Rudder

Hi,

In a multiserver setup, the server roles are overwritten with data loss at some regular (?) interval.

Here's an excerpt of the script I use to init the rudderweb instance, based on the one from rudder-tools git repo:

## Inform Rudder about the various roles so installation can proceed successfully
mkdir -p /var/rudder/cfengine-community/inputs
cat >/var/rudder/cfengine-community/inputs/rudder-server-roles.conf <<EOF
rudder-ldap:rudderldap01
rudder-inventory-endpoint:rudderldap01
rudder-db:rudderdb
rudder-front:rudderrelaytop01
rudder-relay-top:rudderrelaytop01
rudder-webapp:rudderweb
rudder-web:rudderweb
EOF

## Install components
yum install -y rudder-agent rudder-webapp rudder-techniques rsyslog rsyslog-pgsql httpd httpd-tools

## Give more RAM to jetty
#<snip>

## Init DB (/opt/rudder/etc/postgresql/reportsSchema.sql from rudder-reports-3.0.2.release-1.EL.6.noarch)
#<snip, this performs the init our rds instance, set the proper psql password in the conf, etc>

## Initialize the server: hostname, demodata?, ldapreset?, initialpromisesreset?, allowednet
sed -i '/^Pause$/d' /opt/rudder/bin/rudder-init # ensure that it won't hang waiting for user input
/opt/rudder/bin/rudder-init yes <our networks>

After that, rudder seems to work fine for some time, at least the UI. But at some point we can't get any report.

Indeed, because /etc/rsyslog.d/rudder.conf contains:

:ompgsql:localhost,rudder,rudder,password;RudderReportsFormat

localhost instead of rudderdb!
rudder.jdbc.* keys in /opt/rudder/etc/rudder-web-properties have the proper values.

Now, if I look at /var/rudder/cfengine-community/inputs/rudder-server-roles.conf, it has been overwritten by something:

# This file contains, for each role, the list of nodes that have this role
# These roles are auto-detected by default, based on inventories of nodes that have role files in /opt/rudder/etc/server-roles.d/
# Format is the following
# role:hostname1,hostname2,hostname3
rudder-ldap:rudderldap01
rudder-inventory-endpoint:rudderldap01
rudder-relay-top:rudderrelaytop01

rudderdb is gone!

If I put it back and run cf-agent, then the rsyslog config is ok.

However, it's somehow overwritten again at some point, because I restored the proper value yesterday evening, and this morning it's broken again.

Thanks.