[
https://issues.apache.org/jira/browse/TRAFODION-1449?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14714425#comment-14714425
]
Qifan Chen commented on TRAFODION-1449:
---------------------------------------
The goal of the fix is to remove redundancy that was reported in previous
posing.
1. The need of full key predicates that refer to any key column.
This set is required to compute the keys for HBase in
SearchKey::makeHBaseSearchKey(). The normal key predicates (stored in
SearchKey::keyPredicates_) is not suitable since it is an intersection of
commonPredicates and may not contain all the predicates. This is the reason
that the select predicates were included as the input to makeHbaseSearchKey()
call. However, the inclusion of the selection predicates introduces
redundancy.
The full key predicates are computed in SearchKey::init() and stored in
SearchKey as a new data member fullyKeyPredicates_. The full key predicate is
a set of predicates in which each predicate refers to some key column,
2. Removal of the redundancy elsewhere.
The two places in the code that introduce redundant representation of the same
original predicates have been modified. One is the scan optimizer before we
call "new SearchKey" to include another representation of a RANGE SPEC. The
second place is the setup steps before calling SearchKey::makeHBaseSaerchKey().
The final executor predicate used for HBase access or IUD operators will be the
predicates survived from makeHbaseSearchKey() call, plus the executor predicate
passed-in plus the selection predicate minus the full key predicates.
Evidences that the fix works.
1. The original query.
TRAFODION_SCAN ============================ SEQ_NO 1 NO CHILDREN
TABLE_NAME ............... FOO
REQUESTS_IN .............. 1
ROWS_OUT ................ 11
EST_OPER_COST ............ 0.01
EST_TOTAL_COST ........... 0.01
DESCRIPTION
max_card_est .......... 11
fragment_id ............ 0
parent_frag ............ (none)
fragment_type .......... master
scan_type .............. subset scan of table TRAFODION.SEABASE.FOO
object_type ............ Trafodion
columns ................ all
begin_keys(excl) ....... 2
end_keys(excl) ......... 5 <== the end key is present with the fix now.
cache_size ........... 100
probes ................. 1
rows_accessed ......... 11
key_columns ............ UNIQ
executor_predicates .... (UNIQ > 2) and (UNIQ < 5)
2. Query in regression SEABASE/TEST010
TRAFODION_SCAN ============================ SEQ_NO 1 NO CHILDREN
TABLE_NAME ............... T010T2
REQUESTS_IN .............. 1
ROWS_OUT ................. 3
EST_OPER_COST ............ 0.01
EST_TOTAL_COST ........... 0.01
DESCRIPTION
max_card_est ........... 5
fragment_id ............ 0
parent_frag ............ (none)
fragment_type .......... master
scan_type .............. subset scan of table TRAFODION.SEABASE.T010T2
object_type ............ Trafodion
columns ................ all
unique_rows ............ 1,a,1 <== we do not miss these two keys either.
unique_rows ............ 4,a,1
cache_size ........... 100
probes ................. 1
rows_accessed ......... 10
key_columns ............ A, B, C
executor_predicates .... (B = 'a') and (C = 1) and ((A = 1) or (A = 4))
> End-key is not specified for a scan node when range spec feature is on.
> ------------------------------------------------------------------------
>
> Key: TRAFODION-1449
> URL: https://issues.apache.org/jira/browse/TRAFODION-1449
> Project: Apache Trafodion
> Issue Type: Bug
> Components: sql-cmp
> Affects Versions: 1.1 (pre-incubation)
> Reporter: Qifan Chen
> Assignee: Qifan Chen
> Labels: performance
>
> Problem: the end key is not specified for the scan operator and is evaluated
> in the executor predicate.
> MODULE_NAME .............. DYNAMICALLY COMPILED
> STATEMENT_NAME ........... XX
> PLAN_ID .................. 212306167959521210
> ROWS_OUT ................. 1
> EST_TOTAL_COST ........... 0.01
> STATEMENT ................ select *
> from foo << + cardinality 10e8 >>
> where uniq > 2 and uniq < 5;
> TRAFODION_SCAN ============================ SEQ_NO 1 NO CHILDREN
> TABLE_NAME ............... FOO
> REQUESTS_IN .............. 1
> ROWS_OUT ................ 11
> EST_OPER_COST ............ 0.01
> EST_TOTAL_COST ........... 0.01
> DESCRIPTION
> max_card_est .......... 11
> fragment_id ............ 0
> parent_frag ............ (none)
> fragment_type .......... master
> scan_type .............. subset scan of table TRAFODION.SEABASE.FOO
> object_type ............ Trafodion
> columns ................ all
> begin_keys(excl) ....... 2
> end_keys(incl)
> cache_size ........... 100
> probes ................. 1
> rows_accessed ......... 11
> key_columns ............ UNIQ
> executor_predicates .... (UNIQ < 5) and (UNIQ > 2) and (UNIQ < 5)
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
