[
https://issues.apache.org/jira/browse/CASSANDRA-13436?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16067859#comment-16067859
]
Murukesh Mohanan commented on CASSANDRA-13436:
----------------------------------------------
I had a look at the current init script, and tried writing a systemd
equivalent. This raised some questions:
1. The part that doesn't easily translate to systemd is the JVM search code
(which searches for Java in certain directories and sets the command and
{{JAVA_HOME}} appropriately. Perhaps we should leave it to the system
administrator to set it in {{/etc/default/cassandra}}?
2. The PID file becomes unnecessary since systemd can track the processes
started by a service. But I'm not sure if nothing outside the init script uses
the PID file.
3. The log file could also be considered unnecessary since systemd-journald
will gather up any output. But, again, people might expect the logs to stored
in {{/var/log/cassandra/cassandra.log}}.
So, we could have a minimal systemd unit like so (running cassandra in the
foreground making it a simple service):
{code}
[Unit]
Description=Cassandra is a distributed storage system for structured data.
[Service]
Environment=CASSANDRA_HOME=/usr/share/cassandra
CASSANDRA_CONF=/etc/cassandra/conf
CASSANDRA_INCLUDE=/usr/share/cassandra/cassandra.in.sh
EnvironmentFile=-/etc/default/cassandra
User=cassandra
ExecStart=/usr/sbin/cassandra -f
TimeoutStopSec=25
SuccessExitStatus=0 143
[Install]
WantedBy=multi-user.target
{code}
Or more verbose:
{code}
[Unit]
Description=Cassandra is a distributed storage system for structured data.
[Service]
Environment=CASSANDRA_HOME=/usr/share/cassandra
CASSANDRA_CONF=/etc/cassandra/conf
CASSANDRA_INCLUDE=/usr/share/cassandra/cassandra.in.sh
EnvironmentFile=-/etc/default/cassandra
User=cassandra
PermissionsStartOnly=true
ExecStartPre=/usr/bin/install -m 755 -o cassandra -g cassandra -d
/var/run/cassandra
ExecStart=/bin/sh -c '/usr/sbin/cassandra -p /var/run/cassandra/cassandra.pid >
/var/log/cassandra/cassandra.log 2>&1'
Type=forking
PIDFile=/var/run/cassandra/cassandra.pid
TimeoutStopSec=25
SuccessExitStatus=0 143
[Install]
WantedBy=multi-user.target
{code}
I prefer the simpler unit. Thoughts?
> Stopping Cassandra shows status "failed" due to non-zero exit status
> --------------------------------------------------------------------
>
> Key: CASSANDRA-13436
> URL: https://issues.apache.org/jira/browse/CASSANDRA-13436
> Project: Cassandra
> Issue Type: Sub-task
> Components: Packaging
> Reporter: Stefan Podkowinski
>
> Systemd will monitor the process from the pid file and save the return status
> once if has been stopped. In case the process terminates with a status other
> than zero, it will assume the process terminated abnormaly. Stopping
> Cassandra using the cassandra script will send a kill signal to the JVM
> causing it to terminate. If this happen, the JVM will exit with status 143,
> no matter if shutdown hooks have been executed or not. In order to make
> systemd recognize this as a normal exit code, the following should be added
> to the yet to be created unit file:
> {noformat}
> [Service]
> ...
> SuccessExitStatus=0 143
> ...
> {noformat}
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
