48

I was trying to run flume on Ubuntu 16.04 as systemd service and have following in /etc/systemd/system/flume-ng.service

[Unit] Description=Apache Flume [Service] ExecStart=/usr/bin/nohup /opt/flume/current/bin/flume-ng agent -c /etc/flume-ng/conf -f /etc/flume-ng/conf/flume.conf --name a1 & ExecStop=/opt/flume/current/bin/flume-ng agent stop [Install] WantedBy=multi-user.target

I tried adding following lines

StandardOutput=/var/log/flume-ng/log1.log StandardError=/var/log/flume-ng/log2.log

which didn't work for me. I did run systemctl daemon-reload and systemctl restart flume-ng

anyone know how this works ?

Improve this question

6 Answers 6

Reset to default
72

Redirect

The following works only in systemd v236 and newer:

StandardOutput=file:/var/log/flume-ng/log1.log StandardError=file:/var/log/flume-ng/log2.log

as documented here.

In case of systemd older than v236, you can use:

ExecStart=/bin/sh -c 'exec /usr/bin/my_binary [arguments] >/var/log/flume-ng/log1.log 2>/var/log/flume-ng/log2.log'

Note that this way the whole log files contents will be overwritten each time service restarts.

Append

If you want to maintain file log between service restarts and just append new logged lines to it:

# Works only in systemd v240 and newer! # (but backported to e.g. RHEL8) StandardOutput=append:/var/log/flume-ng/log1.log StandardError=append:/var/log/flume-ng/log2.log

In case of systemd older than v240, you can use:

ExecStart=/bin/sh -c 'exec /usr/bin/my_binary [arguments] >>/var/log/flume-ng/log1.log 2>>/var/log/flume-ng/log2.log'

exec means that shell program will be substituted with my_binary program after setting up redirections without forking. So there will be no difference from running my_binary directly after ExecStart=.

Improve this answer
6
  • 4
    Note: Writing to file was only added in systemd version 236. Commented Dec 20, 2018 at 5:54
  • @Dave Why log file name ends with a number ? Commented Jan 3, 2019 at 17:30
  • 6
    the file: option creates a new file the first time and then fails on subsequent restarts of the app for me, i tried append: and it didn't work at all Commented Jan 29, 2019 at 17:07
  • Last solution works fine, although it has to be /bin/sh -c ... (at least on Ubuntu). Commented Oct 8, 2019 at 21:32
  • 1
    This worked perfectly for me with Systemd v237. Commented Dec 13, 2021 at 20:14
27

ExecStart=/usr/bin/nohup …

This is wrong. Remove it. This service is not running in an interactive login session. There is no controlling terminal, or session leader, to send a hangup signal to it in the first place.

ExecStart=… &

This is wrong. Remove it. This is not shell script. & has no special shell-like meaning, and in any case would be the wrong way to start a service.

StandardOutput=/var/log/flume-ng/log1.log StandardError=/var/log/flume-ng/log2.log

These are wrong. Do not use these. systemd already sends the standard output and error of the service process(es) to its journal, without any such settings in the service unit. You can view it with 

journalctl -e -u flume-ng.service

Improve this answer
4
  • 74
    What if I wanted to send logs to a different file - is it possible or not? if possible, what is the correct way of doing this? Commented Mar 31, 2017 at 2:13
  • What is the location of the log file? I thought the question was about it. Commented Aug 12, 2020 at 13:56
  • @ka3ak assume the logfile is stored at $logfile Commented Sep 26, 2020 at 13:27
  • 7
    how is this selected as right answer? the question is how can we redirect service specific logs to a file that user want. not how to tail it. Commented Jun 29, 2021 at 6:40
18

For logging to file without overwriting (system.d version 240+ required):

StandardOutput=append:/var/log/flume-ng/log1.log StandardError=append:/var/log/flume-ng/log2.log

or

StandardOutput=append:/var/log/flume-ng/log.log StandardError=inherit
Improve this answer
2
  • 9
    This is unfortunately not working for me. The service still starts but nothing appears in the log file and if deleted it is not recreated. Ununtu 18.04, system.d 237 Commented Apr 9, 2019 at 11:08
  • 2
    @SystemicPlural you're right. This was only added in system.d 240 with a PullRequest for this issue: github.com/systemd/systemd/issues/10875 Commented Oct 8, 2019 at 21:01
7

I hope this helps someone. for Ununtu

[Unit] Description=name-service After=syslog.target After=mysql.service Requires=mysql.service [Service] Type=idle PIDFile=/path_to_PID/****.pid WorkingDirectory=/path_to_directory User=root Group=root ExecStart=****.py Restart=always TimeoutStartSec=1min RestartSec=1min KillMode=process StandardOutput=append:/path_to_log/service.log StandardError=append:/path_to_log/service_error.log [Install] WantedBy=multi-user.target

for CENTOS

[Unit] Description=name-service After=syslog.target After=mysql.service Requires=mysql.service [Service] Type=idle PIDFile=/path_to_PID/****.pid WorkingDirectory=/path_to_directory User=root Group=root ExecStart=****.py >>/path_to_log/log.log Restart=always TimeoutStartSec=1min RestartSec=1min KillMode=process [Install] WantedBy=multi-user.target

"After" and "Requires" of necessity. Attension "User" and "Group"

Another option: add logging inside executed scripts.

Improve this answer
2
  • What's your question? You should start a new question instead of posting it as an answer here. Commented Oct 9, 2019 at 8:53
  • 2
    I'm sorry. I have written it wrong. "I hope someone will help"->"I hope this helps someone." Commented Oct 9, 2019 at 13:33
2

If you don't get any output at all you should test your script first.

For example a Python script will buffer the output streams (but only when they are piped, not when you test on the console). So you will not see any log messages until the buffer is flushed.

Specify the -u switch ( python3 -u script) or set the PYTHONUNBUFFERED variable to avoid this:

Environment=PYTHONUNBUFFERED=1
Improve this answer
1
  • 1
    Wow, this is very helpful. In a perl script setting $|=1; also unbuffers. Fixed two of my units. Commented Apr 28, 2022 at 19:08
1 
[Unit] Description=ngrokd server [Service] Type=forking User=ngrok Group=ngrok ExecStart=/bin/sh -c '/usr/local/ngrok/bin/ngrokd \ -tlsCrt /usr/local/ngrok/tls/srv-ngrok.cert \ -tlsKey /usr/local/ngrok/tls/srv-ngrok.key \ -domain="ioio.cool" \ -httpAddr=:8480 \ -httpsAddr=:8433 \ -tunnelAddr=:8422 >> /var/log/ngrokd.log 2>&1 &' Restart=always KillMode=process [Install] WantedBy=multi-user.target

You can see man systemd.exec, search keywords StandardOutput, and man systemd.service, serach keywords COMMAND LINES

This syntax is intended to be very similar to shell syntax, but only the meta-characters and expansions described in the following paragraphs are understood. Specifically, redirection using "<", "<<", ">", and ">>", pipes using "|", running programs in the background using "&", and other elements of shell syntax are not supported.

Improve this answer
1
  • You can remove the trailing ampersand & from the commandline because this is not an interactive shell. The startup will be forked anyway. Commented Sep 5, 2022 at 10:50

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.