Define the routing of the log output
When using started tasks to start and stop Spark processes, the routing of the log output is dependent upon a combination of settings in the $SPARK_CONF_DIR/log4j.properties file and the values for the STDERR or STDOUT DD cards in the master and worker procedures.
log4j.rootCategory=INFO, console
log4j.appender.console=org.apache.log4j.ConsoleAppender
log4j.appender.console.target=System.err
Using the default values for both the $SPARK_CONF_DIR/log4j.properties file
and the STDERR DD card of //STDERR DD SYSOUT=&SOUT
directs both the JZOS output
and the Spark output to
SYSOUT
.
Changing the log4j.appender.console.target
value to System.out
will route the log output to the value in the STDOUT DD card.
Alternatively, to route the Spark log output to a USS file in
the $SPARK_LOG_DIR directory, if you are using the default values for the
$SPARK_CONF_DIR/log4j.properties file, you must specify the full path for the
log file by using the value that is set for $SPARK_LOG_DIR on the STDERR DD
card. The following examples show the specifications for the master and the worker log files for a
cluster that is denoted as Cluster1
where the value of
$SPARK_LOG_DIR is /var/spark/logs
. You must specify a unique
file name for the master and a unique file name for the worker. If you are starting more than one
cluster, you must specify unique file names for each master and worker across all clusters.
//STDERR DD PATH='/var/spark/logs/azkmstrCluster1.err',
// PATHOPTS=(OWRONLY,OCREAT,OTRUNC),PATHMODE=SIRWXU
//STDERR DD PATH='/var/spark/logs/azkwrkrCluster1.err',
// PATHOPTS=(OWRONLY,OCREAT,OTRUNC),PATHMODE=SIRWXU
By default, the master and worker log files are created with
PATHMODE=SIRWXU
, which is equivalent to z/OS UNIX file access attributes of “700”
(“rwx------
“). As such, the user ID that starts/stops the Spark cluster (SPARKID in
these examples) has full access to the log files, and no access for anyone else. You may change this
behavior by modifying the PATHMODE parameter value. For more information about the PATHMODE
parameter, see "PATHMODE parameter" in z/OS MVS JCL Reference.
When using started tasks to start and stop Spark processes, the log files are
not rotated as when using other methods for starting the master and worker. It is recommended that
you not specify OAPPEND
for the PATHOPTS
to avoid file growth.