Error Messages and Various Failures

Symptom: Cluster previously configured without Kerberos authentication may fail to run jobs for certain users on certain TaskTrackers (MRv1) or NodeManagers (YARN) after enabling Kerberos for the cluster. Errors may display in the TaskTracker or NodeManager logs. The following example errors are from TaskTracker on MRv1:

10/11/03 01:29:55 INFO mapred.JobClient: Task Id : attempt_201011021321_0004_m_000011_0, Status : FAILED
Error initializing attempt_201011021321_0004_m_000011_0: 
java.io.IOException: org.apache.hadoop.util.Shell$ExitCodeException: 
at org.apache.hadoop.mapred.LinuxTaskController.runCommand(LinuxTaskController.java:212) 
at org.apache.hadoop.mapred.LinuxTaskController.initializeUser(LinuxTaskController.java:442) 
at org.apache.hadoop.mapreduce.server.tasktracker.Localizer.initializeUserDirs(Localizer.java:272) 
at org.apache.hadoop.mapred.TaskTracker.localizeJob(TaskTracker.java:963) 
at org.apache.hadoop.mapred.TaskTracker.startNewTask(TaskTracker.java:2209) 
at org.apache.hadoop.mapred.TaskTracker$TaskLauncher.run(TaskTracker.java:2174) 
Caused by: org.apache.hadoop.util.Shell$ExitCodeException: 
at org.apache.hadoop.util.Shell.runCommand(Shell.java:250) 
at org.apache.hadoop.util.Shell.run(Shell.java:177) 
at org.apache.hadoop.util.Shell$ShellCommandExecutor.execute(Shell.java:370) 
at org.apache.hadoop.mapred.LinuxTaskController.runCommand(LinuxTaskController.java:203) 
... 5 more

Steps to resolve: Delete mapred.local.dir or yarn.nodemanager.local-dirs directories across the cluster for affected users.

Caused by: KrbException: Integrity check on decrypted field failed (31) - PREAUTH_FAILED}}

Caused by: KrbException: Identifier does not match expected value (906)

Possible cause: This issue may be due to incorrect configuration for AES-256. By default, certain operating systems—CentOS/Red Hat Enterprise Linux 5.6 (and higher), Ubuntu—use AES-256 encryption which requires installing the Java Cryptography Extension (JCE) Unlimited Strength Jurisdiction Policy File on all hosts (see JCE Policy File for AES-256 Encryption for details), or disabling AES-256 support in the kdc.conf or krb5.com (see disable AES-256 encryption from the Kerberos instance for details).

Steps to resolve: KrbException 31 and KrbException 906 can be caused by various issues, but the most likely cause is incorrectly configured AES-256 encryption. Resolving the issue should start by determining the type of encryption configured for the cluster.

To verify the type of encryption configured for the cluster:

1. On the local KDC host, type this command to create a test principal:

   kadmin -q "addprinc test" 

2. On a cluster host, type this command to start a Kerberos session as test:

   kinit test 

3. On a cluster host, type this command to view the encryption type in use:

   klist -e 

   If AES-256 is being used, output such as the following displays:

   Ticket cache: FILE:/tmp/krb5cc_0
   Default principal: test@SCM
   Valid starting     Expires            Service principal
   05/19/11 13:25:04  05/20/11 13:25:04  krbtgt/SCM@SCM
       Etype (skey, tkt): AES-256 CTS mode with 96-bit SHA-1 HMAC, AES-256 CTS mode with 96-bit SHA-1 HMAC 

To remove AES-256 encryption from the Kerberos configuration files:

• Remove aes256-cts:normal from the supported_enctypes field of the kdc.conf or krb5.conf file.
• After changing the configuration, restart the KDC and the kadmin servers.
• Change TGT principal (krbtgt/REALM@REALM) and other principal passwords as needed.
• In the [realms] section of the kdc.conf file, for the realm associated with HADOOP.LOCALDOMAIN, add (or replace if it exists already) the following variable:

  supported_enctypes = des3-hmac-sha1:normal arcfour-hmac:normal des-hmac-sha1:normal des-cbc-md5:normal des-cbc-crc:normal des-cbc-crc:v4 des-cbc-crc:afs3

• Recreate the hdfs keytab file and mapred keytab file using the instructions in Viewing or Regenerating Kerberos Credentials Using Cloudera Manager.

Symptom: The NameNode keytab file does not have an AES-256 entry but the client tickets do. The NameNode starts but clients cannot connect to it. The error message does not specify "AES256" but rather contains enctype code "18."

Possible cause: Issue related to AES-256 encryption and the JCE library.

Steps to resolve: Verify that Java Cryptography Extension (JCE) Unlimited Strength Jurisdiction Policy File is installed or remove aes256-cts:normal from the supported_enctypes field of the kdc.conf or krb5.conf file, as detailed above.

Symptom: After enabling cross-realm trust, authenticating as a principal in the local realm lets you successfully run Hadoop commands, but authenticating as a principal in the remote realm does not.

Possible cause: This issue is often due to principals in the two realms having different encryption types or different passwords for the cross-realm principal in each realm. Because the local and remote realm each issue their own TGTs, the local commands run but the service ticket needed for the local and remote realms to communicate cannot be granted.

kadmin: addprinc -e "enc_type_list" krbtgt/LOCAL-REALM.EXAMPLE.COM@MAIN-REALM.COMPANY.COM

kadmin:  addprinc -e "aes256-cts:normal rc4-hmac:normal des3-hmac-sha1:normal" krbtgt/LOCAL-REALM.EXAMPLE.COM@MAIN-REALM.COMPANY.COM

13/01/15 17:44:48 DEBUG ipc.Client: Exception encountered while connecting to the server : javax.security.sasl.SaslException:
GSS initiate failed [Caused by GSSException: No valid credentials provided (Mechanism level: Fail to create credential. 
(63) - No service creds)]

Possible cause: Ticket message may be too large for the UDP protocol (which is used by SASL by default).

[libdefaults]
udp_preference_limit = 1

Configure krb5.conf through Cloudera Manager, this will automatically get added to krb5.conf.

Symptom: Multiple valid requests to Kerberos protected services are identified as replay attempts when they are not. The following exception shows up in the logs for one or more of the Hadoop daemons:

2013-02-28 22:49:03,152 INFO  ipc.Server (Server.java:doRead(571)) - IPC Server listener on 8020: readAndProcess threw exception javax.security.sasl.SaslException: GSS initiate failed [Caused by GSSException: Failure unspecified at GSS-API level (Mechanism l
javax.security.sasl.SaslException: GSS initiate failed [Caused by GSSException: Failure unspecified at GSS-API level (Mechanism level: Request is a replay (34))]
        at com.sun.security.sasl.gsskerb.GssKrb5Server.evaluateResponse(GssKrb5Server.java:159)
        at org.apache.hadoop.ipc.Server$Connection.saslReadAndProcess(Server.java:1040)
        at org.apache.hadoop.ipc.Server$Connection.readAndProcess(Server.java:1213)
        at org.apache.hadoop.ipc.Server$Listener.doRead(Server.java:566)
        at org.apache.hadoop.ipc.Server$Listener$Reader.run(Server.java:363)
Caused by: GSSException: Failure unspecified at GSS-API level (Mechanism level: Request is a replay (34))
        at sun.security.jgss.krb5.Krb5Context.acceptSecContext(Krb5Context.java:741)
        at sun.security.jgss.GSSContextImpl.acceptSecContext(GSSContextImpl.java:323)
        at sun.security.jgss.GSSContextImpl.acceptSecContext(GSSContextImpl.java:267)
        at com.sun.security.sasl.gsskerb.GssKrb5Server.evaluateResponse(GssKrb5Server.java:137)
        ... 4 more
Caused by: KrbException: Request is a replay (34)
        at sun.security.krb5.KrbApReq.authenticate(KrbApReq.java:300)
        at sun.security.krb5.KrbApReq.<init>(KrbApReq.java:134)
        at sun.security.jgss.krb5.InitSecContextToken.<init>(InitSecContextToken.java:79)
        at sun.security.jgss.krb5.Krb5Context.acceptSecContext(Krb5Context.java:724)
        ... 7 more

This issue can also manifest as poor performance for clients of the cluster, including dropped connections, timeouts attempting to make RPC calls, and so on.

Possible cause: Kerberos uses a second-resolution timestamp to protect against replay attacks (where an attacker can record network traffic, and play back recorded requests later to gain elevated privileges). That is, incoming requests are cached by Kerberos for a little while, and if there are similar requests within a few seconds, Kerberos will be able to detect them as replay attack attempts (see MIT Kerberos replay cache for more information).

However, if there are multiple valid Kerberos requests coming in at the same time, these may also be misjudged as attacks for the following reasons:

• Multiple services in the cluster are using the same Kerberos principal. All secure clients that run on multiple machines should use unique Kerberos principals for each machine. For example, rather than connecting as a service principal myservice@EXAMPLE.COM, services should have per-host principals such as myservice/host123.example.com@EXAMPLE.COM.
• Clocks not synchronized: All hosts should run NTP so that clocks are kept in sync between clients and servers.

Steps to resolve:

While having different principals for each service, and clocks in sync helps mitigate the issue, there are, however, cases where even if all of the above are implemented, the problem still persists. In such a case, disabling the cache (and the replay protection as a consequence), will allow parallel requests to succeed. This compromise between usability and security can be applied by setting the KRB5RCACHETYPE environment variable to none.

Note that the KRB5RCACHETYPE is not automatically detected by Java applications. For Java-based components:

• Ensure that the cluster runs on JDK 8.
• To disable the replay cache, add -Dsun.security.krb5.rcache=none to the Java Opts/Arguments of the targeted JVM. For example, HiveServer2 or the Sentry service.

Exception in secureMain 
java.lang.ExceptionInInitializerError 
at javax.crypto.KeyGenerator.nextSpi(KeyGenerator.java:324) 
at javax.crypto.KeyGenerator.<init>(KeyGenerator.java:157) 
...
Caused by: java.lang.SecurityException: The jurisdiction policy files are not signed by a trusted signer! 
at javax.crypto.JarVerifier.verifyPolicySigned(JarVerifier.java:289) 
at javax.crypto.JceSecurity.loadPolicies(JceSecurity.java:316) 
at javax.crypto.JceSecurity.setupJurisdictionPolicies(JceSecurity.java:261) 
...

Possible cause: This is another example of a mismatch for AES-256 encryption. Services cannot start when the version of the JCE policy file does not match the version of Java installed on a node because the cryptographic signatures for the JCE policy files cannot be verified, resulting in the message shown above.

Download and unpack the zip file. Copy the two JAR files to the $JAVA_HOME/jre/lib/security directory on each node within the cluster.