[
https://issues.apache.org/jira/browse/HIVE-12612?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Paulo Sequeira updated HIVE-12612:
----------------------------------
Description:
Similar to what was reported on HIVE-6978, but now it only happens when the
query is read from the standard input. For example, the following fails as
expected:
{code}
bash$ if beeline -u "jdbc:hive2://..." -e "boo;" ; then echo "Ok?!" ; else echo
"Failed!" ; fi
Connecting to jdbc:hive2://...
Connected to: Apache Hive (version 1.1.0-cdh5.5.0)
Driver: Hive JDBC (version 1.1.0-cdh5.5.0)
Transaction isolation: TRANSACTION_REPEATABLE_READ
Error: Error while compiling statement: FAILED: ParseException line 1:0 cannot
recognize input near 'boo' '<EOF>' '<EOF>' (state=42000,code=40000)
Closing: 0: jdbc:hive2://...
Failed!
{code}
But the following does not:
{code}
bash$ if echo "boo;"|beeline -u "jdbc:hive2://..." ; then echo "Ok?!" ; else
echo "Failed!" ; fi
Connecting to jdbc:hive2://...
Connected to: Apache Hive (version 1.1.0-cdh5.5.0)
Driver: Hive JDBC (version 1.1.0-cdh5.5.0)
Transaction isolation: TRANSACTION_REPEATABLE_READ
Beeline version 1.1.0-cdh5.5.0 by Apache Hive
0: jdbc:hive2://...:8> Error: Error while compiling statement: FAILED:
ParseException line 1:0 cannot recognize input near 'boo' '<EOF>' '<EOF>'
(state=42000,code=40000)
0: jdbc:hive2://...:8> Closing: 0: jdbc:hive2://...
Ok?!
{code}
This was misleading our batch scripts to always believe that the execution of
the queries succeded, when sometimes that was not the case.
h2. Workaround
We found we can work around the issue by always using the -e or the -f
parameters, and even reading the standard input through the /dev/stdin device
(this was useful because a lot of the scripts fed the queries from here
documents), like this:
{code:title=some-script.sh}
#!/bin/sh
set -o nounset -o errexit -o pipefail
# As beeline is failing to report an error status if reading the query
# to be executed from STDIN, check whether no -f or -e option is used
# and, in that case, pretend it has to read the query from a regular
# file using -f to read from /dev/stdin
function beeline_workaround_exit_status () {
for arg in "$@"
do if [ "$arg" = "-f" -o "$arg" = "-e" ]
then beeline -u "..." "$@"
return
fi
done
beeline -u "..." "$@" -f /dev/stdin
}
beeline_workaround_exit_status <<EOF
boo;
EOF
{code}
was:
Similar to what was reported on HIVE-6978, but now it only happens when the
query is read from the standard input. For example, the following fails as
expected:
{code}
bash$ if beeline -u "jdbc:hive2://..." -e "boo;" ; then echo "Ok?!" ; else echo
"Failed!" ; fi
Connecting to jdbc:hive2://...
Connected to: Apache Hive (version 1.1.0-cdh5.5.0)
Driver: Hive JDBC (version 1.1.0-cdh5.5.0)
Transaction isolation: TRANSACTION_REPEATABLE_READ
Error: Error while compiling statement: FAILED: ParseException line 1:0 cannot
recognize input near 'boo' '<EOF>' '<EOF>' (state=42000,code=40000)
Closing: 0: jdbc:hive2://...
Failed!
{code}
But the following does not:
{code}
bash$ if echo "boo;"|beeline -u "jdbc:hive2://..." ; then echo "Ok?!" ; else
echo "Failed!" ; fi
Connecting to jdbc:hive2://...
Connected to: Apache Hive (version 1.1.0-cdh5.5.0)
Driver: Hive JDBC (version 1.1.0-cdh5.5.0)
Transaction isolation: TRANSACTION_REPEATABLE_READ
Beeline version 1.1.0-cdh5.5.0 by Apache Hive
0: jdbc:hive2://...:8> Error: Error while compiling statement: FAILED:
ParseException line 1:0 cannot recognize input near 'boo' '<EOF>' '<EOF>'
(state=42000,code=40000)
0: jdbc:hive2://...:8> Closing: 0: jdbc:hive2://...
Ok?!
{code}
This was misleading our batch scripts to always believe that the execution of
the queries always succeded, when sometimes that was not the case.
h2. Workaround
We found we can work around the issue by always using the -e or the -f
parameters, and even reading the standard input through the /dev/stdin device
(this was useful because a lot of the scripts fed the queries from here
documents), like this:
{code:title=some-script.sh}
#!/bin/sh
set -o nounset -o errexit -o pipefail
# As beeline is failing to report an error status if reading the query
# to be executed from STDIN, check whether no -f or -e option is used
# and, in that case, pretend it has to read the query from a regular
# file using -f to read from /dev/stdin
function beeline_workaround_exit_status () {
for arg in "$@"
do if [ "$arg" = "-f" -o "$arg" = "-e" ]
then beeline -u "..." "$@"
return
fi
done
beeline -u "..." "$@" -f /dev/stdin
}
beeline_workaround_exit_status <<EOF
boo;
EOF
{code}
> beeline always exits with 0 status when reading query from standard input
> -------------------------------------------------------------------------
>
> Key: HIVE-12612
> URL: https://issues.apache.org/jira/browse/HIVE-12612
> Project: Hive
> Issue Type: Bug
> Components: Beeline
> Affects Versions: 1.1.0
> Environment: CDH5.5.0
> Reporter: Paulo Sequeira
> Priority: Minor
>
> Similar to what was reported on HIVE-6978, but now it only happens when the
> query is read from the standard input. For example, the following fails as
> expected:
> {code}
> bash$ if beeline -u "jdbc:hive2://..." -e "boo;" ; then echo "Ok?!" ; else
> echo "Failed!" ; fi
> Connecting to jdbc:hive2://...
> Connected to: Apache Hive (version 1.1.0-cdh5.5.0)
> Driver: Hive JDBC (version 1.1.0-cdh5.5.0)
> Transaction isolation: TRANSACTION_REPEATABLE_READ
> Error: Error while compiling statement: FAILED: ParseException line 1:0
> cannot recognize input near 'boo' '<EOF>' '<EOF>' (state=42000,code=40000)
> Closing: 0: jdbc:hive2://...
> Failed!
> {code}
> But the following does not:
> {code}
> bash$ if echo "boo;"|beeline -u "jdbc:hive2://..." ; then echo "Ok?!" ; else
> echo "Failed!" ; fi
> Connecting to jdbc:hive2://...
> Connected to: Apache Hive (version 1.1.0-cdh5.5.0)
> Driver: Hive JDBC (version 1.1.0-cdh5.5.0)
> Transaction isolation: TRANSACTION_REPEATABLE_READ
> Beeline version 1.1.0-cdh5.5.0 by Apache Hive
> 0: jdbc:hive2://...:8> Error: Error while compiling statement: FAILED:
> ParseException line 1:0 cannot recognize input near 'boo' '<EOF>' '<EOF>'
> (state=42000,code=40000)
> 0: jdbc:hive2://...:8> Closing: 0: jdbc:hive2://...
> Ok?!
> {code}
> This was misleading our batch scripts to always believe that the execution of
> the queries succeded, when sometimes that was not the case.
> h2. Workaround
> We found we can work around the issue by always using the -e or the -f
> parameters, and even reading the standard input through the /dev/stdin device
> (this was useful because a lot of the scripts fed the queries from here
> documents), like this:
> {code:title=some-script.sh}
> #!/bin/sh
> set -o nounset -o errexit -o pipefail
> # As beeline is failing to report an error status if reading the query
> # to be executed from STDIN, check whether no -f or -e option is used
> # and, in that case, pretend it has to read the query from a regular
> # file using -f to read from /dev/stdin
> function beeline_workaround_exit_status () {
> for arg in "$@"
> do if [ "$arg" = "-f" -o "$arg" = "-e" ]
> then beeline -u "..." "$@"
> return
> fi
> done
> beeline -u "..." "$@" -f /dev/stdin
> }
> beeline_workaround_exit_status <<EOF
> boo;
> EOF
> {code}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
