[
https://issues.apache.org/jira/browse/TRAFODION-27?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14624918#comment-14624918
]
ASF GitHub Bot commented on TRAFODION-27:
-----------------------------------------
Github user zellerh commented on a diff in the pull request:
https://github.com/apache/incubator-trafodion/pull/25#discussion_r34482687
--- Diff: core/sql/generator/GenProbeCache.cpp ---
@@ -361,6 +361,38 @@ short ProbeCache::codeGen(Generator *generator)
// that there is room enough for one row.
}
+ double memoryLimitPerCpu =
+
ActiveSchemaDB()->getDefaults().getAsLong(EXE_MEMORY_FOR_PROBE_CACHE_IN_MB) *
1024 * 1024;
+ double estimatedMemory;
+
+ if (numInnerTuples_ > 0) {
+ estimatedMemory = numInnerTuples_ * innerRecLength;
+ if (estimatedMemory > memoryLimitPerCpu) {
+ numInnerTuples_ = memoryLimitPerCpu / innerRecLength;
+ queue_index pUpSize_calc;
+ queue_index pDownSize_calc;
+
+ pUpSize_calc = numInnerTuples_ * pUpSize / (pUpSize + pDownSize);
+ pDownSize_calc = numInnerTuples_ * pDownSize / (pUpSize +
pDownSize);
+ queue_index pq2 = 1;
+ queue_index bits = pUpSize_calc;
+ while (bits && pq2 < pUpSize_calc) {
+ bits = bits >> 1;
+ pq2 = pq2 << 1;
+ }
+ pUpSize = pq2;
--- End diff --
The code on lines 350-356 makes sure that numInnerTuples_ is greater or
equal pUpSize. This while loop seems to me to be doing the opposite, make sure
that pUpSize (and pDownSize) below are greater or equal numInnerTuples_. Should
this say: pUpSize = pq2 / 2 instead?
> Time consuming compilation for queries of BLOB with JOINs
> ----------------------------------------------------------
>
> Key: TRAFODION-27
> URL: https://issues.apache.org/jira/browse/TRAFODION-27
> Project: Apache Trafodion
> Issue Type: Bug
> Environment: Trafodion 1.2, HDP 2.2, CentOS 6
> Reporter: Haifeng Li
> Priority: Minor
> Attachments: trafodion27_qry.sql, trafodion27_setup.sql
>
>
> It takes 3 ~ 5 seconds to compile a query that returns a BLOB field by
> joining two tables. If a query is cached and subsequent queries are 'similar'
> to
> the initial query, those subsequence compilation times will usually be
> much faster. However, sometimes we observe that subsequent queries with
> literal difference still take long time in sqlci.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
