Frequently Asked Questions


How can I set up LDAP authentication?

You will need to install django-auth-ldap and python-ldap using pip. Then add LDAP as an authentication backend in your local settings:

LOGGING['loggers']['django_auth_ldap'] = {
    'handlers': ['file'],
    'propagate': True,
    'level': 'DEBUG',

You will need to configure the LDAP connection as well as mapping remote users and groups to local ones. For details consult the official django-auth-ldap documentation. For example, connecting to an Active Directory service might look like this:

import ldap
from django_auth_ldap.config import LDAPSearch, GroupOfNamesType
AUTH_LDAP_SERVER_URI = "ldap://activedirectory.domain:389"
AUTH_LDAP_BIND_DN = "secret"
AUTH_LDAP_USER_SEARCH_BASE = "DC=allegrogroup,DC=internal"
AUTH_LDAP_USER_SEARCH_FILTER = '(&(objectClass=*)({0}=%(user)s))'.format(
AUTH_LDAP_USER_ATTR_MAP = {"first_name": "givenName", "last_name": "sn",
    "email": "mail"}
    "company": "company",
    "manager": "manager",
    "department": "department",
    "employee_id": "employeeID",
    "location": "officeName",
    "country": "ISO-country-code",

However, when using OpenDJ as a LDAP server, AUTH_LDAP_USER_USERNAME_ATTR should be equal to uid:


For other implementations objectClass may have the following values:

  • Active Directory: objectClass=user,
  • Novell eDirectory: objectClass=inetOrgPerson,
  • Open LDAP: objectClass=posixAccount

Manager is special field and is treated as reference to another user, for example “CN=John Smith,OU=TOR,OU=Corp-Users,DC=mydomain,DC=internal” is mapped to “John Smith” text.

Country is special field, the value of this field must be a country code in the ISO 3166-1 alfa-2 format.

Ralph provides ldap groups to django groups mapping. All what you need to do are:

  • import custom MappedGroupOfNamesType,
  • set up group mirroring,
  • declare mapping.
from ralph.account.ldap import MappedGroupOfNamesType
  'CN=_gr_ralph,OU=Other,DC=mygroups,DC=domain': "active",
  'CN=_gr_ralph_assets_buyer,OU=Other,DC=mygroups,DC=domain': "assets-buyer",
  'CN=_gr_ralph_assets_helper,OU=Other,DC=mygroups,DC=domain': "assets-helper",
  'CN=_gr_ralph_assets_staff,OU=Other,DC=mygroups,DC=domain': "assets-staff",
  'CN=_gr_ralph_admin,OU=Other,DC=mygroups,DC=domain': "superuser",
  'CN=_gr_ralph_staff,OU=Other,DC=mygroups,DC=domain': "staff",
AUTH_LDAP_GROUP_TYPE = MappedGroupOfNamesType(name_attr="cn")
AUTH_LDAP_GROUP_SEARCH = LDAPSearch("DC=organization,DC=internal",
    ldap.SCOPE_SUBTREE, '(objectClass=group)')

Note: For OpenDJ implementation AUTH_LDAP_GROUP_MAPPING is not obligatory. AUTH_LDAP_GROUP_TYPE and AUTH_LDAP_GROUP_SEARCH should be set as follows:

from django_auth_ldap.config import GroupOfUniqueNamesType
AUTH_LDAP_GROUP_TYPE = GroupOfUniqueNamesType()
AUTH_LDAP_GROUP_SEARCH = LDAPSearch("DC=organization,DC=internal",
  ldap.SCOPE_SUBTREE, '(structuralObjectClass=groupOfUniqueNames)')

If you want to define ldap groups with names identical to ralph roles, you shouldn’t declare mapping AUTH_LDAP_GROUP_MAPPING. If there are any one mapping defined another groups will be filtered. Some groups have special meanings. For example users need to be in active to log in, superuser gives superuser privileges. You can read more info in Groups.

You can define users filter, if you don’t want to import all users to ralph:

AUTH_LDAP_USER_FILTER = '(|(memberOf=CN=_gr_ralph_group1,OU=something,'\
    'DC=mygroup,DC=domain)(memberOf=CN=_gr_ralph_group2,OU=something else,'\

In case of OpenDJ please use isMemberOf instead of memberOf.

To synchronize user list you must run command:

ralph ldap_sync

During the process, script will report progress on every 100-th item loaded.


Can I start Gunicorn using the traditional start-stop-daemon?

Sure, here is the init.d recipe for Debian/Ubuntu: /etc/init.d/gunicorn. Just alter the CONFDIR and VENV_ACTIVATE variables. Note: service gunicorn restart only HUPs the server. If an actual restart is necessary, use the service gunicorn force-restart command.

If you happen to have a script like this for another operating system, contact us so to include it here.


The web app shows error screens with database warnings about incorrect characters.

On many systems the default configuration of MySQL is not optimal. Make sure that your tables are using UTF-8 as the character set, with compatible collation and InnoDB storage engine. Example:

mysql> alter table TABLENAME engine=innodb;
mysql> alter table TABLENAME convert to character set utf8 collate utf8_general_ci;

My worker creates a new database connection on each task.

This is a known limitation of Django. The best solution is to set up a mysql-proxy instance which will reuse actual database connections and set up Ralph to use that. First install mysql-proxy:

$ sudo apt-get install mysql-proxy

Then edit /etc/default/mysql-proxy so it says:

OPTIONS="--proxy-backend-addresses=mysqlserverhost.local:3306 --log-level=info --log-use-syslog --proxy-address= --admin-username=ralph --admin-password=ralph --admin-lua-script=/usr/lib/mysql-proxy/lua/admin.lua"

Start the proxy:

$ sudo service mysql-proxy start

In /var/log/syslog you should see:

Jul 25 10:44:14 s10337 mysql-proxy: 2012-07-25 10:44:14: (message) mysql-proxy 0.8.1 started
Jul 25 10:44:14 s10337 mysql-proxy: 2012-07-25 10:44:14: (message) proxy listening on port
Jul 25 10:44:14 s10337 mysql-proxy: 2012-07-25 10:44:14: (message) added read/write backend: mysqlserverhost.local:3306

Then alter your settings so the DATABASES dictionary points at the proxy address and not at the actual database, restart Ralph and you’re done.

The web app creates a new database connection on each request.

See above.

My worker leaves too many connections to the database open.

See above.

RQ workers

How to check how many tasks are waiting on the queue?

Just install rq-dashboard to control RQ queues:

$ pip install rq-dashboard

To use it, just run rq-dashboard from commandline, and fire up browser on port 9181:

$ rq-dashboard
Running on


There are large amounts of sockets in TIME_WAIT state on the worker machine. What is this?

Sockets that are closed wait for 60 more seconds to handle possible duplicate packets and ensure the other party received the ACK. For massively concurrent workers this can lead to tens of thousands of sockets in the TIME_WAIT state. The worker machines are dedicated to scan the local network so you can safely shorten keepalive to 5 * 30 seconds and the timeout interval to 10 seconds and by issuing:

$ sysctl -w net.ipv4.tcp_fin_timeout=10
$ sysctl -w net.ipv4.tcp_keepalive_probes=5
$ sysctl -w net.ipv4.tcp_keepalive_intvl=30

Additionally, if you don’t use a load balancer on the worker machine, you can safely recycle TIME_WAIT sockets:

$ sysctl -w net.ipv4.tcp_tw_reuse=1
$ sysctl -w net.ipv4.tcp_tw_recycle=1

The current number of waiting connections can be checked by:

$ sudo netstat -natup | grep "^tcp" | wc -l

On a large subnetwork I’m getting ipv4: Neighbour table overflow. in dmesg.

Your ARP table is overflowing. Increase the limits:

$ sudo sysctl -w net.ipv4.neigh.default.gc_thresh3=8192
$ sudo sysctl -w net.ipv4.neigh.default.gc_thresh2=8192
$ sudo sysctl -w net.ipv4.neigh.default.gc_thresh1=4096
$ sudo sysctl -w net.ipv4.neigh.default.base_reachable_time=86400
$ sudo sysctl -w net.ipv4.neigh.default.gc_stale_time=86400

How to handle "No buffer space available" errors on sockets?

See the two above.