[ https://issues.apache.org/jira/browse/TRAFODION-3192?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16596661#comment-16596661 ]
Sandhya Sundaresan commented on TRAFODION-3192: ----------------------------------------------- |an old email showing other queue resizing: From: Eric Owhadi Sent: Monday, September 11, 2017 3:25 PM To: Sandhya Sundaresan <[sandhya.sundare...@esgyn.com|mailto:sandhya.sundare...@esgyn.com]>; Hans Zeller <[hans.zel...@esgyn.com|mailto:hans.zel...@esgyn.com]>; Suresh Subbiah <[suresh.subb...@esgyn.com|mailto:suresh.subb...@esgyn.com]>; Anoop Sharma <[anoop.sha...@esgyn.com|mailto:anoop.sha...@esgyn.com]> Subject: RE: update path may need some work? I have seen queue sizing issues on couple of places. cqd GEN_DP2I_SIZE_DOWN '1024'; cqd UPD_ORDERED 'OFF'; cqd HBASE_ROWSET_VSBB_SIZE '20480'; for an upsert using load-under performing clearly because auto sizing was not kickin-in (this is for my bottle of wine bet with Hans POC ). and queue of a UNION ALL operator in my HotCold also failing to resize, had to hardcode it using CQD cqd gen_un_size_up '2048'; cqd gen_un_size_down '2048'; So given the queueing mechanism is in a shared class inherited by many operator, it would not surprise me if a common issue affecting multiple operators? It would not surprise me that if people start looking at dispatch metrics, this would be a widespread problem, \| > UNION ALL up and down queue size default value too small, or not dynamically > adjusted > ------------------------------------------------------------------------------------- > > Key: TRAFODION-3192 > URL: https://issues.apache.org/jira/browse/TRAFODION-3192 > Project: Apache Trafodion > Issue Type: Bug > Components: sql-exe > Affects Versions: any > Reporter: Sandhya Sundaresan > Assignee: Sandhya Sundaresan > Priority: Major > Fix For: 2.4 > > > The union up queue size is hardcoded in nadefaults.cpp as a start value as 16 > which > is very low if large number of rows are being returned. Scan operator has it > set to 2048. > Some operators further adjust this up size from the original value based on > estimate of > the number of rows being returned. We can do that for union codegen method and > also change the original up queue size to 2048. > > Noticed that Transpose, Sequence also have the same issue and dynamic queue > resizing is missing. -- This message was sent by Atlassian JIRA (v7.6.3#76005)