Possible causes of losing log messages
During the course of a message from the sending application to the final destination of the message, there are a number of locations where a message may be lost, even though syslog-ng does its best to avoid message loss. Usually losing messages can be avoided with careful planning and proper configuration of syslog-ng and the hosts running syslog-ng. The following list shows the possible locations where messages may be lost, and provides methods to minimize the risk of losing messages.
-
Between the application and the syslog-ng client: Make sure to use an appropriate source to receive the logs from the application (for example from /dev/log). For example, use unix-stream instead of unix-dgram whenever possible.
-
When syslog-ng is sending messages: If syslog-ng cannot send messages to the destination and the output buffer gets full, syslog-ng will drop messages.
Use flags (flow-control) to avoid it (for details, see Configuring flow-control). For more information about the error caused by the missing flow-control, see Destination queue full.
The number of dropped messages is displayed per destination in the log message statistics of syslog-ng (for details, see Statistics of syslog-ng).
-
On the network: When transferring messages using the UDP protocol, messages may be lost without any notice or feedback — such is the nature of the UDP protocol. Always use the TCP protocol to transfer messages over the network whenever possible.
-
In the socket receive buffer: When transferring messages using the UDP protocol, the UDP datagram (that is, the message) that reaches the receiving host placed in a memory area called the socket receive buffer. If the host receives more messages than it can process, this area overflows, and the kernel drops messages without letting syslog-ng know about it. Using TCP instead of UDP prevents this issue. If you must use the UDP protocol, increase the size of the receive buffer using the so-rcvbuf() option.
-
When syslog-ng is receiving messages:
-
The receiving syslog-ng (for example the syslog-ng server or relay) may drop messages if the fifo of the destination file gets full. The number of dropped messages is displayed per destination in the log message statistics of syslog-ng (for details, see Statistics of syslog-ng).
-
-
When the destination cannot handle large load: When syslog-ng is sending messages at a high rate into an SQL database, a file, or another destination, it is possible that the destination cannot handle the load, and processes the messages slowly. As a result, the buffers of syslog-ng fill up, syslog-ng cannot process the incoming messages, and starts to loose messages. For details, see the previous entry. Use the throttle parameter to avoid this problem.
-
As a result of an unclean shutdown of the syslog-ng server: If the host running the syslog-ng server experiences an unclean shutdown, it takes time until the clients realize that the connection to the syslog-ng server is down. Messages that are put into the output TCP buffer of the clients during this period are not sent to the server.
-
When syslog-ng OSE is writing messages into files: If syslog-ng OSE receives a signal (SIG) while writing log messages to file, the log message that is processed by the write call can be lost if the flush_lines parameter is higher than 1.
Creating syslog-ng core files
Purpose:
When syslog-ng crashes for some reason, it can create a core file that contains important troubleshooting information. To enable core files, complete the following procedure:
Steps:
-
Core files are produced only if the maximum core file size ulimit is set to a high value in the init script of syslog-ng.Add the following line to the init script of syslog-ng:
ulimit -c unlimited
-
Verify that syslog-ng has permissions to write the directory it is started from, for example /opt/syslog-ng/sbin/.
-
If syslog-ng crashes, it will create a core file in the directory syslog-ng was started from.
-
To test that syslog-ng can create a core file, you can create a crash manually. For this, determine the PID of syslog-ng (for example using the ps -All|grep syslog-ng command), then issue the following command: kill -ABRT <syslog-ng pid>
This should create a core file in the current working directory.
Collecting debugging information with strace, truss, or tusc
To properly troubleshoot certain situations, it can be useful to trace which system calls syslog-ng OSE performs. How this is performed depends on the platform running syslog-ng OSE. In general, note the following points:
-
When syslog-ng OSE is started, a supervisor process might stay in the foreground, while the actual syslog-ng daemon goes to the background. Always trace the background process.
-
Apart from the system calls, the time between two system calls can be important as well. Make sure that your tracing tool records the time information as well. For details on how to do that, refer to the manual page of your specific tool (for example, strace on Linux, or truss on Solaris and BSD).
-
Run your tracing tool in verbose mode, and if possible, set it to print long output strings, so the messages are not truncated.
-
When using strace, also record the output of lsof to see which files are accessed.
The following are examples for tracing system calls of syslog-ng on some platforms. The output is saved into the /tmp/syslog-ng-trace.txt file, sufficed with the PID of the related syslog-ng process.The path of the syslog-ng binary may be different for your installation, as distribution-specific packages may use different paths.
-
Linux: strace -o /tmp/trace.txt -s256 -ff -ttT /opt/syslog-ng/sbin/syslog-ng -f /opt/syslog-ng/etc/syslog-ng.conf -Fdv
-
HP-UX: tusc -f -o /tmp/syslog-ng-trace.txt -T /opt/syslog-ng/sbin/syslog-ng
-
IBM AIX and Solaris: truss -f -o /tmp/syslog-ng-trace.txt -r all -w all -u libc:: /opt/syslog-ng/sbin/syslog-ng -d -d -d
|
|
TIP:
To execute these commands on an already running syslog-ng OSE process, use the -p <pid_of_syslog-ng> parameter. |
Running a failure script
Purpose:
You can create a failure script that is executed when syslog-ng OSE terminates abnormally, that is, when it exits with a non-zero exit code. For example, you can use this script to send an automatic e-mail notification.
Prerequisites:
The failure script must be the following file: /opt/syslog-ng/sbin/syslog-ng-failure, and must be executable.
To create a sample failure script, complete the following steps.
Steps:
-
Create a file named /opt/syslog-ng/sbin/syslog-ng-failure with the following content:
#!/usr/bin/env bash cat >>/tmp/test.txt <<EOF $(date) Name............$1 Chroot dir......$2 Pid file dir....$3 Pid file........$4 Cwd.............$5 Caps............$6 Reason..........$7 Argbuf..........$8 Restarting......$9 EOF
-
Make the file executable: chmod +x /opt/syslog-ng/sbin/syslog-ng-failure
-
Run the following command in the /opt/syslog-ng/sbin directory: ./syslog-ng --process-mode=safe-background; sleep 0.5; ps aux | grep './syslog-ng' | grep -v grep | awk '{print $2}' | xargs kill -KILL; sleep 0.5; cat /tmp/test.txt
The command starts syslog-ng OSE in safe-background mode (which is needed to use the failure script) and then kills it. You should see that the relevant information is written into the /tmp/test.txt file, for example:
Thu May 18 12:08:58 UTC 2017 Name............syslog-ng Chroot dir......NULL Pid file dir....NULL Pid file........NULL Cwd.............NULL Caps............NULL Reason..........signalled Argbuf..........9 Restarting......not-restarting
-
You should also see messages similar to the following in system syslog. The exact message depends on the signal (or the reason why syslog-ng OSE stopped):
May 18 13:56:09 myhost supervise/syslog-ng[10820]: Daemon exited gracefully, not restarting; exitcode='0' May 18 13:57:01 myhost supervise/syslog-ng[10996]: Daemon exited due to a deadlock/signal/failure, restarting; exitcode='131' May 18 13:57:37 myhost supervise/syslog-ng[11480]: Daemon was killed, not restarting; exitcode='9'
The failure script should run on every non-zero exit event.