Community Articles

Find and share helpful community-sourced technical articles.
Celebrating as our community reaches 100,000 members! Thank you!
  • Goal
    • Create and Setup IPA Keytabs for HDP
    • Tested on CentOS 7 with IPA 4, HDP 2.3.0 and Ambari 2.1.1
  • Assumptions
    • Ambari does NOT currently (2.1.x) support the automatic generation of keytabs against IPA
    • IPA Server is already installed and IPA clients have been installed/configured on all HDP cluster nodes
    • If you are using Accumulo, you need to create the user 'tracer' in IPA. A keytab for this user will be requested for Accumulo Tracer.
    • We'll use Ambari's 'resources' directory to simplify the distribution of scripts and keys.

Enable kerberos using wizard

  • In Ambari, start security wizard by clicking Admin -> Kerberos and click Enable Kerberos. Then select "Manage Kerberos principals and key tabs manually" option Image
  • Enter your realm Image
  • (Optional, Read on before doing this...) Remove clustername from smoke/hdfs principals to remove the -${cluster_name} references to look like below
    • smoke user principal: ${cluster-env/smokeuser}@${realm}
    • HDFS user principal: ${hadoop-env/hdfs_user}@${realm}
    • HBase user principal: ${hbase-env/hbase_user}@${realm}
    • Spark user principal: ${spark-env/spark_user}@${realm}
    • Accumulo user principal: ${accumulo-env/accumulo_user}@${realm}
    • Accumulo Tracer User: tracer@${realm}

    If you don't remove the "cluster-name" from above, Ambari will generate/use principal names that are specific to your cluster. This could be very important if you are supporting multiple clusters with the same IPA implementation.


  • On next page download csv file but DO NOT click Next yet! Image
    • If you are deploying storm, the storm user maybe missing from the storm USER row. If you see something like the below:

replace the ,, with ,storm,

  • Copy above csv to the Ambari Server and place it in the /var/lib/ambari-server/resources directory, making sure to remove the header and any empty lines at the end.
vi kerberos.csv
  • From any IPA Client Node Create principals and service accounts using csv file.
## authenticate
kinit admin
# Get the kerberos.csv file from Ambari
wget http://${AMBARI_HOST}:8080/resources/kerberos.csv -O /tmp/kerberos.csv
# Create IPA service entries.
awk -F"," '/SERVICE/ {print "ipa service-add --force "$3}' /tmp/kerberos.csv | sort -u >
# Create IPA User accounts
awk -F"," '/USER/ {print "ipa user-add "$5" --first="$5" --last=Hadoop --shell=/sbin/nologin"}' /tmp/kerberos.csv | sort | uniq >
  • On Ambari node authenticate and create the keytabs for the headless user accounts and initialize the service keytabs.
## authenticate
sudo echo '<kerberos_password>' | kinit --password-file=STDIN admin
## or (IPA 4)
sudo echo '<kerberos_password>' | kinit -X password-file=STDIN admin

Should be run as root (adjust for your Ambari Host/Port).

wget http://${AMBARI_HOST_PORT}/resources/kerberos.csv -O /tmp/kerberos.csv
ipa_server=$(grep server /etc/ipa/default.conf |  awk -F= '{print $2}')
if [ "${ipa_server}X" == "X" ]; then
    ipa_server=$(grep host /etc/ipa/default.conf |  awk -F= '{print $2}')
if [ -d /etc/security/keytabs ]; then
    mv -f /etc/security/keytabs /etc/security/keytabs.`date +%Y%m%d%H%M%S`
mkdir -p /etc/security/keytabs
chown root:hadoop /etc/security/keytabs/
if [ ! -d /var/lib/ambari-server/resources/etc/security/keytabs ]; then
    mkdir -p /var/lib/ambari-server/resources/etc/security/keytabs
grep USER /tmp/kerberos.csv | awk -F"," '{print "ipa-getkeytab -s '${ipa_server}' -p "$3" -k "$6";chown "$7":"$9,$6";chmod "$11,$6}' | sort -u >
# Copy the 'user' keytabs to the Ambari Resources directory for distribution.
echo "cp -f /etc/security/keytabs/*.* /var/lib/ambari-server/resources/etc/security/keytabs/" >>
# ReGenerate Keytabs for all the required Service Account, EXCEPT for the HTTP service account on the IPA Server host.
grep SERVICE /tmp/kerberos.csv | awk -F"," '{print "ipa-getkeytab -s '${ipa_server}' -p "$3" -k "$6";chown "$7":"$9,$6";chmod "$11,$6}' | sort -u | grep -v HTTP\/${ipa_server} >>
# Allow the 'admins' group to retrieve the keytabs.
grep SERVICE /tmp/kerberos.csv | awk -F"," '{print "ipa service-allow-retrieve-keytab "$3" --group=admins"}' | sort -u >>
bash ./
# Now remove the keytabs, they'll be replaced by the distribution phase.
mv -f /etc/security/keytabs /etc/security/genedkeytabs.`date +%Y%m%d%H%M%S`
mkdir /etc/security/keytabs
chown root:hadoop /etc/security/keytabs
  • Build a distribution script used to create host specific keytabs (adjust for your Ambari Host/Port).
# Set the location of Ambari
# Retrieve the kerberos.csv file from the wizard
wget http://${AMBARI_HOST_PORT}/resources/kerberos.csv -O /tmp/kerberos.csv
ipa_server=$(grep server /etc/ipa/default.conf |  awk -F= '{print $2}')
if [ "${ipa_server}X" == "X" ]; then
    ipa_server=$(grep host /etc/ipa/default.conf |  awk -F= '{print $2}')
if [ ! -d /etc/security/keytabs ]; then
    mkdir -p /etc/security/keytabs
chown root:hadoop /etc/security/keytabs/
# Retrieve WITHOUT recreating the existing keytabs for each account.
grep USER /tmp/kerberos.csv | awk -F"," '/'$(hostname -f)'/ {print "wget http://'$( echo ${AMBARI_HOST_PORT})'/resources"$6" -O "$6";chown "$7":"$9,$6";chmod "$11,$6}' | sort -u >
grep SERVICE /tmp/kerberos.csv | awk -F"," '/'$(hostname -f)'/ {print "ipa-getkeytab -s '$(echo $ipa_server)' -r -p "$3" -k "$6";chown "$7":"$9,$6";chmod "$11,$6}' | sort -u >>
bash ./

Copy file to the Ambari Servers resource directory for distribution.

scp root@<ambari_host>:/var/lib/ambari-server/resources
  • On the Each HDP node via 'pdsh' authenticate and create the keytabs
# Should be logging in as 'root'
pdsh -g <host_group> -l root

> ## authenticate as the KDC Admin
> echo '<kerberos_password>' | kinit --password-file=STDIN admin
> # or (for IPA 4)
> echo '<kerberos_password>' | kinit -X password-file=STDIN admin
> wget http://<ambari_host>:8080/resources/ -O /tmp/
> bash /tmp/
> ## Verify kinit works before proceeding (should not give errors)
> # Service Account Check (replace REALM with yours)
> sudo -u hdfs kinit -kt /etc/security/keytabs/nn.service.keytab nn/$(hostname -f)@HORTONWORKS.COM
> # Headless Check (check on multiple hosts)
> sudo -u ambari-qa kinit -kt /etc/security/keytabs/smokeuser.headless.keytab ambari-qa@HORTONWORKS.COM
> sudo -u hdfs kinit -kt /etc/security/keytabs/hdfs.headless.keytab hdfs@HORTONWORKS.COM

WARNING: IPA UI maybe broken after above procedure

When the process to build keytabs for services is run on the same host that IPA lives on, it will invalidate the keytab used by Apache HTTPD to authenticate. I've added a step that should eliminate the "re"-creation of the key tab, but just incase..

Replace /etc/httpd/conf/ipa.keytab with /etc/security/keytabs/spnego.service.keytab

cd /etc/httpd/conf
mv ipa.keytab ipa.keytab.orig
cp /etc/security/keytabs/spnego.service.keytab ipa.keytab
chown apache:apache ipa.keytab
service httpd restart

Remove the headless.keytabs.tgz file from /var/lib/ambari-server/resources on the Ambari-Server.

  • Press next on security wizard and proceed to stop services Image
  • Proceed with the next steps of wizard by clicking Next ImageImage
  • Once completed, click Complete and now the cluster is kerborized Image

Using your Kerberized cluster

  1. Try to run commands without authenticating to kerberos.
    $ hadoop fs -ls /
    15/07/15 14:32:05 WARN ipc.Client: Exception encountered while connecting to the server : GSS initiate failed [Caused by GSSException: No valid credentials provided (Mechanism level: Failed to find any Kerberos tgt)]
    $ curl -u someuser -skL "http://$(hostname -f):50070/webhdfs/v1/user/?op=LISTSTATUS"
    <title>Error 401 Authentication required</title>
  2. Get a token
    ## for the current user
    sudo su - gooduser
    ## for any other user
    kinit someuser
  3. Use the cluster
  • Hadoop Commands
    $ hadoop fs -ls /
    Found 8 items
  • WebHDFS
    ## note the addition of `--negotiate -u : `
    curl -skL --negotiate -u : "http://$(hostname -f):50070/webhdfs/v1/user/?op=LISTSTATUS"
  • Hive (using Beeline or another Hive JDBC client)
    • Hive in Binary mode (the default)
    beeline -u "jdbc:hive2://localhost:10000/default;principal=hive/$(hostname -f)@HORTONWORKS.COM"
    • Hive in HTTP mode
    ## note the update to use HTTP and the need to provide the kerberos principal.
    beeline -u "jdbc:hive2://localhost:10001/default;transportMode=http;httpPath=cliservice;principal=HTTP/$(hostname -f)@HORTONWORKS.COM"

Thank you to (Ali Bajwa) for his original workshop this is intended to extend.


Great guide found this very useful thanks!

Very helpful, thank you.

For a while, I thought that the "getkeytab retrieve mode" would allow getting headless keytab on all nodes without incrementing the KVNO. But, if I get it well, the retrieve mode only can be used for service keytabs, and not headless ones, right?

Expert Contributor

Thanks @David Streever

Very helpful post.

For spnego, i have added

ipa-getkeytab -s IPA_SERVER -p HTTP/NODE1@REALM -k /etc/security/keytabs/spnego.service.keytab ipa-getkeytab -s IPA_SERVER -p HTTP/NODE2@REALM -k /etc/security/keytabs/spnego.service.keytab ipa-getkeytab -s IPA_SERVER -p HTTP/MASTER1@REALM -k /etc/security/keytabs/spnego.service.keytab


I am trying to get this working with Ambari 2.6.0 and HDP 2.6.3 with IPA 4.5.2 in centos 7

I have posted some questions about what local users and USER princ names....

Has anyone gotten this to work with Ambari 2.6.0 I am getting errors and I think it is probably something simple.

@David Streever

Hi David, I am trying to enable kerberos on a cluster running Ambari 2.6.0 with HDP 2.6.3 and IPA 4.5.2

I want to keep the cluster name in the Ambari USER names.... When I use the above procedure I run into problems when the USER principals are created and subsequently when the keytabs are generated. It looks like the Ambari wizard does not change the local user name (%5) to also have the lower case cluster name... so that when the USER principals are created they are created as local user name without the cluster name. Then when running I get the following:

Failed to parse result: PrincipalName not found.

Retrying with pre-4.0 keytab retrieval method...

Failed to parse result: PrincipalName not found.

Failed to get keytab!

Failed to get keytab

chown: cannot access ‘/etc/security/keytabs/smokeuser.headless.keytab’: No such file or directory

chmod: cannot access ‘/etc/security/keytabs/smokeuser.headless.keytab’: No such file or directory

Failed to parse result: PrincipalName not found.

I can see why this happens but I am unsure as to what the USER name should be... in other words do I edit the kerberos.csv so that local username matches the new Kerberos principal? Do the host's local usernames that are local to each host in the cluster need to match the Kerberos USER Princ names? I have tried with and without cluster name and I still run into errors during the Start and Test phase having to do with credentials not working.

I am hoping once I figure this all out I can create a new HOWTO for IPA-Manual Princ process. FYI I was unable to get the Ambari Automatic Kerberization to work using the FreeIPA experimental feature before moving on to attempting your manual process.

Any insights or assistance is much appreciated.