table spaceused (a int);
> -
> --- check space usage on the index
> -insert into spaceused
> -select numallocatedpages
> -from new org.apache.derby.diag.SpaceTable('FOO') t
> -where conglomeratename = 'A';
> -
> --- delete every other row
> -delete from foo where foo.c = 0 and foo.a < 1000;
> --- turn all the deletes into "committed deletes"
> -commit;
> --- wait for post commit on heap deleted rows to finish, so that there is
> --- no interference between post commit heap processing, and btree reclamation
> --- during split.
> -CALL WAIT_FOR_POST_COMMIT();
> -
> --- reinsert
> -insert into foo (select * from foo2);
> -commit;
> --- do the above several times
> -delete from foo where foo.c = 0 and foo.a < 1000;
> -commit;
> --- wait for post commit on heap deleted rows to finish, so that there is
> --- no interference between post commit heap processing, and btree reclamation
> --- during split.
> -CALL WAIT_FOR_POST_COMMIT();
> -
> -insert into foo (select * from foo2);
> -delete from foo where foo.c = 0 and foo.a < 1000;
> -commit;
> --- wait for post commit on heap deleted rows to finish, so that there is
> --- no interference between post commit heap processing, and btree reclamation
> --- during split.
> -CALL WAIT_FOR_POST_COMMIT();
> -
> -insert into foo (select * from foo2);
> -delete from foo where foo.c = 0 and foo.a < 1000;
> -commit;
> --- wait for post commit on heap deleted rows to finish, so that there is
> --- no interference between post commit heap processing, and btree reclamation
> --- during split.
> -CALL WAIT_FOR_POST_COMMIT();
> -
> -insert into foo (select * from foo2);
> -delete from foo where foo.c = 0 and foo.a < 1000;
> -commit;
> --- wait for post commit on heap deleted rows to finish, so that there is
> --- no interference between post commit heap processing, and btree reclamation
> --- during split.
> -CALL WAIT_FOR_POST_COMMIT();
> -
> -insert into foo (select * from foo2);
> -delete from foo where foo.c = 0 and foo.a < 1000;
> -commit;
> --- wait for post commit on heap deleted rows to finish, so that there is
> --- no interference between post commit heap processing, and btree reclamation
> --- during split.
> -CALL WAIT_FOR_POST_COMMIT();
> -
> -
> -insert into foo (select * from foo2);
> -
> -commit;
> -
> --- check space usage on the index
> -insert into spaceused
> -select numallocatedpages
> -from new org.apache.derby.diag.SpaceTable('FOO') t
> -where conglomeratename = 'A';
> -
> --- should only be 1 distinct row in this table since space used by the
> --- index should be the same before and after
> -select count(distinct a)
> -from spaceused;
> -
> --- cleanup
> -drop table foo;
> -drop table foo2;
> -drop table spaceused;
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
