Basic Logging
Introduction
This page describes the basic logging function provided by Alluxio server processes (e.g., masters, workers and etc) and application processes utilizing Alluxio clients (e.g., Spark or MapReduce jobs running on Alluxio). Note that there is an experimental feature that streams server-side and client-side logs to separate Alluxio logservers (see remote logging for more details).
Alluxio logging is implemented using log4j
and thus most of the configuration is done through modifying log4j.properties
.
Log Location
Server Logs
Log files for each individual Alluxio server process (e.g., master, worker, FUSE, proxy) can be found under ${ALLUXIO_HOME}/logs/
. Files suffixed with .log
like master.log
or worker.log
are generated by log4j
, logging the events that Alluxio system is recording through JVM. These files are typically the main target for users to investigate logs. Files suffixed with .out
like master.out
or worker.out
are the redirection of stdout
and stderr
of the corresponding process. Fatal error messages (e.g., killed by the OS), will most likely go to these files.
The log location can be customized by setting environment variable ALLUXIO_LOGS_DIR
. See the configuration settings page for more information.
By default, the *.log
files rotate. For example this is the default log4j configuration for master.log:
However, the *.out
files do not rotate. So it makes sense to regularly check the size of these files, and clean them up if necessary.
Application Logs
Log files for the Alluxio client utilized by different applications are located with their respective application logs. Please check out particular compute frameworks on where their logs may be found.
Here are the documentation to configure individual application logs including Apache Hadoop, Apache Hive, Apache Spark.
For the Alluxio command line interface, the user log is located at ${ALLUXIO_HOME}/logs/user/user_${user_name}.log
.
Configuring Log Levels
Alluxio uses the following five logging levels:
TRACE
: verbose information, only useful for debugging a certain method or classDEBUG
: fine-grained information, most useful for debugging purposesINFO
: messages that highlight the status or progressWARN
: potentially harmful events that users may need to know about, but the process will still continue runningERROR
: system errors that users should pay attention to
By default, Alluxio server processes write logs at INFO
level, which includes all events at INFO
, WARN
and ERROR
levels.
Modifying Logging with log4j.properties
log4j.properties
You can modify ${ALLUXIO_HOME}/conf/log4j.properties
to customize logging levels and restart corresponding server processes to apply new changes. This is the recommended way to modify logging configurations.
For example, to modify the level for all logs to DEBUG
, change the rootLogger
level by modifying the first line of log4j.properties
as the following:
To modify the logging level for a particular Java class (e.g., set alluxio.client.file.AlluxioFileInStream
to DEBUG
), add a new line at the end of this file:
To modify the logging level for a package (e.g., set all classes under alluxio
to DEBUG
), add a new line in the end of this file as below. This can be helpful when you do not know what the target classes are, or many classes are relevant.
Modifying Server Logging at Runtime
An alternative way to modify logging configurations is use the logLevel
command. This allows someone to modify the configuration at runtime without needing to restart processes. This is not the recommended way as any modifications will not be persisted across restart, and causes a configuration mismatch between the running process and its log4j.properties
file. See the logLevel command documentation for the command options.
For example, the following command sets the logger level of the class alluxio.underfs.hdfs.HdfsUnderFileSystem
to DEBUG
on master as well as a worker at 192.168.100.100:30000
:
And the following command returns the log level of the class alluxio.underfs.hdfs.HdfsUnderFileSystem
among all the workers:
You can also update the log level at a package level. For example, you can update the log level of all classes in alluxio.underfs
package with the following command:
This works because log4j loggers will inherit the log level from their ancestors. In this case alluxio.underfs.hdfs.HdfsUnderFileSystem
inherits the log level if it is set on alluxio.underfs
or alluxio.underfs.hdfs
.
Furthermore, you can turn on Alluxio debug logging when you are troubleshooting a certain issue in a running cluster, and turn it off when you are done.
Finally, if your Alluxio deployment uses custom web ports (e.g. alluxio.master.web.port
is different from 19999, or alluxio.worker.web.port
is different from 30000), you can use the format host:port:role
for your target. role
can be one of master
or worker
or job_master
or job_worker
. For example, if your master running on 10.10.10.10
has alluxio.master.web.port=2181
configured, you would use:
If your worker is running on 127.0.0.1
with alluxio.worker.web.port=25252
configured, you would use:
Enabling Advanced Logging
Logging JVM GC (Garbage Collection) events
Add the following line to conf/allulxio-env.sh
to enable logging GC events for server processes in log files with .out
suffix like master.out
and worker.out
:
ALLUXIO_JAVA_OPTS
is included in Java VM options for all Alluxio server processes. Alternatively, modify ALLUXIO_MASTER_JAVA_OPTS
, ALLUXIO_WORKER_JAVA_OPTS
to turn on GC for each individual process.
Logging all FUSE API calls
Set in conf/log4j.properties
:
You will see debug logs at both the start and end of each FUSE API call with its arguments and result in logs/fuse.log
:
Logging RPCs Calls Sent by Client
Add the following to your application-side log4j.properties
to capture RPCs between the Alluxio client and FileSystem Master:
Similarly, capture lower-level RPCs between Alluxio client and Block Master:
You will see debug logs at the beginning and end of each RPC with its arguments and result in the client logs like the following:
Logging RPC Calls Received by Masters
On the master, debug-level logging for incoming File System level RPC calls can be turned on (e.g., creating/reading/writing/removing files, updating file attributions) using the logLevel
command:
Similarly, turn on the debug-level logging for block related RPC calls (e.g., adding/removing blocks):
Similarly, you can update the log level for these classes in the conf/log4j.properties
file. You need to restart the relevant processes for the log4j properties to take effect.
Logging UnderFileSystem Operations
Sometimes it can be useful to log all operations on under storage. On the master, debug-level logging for UnderFileSystem operations can be turned on (e.g., creating/reading/writing/removing files on UFS) using the logLevel
command:
One can similarly turn on UFS operations on workers by passing --target workers
. You can see operations in corresponding master or worker logs like below:
Similarly, you can update the log level for these classes in the conf/log4j.properties
file. You need to restart the relevant processes for the log4j properties to take effect.
Identifying Expensive Calls
When debugging the performance, it is often useful to understand which RPCs take most of the time but without recording all the communication (e.g., enabling all debug logging). Alluxio can record slow calls or RPCs in logs with WARN level by setting the following properties:
Set
alluxio.user.logging.threshold
to record slow client-side RPCs in application logs.Set
alluxio.fuse.logging.threshold
to record slow FUSE API calls in fuse logs (logs/fuse.log
).Set
alluxio.underfs.logging.threshold
to record slow under storage RPCs (logs/master.log
orlogs/worker.log
).
For example, the following setting in conf/alluxio-site.properties
will log each FUSE API call slower than 1s, each UFS call slower than 10s, and each client-side RPC slower than 200ms:
Example results are:
Redirecting debug log for certain classes
Sometimes it is useful to separate the log for certain classes into a separate log. This can be useful for reasons including but not limited to:
Clearly separate the wanted logs for further analysis.
Avoid
master.log
orworker.log
being too big or having too many files created by log rotation.Use a separate logger to send logs to a remote endpoint like a socket.
This can be achieved by adding a separate logger in the conf/log4j.properties
. For example, the below example redirects debug logs of alluxio.master.StateLockManager
to a separate set of files, so the master.log
will not be full of DEBUG logs created by alluxio.master.StateLockManager
.
Disable certain log files
Sometimes it makes sense to disable certain logs files.
One example of use cases is when Alluxio is running in a containerized environment, where logs are written to the writable layer. This has performance penalties and the writable layer may grow indefinitely, causing disk pressure on the host. In that case you can either mount a volume to the log directory so logs are written to the volume, or rely on the Remote Logging and disable local logs.
To disable local log output, you can set the corresponding logger(s) as below:
Last updated