List:Commits« Previous MessageNext Message »
From:jonas oreland Date:May 4 2011 1:00pm
Subject:bzr push into mysql-5.1-telco-7.0 branch (jonas:4355 to 4356)
View as plain text  
 4356 jonas oreland	2011-05-04
      ndb - add --no-defaults to 2 --exec's which would otherwise fail mysteriously if having non-default environment (found by having non standard environment by misstake)

    modified:
      mysql-test/include/ndb_backup_id.inc
      mysql-test/suite/ndb/t/ndb_show_tables_result.inc
 4355 Ole John Aske	2011-05-04
      Extended SPJ block with the result mode 'NI_REPEAT_SCAN_RESULT' which will
      enable us to also push bushy-scan operation (or 'star-join') when integrated
      with mysqld.
      
      This mode comes in addition as an alternative to the already existing way
      of handling pushed queries with multiple child scans.
      
      - Currently new result rows from all child scans related to the same parent 
        operation is returned simultaneously when fetching more scan rows.
        In order for mysqld to produce joined results from these rows, it has
        to buffer all rows from the other scans currently not being iterated 
        in the nested loop join algorith. As #rows returned might be (almost) 
        infinite this was not possible to integrate with mysqld, and has currently
        forced us to enforce an artificial parent dependencies between bushy-scans.
        This had the side effect that cross joined results was materialized on
        the SPJ nodes, and lots of redundant result rows had to be returned.
      
      - The new 'NI_REPEAT_SCAN_RESULT' aims at reducing these redundant rows.
        It moves X-join granularity from 'repeat rows' (as above) to
       'repeat batches' such that: If multiple batches from bushy scans has 
        remaining rows, a sequence of NEXTREQ's will fill new rows into *one* of
        batch buffers, while the other scan batches will be filled with 
        rows which are repeated from the the previous fetches. 
        The combination of these batched rows will be unique, and returned
        such that if the client iterates all current rows in the batched it
        will form the full crossproduct of the start-joined scans.

    modified:
      storage/ndb/include/kernel/signaldata/QueryTree.hpp
      storage/ndb/src/kernel/blocks/dbspj/Dbspj.hpp
      storage/ndb/src/kernel/blocks/dbspj/DbspjMain.cpp
      storage/ndb/src/ndbapi/NdbQueryBuilder.cpp
      storage/ndb/src/ndbapi/NdbQueryBuilderImpl.hpp
      storage/ndb/src/ndbapi/ndberror.c
=== modified file 'mysql-test/include/ndb_backup_id.inc'
--- a/mysql-test/include/ndb_backup_id.inc	2009-07-14 07:52:35 +0000
+++ b/mysql-test/include/ndb_backup_id.inc	2011-05-04 13:00:44 +0000
@@ -1,6 +1,6 @@
 # there is no neat way to find the backupid, this is a hack to find it...
 --let $dump_file= $MYSQLTEST_VARDIR/tmp/select_all.txt
---exec $NDB_SELECT_ALL -d sys --delimiter=, SYSTAB_0 > $dump_file
+--exec $NDB_SELECT_ALL --no-defaults -d sys --delimiter=, SYSTAB_0 > $dump_file
 
 CREATE TEMPORARY TABLE test.backup_info(
   a BIGINT,

=== modified file 'mysql-test/suite/ndb/t/ndb_show_tables_result.inc'
--- a/mysql-test/suite/ndb/t/ndb_show_tables_result.inc	2009-10-09 16:13:54 +0000
+++ b/mysql-test/suite/ndb/t/ndb_show_tables_result.inc	2011-05-04 13:00:44 +0000
@@ -2,7 +2,7 @@
 let $dump_file = $MYSQLTEST_VARDIR/tmp/ndb_show_tables.txt;
 
 # Dump the output of ndb_show_tables to file
---exec $NDB_SHOW_TABLES --p > $dump_file
+--exec $NDB_SHOW_TABLES --no-defaults --p > $dump_file
 
 TRUNCATE ndb_show_tables_results;
 

No bundle (reason: useless for push emails).
Thread
bzr push into mysql-5.1-telco-7.0 branch (jonas:4355 to 4356) jonas oreland4 May