[
https://issues.apache.org/jira/browse/TRAFODION-2301?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15590811#comment-15590811
]
Roberta Marton commented on TRAFODION-2301:
-------------------------------------------
Today the UDF event_log_reader scans all logs and loads into memory and then
discards the rows that is not need to return. It is a known design limitation.
Since there could be thousands of logs, it uses too much memory which cause
system issues.
> Hadoop stack crash with LOGS TMUDF
> ----------------------------------
>
> Key: TRAFODION-2301
> URL: https://issues.apache.org/jira/browse/TRAFODION-2301
> Project: Apache Trafodion
> Issue Type: Bug
> Components: sql-general
> Reporter: Roberta Marton
> Assignee: Roberta Marton
>
> There seems to be a problem is a that shows up when a tdm_udrsrvr process
> takes most of the available Memory. When this happens, too much CPU time is
> being used to handle this memory issue which causes some of Hadoop processes
> to crash due to lack of resources. In addition, Linux is slow to respond
> from the terminal.
>
> It can be reproduced from DB Mgr using log page which calls UDR in the
> backend. Calling this UDF (event_log_reader) from sqlci can also reproduce
> the problem.
>
> When using sqlci, SQL prints out the udrserver process launched info but then
> just hangs.
> >>select [first 1]* from udf(event_log_reader());
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
