[AMBARI-21028] The credential cache for livy is messed up

• Type: Bug
• Status: Resolved
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: trunk
• Fix Version/s: trunk, 2.5.1
• Component/s: ambari-server
• Assignee: Weiqing Yang
• Reporter: Weiqing Yang
• Created: 15/May/17 21:01
• Updated: 17/May/17 03:10
• Resolved: 17/May/17 02:16

Description

This issue was reported by kbadani.

Steps to reproduce this issue:

• Kdestroy and kinit as 'livy' user
• Do spark-submit with --proxy-user as 'hrt_1'
• In the console output, you can see that 'ambari-qa' is trying to impersonate as 'hrt_1' and its failing
• Cancel the running job and do klist again - it will show credentials for 'ambari-qa' user and not the 'livy' user with which it was kinited

[livy@ctr-e133-1493418528701-6489-01-000003 spark]$ kinit -kt /etc/security/keytabs/livy.service.keytab livy/ctr-e133-1493418528701-6489-01-000003.hwx.site

[livy@ctr-e133-1493418528701-6489-01-000003 spark]$ klist
Ticket cache: FILE:/tmp/krb5cc_1808
Default principal: livy/[e-mail]

Valid starting       Expires              Service principal
05/02/2017 23:52:14  05/03/2017 23:52:14  krbtgt/[e-mail]

[livy@ctr-e133-1493418528701-6489-01-000003 spark]$ spark-submit --class org.apache.spark.examples.SparkPi --master yarn-cluster --num-executors 3 --driver-memory 512m --executor-memory 512m --proxy-user hrt_1 --executor-cores 1 /usr/hdp/current/spark-client/lib/spark-examples-1.6.3.2.6.1.0-45-hadoop2.7.3.2.6.1.0-45.jar 10
Multiple versions of Spark are installed but SPARK_MAJOR_VERSION is not set
Spark1 will be picked by default
17/05/02 23:53:10 WARN NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
17/05/02 23:53:12 INFO AHSProxy: Connecting to Application History server at ctr-e133-1493418528701-6489-01-000004.hwx.site/172.27.22.136:10200
17/05/02 23:53:12 INFO RequestHedgingRMFailoverProxyProvider: Looking for the active RM in [rm1, rm2]...
17/05/02 23:53:12 WARN RequestHedgingRMFailoverProxyProvider: Invocation returned exception on [rm1] : org.apache.hadoop.security.authorize.AuthorizationException: User: [e-mail] is not allowed to impersonate hrt_1, so propagating back to caller.
17/05/02 23:53:12 INFO RequestHedgingRMFailoverProxyProvider: Connection lost with rm1, trying to fail over.
17/05/02 23:53:12 INFO RequestHedgingRMFailoverProxyProvider: Looking for the active RM in [rm1, rm2]...
17/05/02 23:53:12 WARN RequestHedgingRMFailoverProxyProvider: Invocation returned exception on [rm1] : org.apache.hadoop.security.authorize.AuthorizationException: User: [e-mail] is not allowed to impersonate hrt_1, so propagating back to caller.
17/05/02 23:53:12 INFO RetryInvocationHandler: org.apache.hadoop.security.authorize.AuthorizationException: User: [e-mail] is not allowed to impersonate hrt_1, while invoking $Proxy9.getClusterMetrics over Failover proxy for [rm1, rm2] after 1 failover attempts. Trying to failover after sleeping for 10442ms.

[livy@ctr-e133-1493418528701-6489-01-000003 spark]$ klist
Ticket cache: FILE:/tmp/krb5cc_1808
Default principal: [e-mail]

Valid starting       Expires              Service principal
05/02/2017 23:53:03  05/03/2017 23:53:03  krbtgt/[e-mail]
05/02/2017 23:53:03  05/03/2017 23:53:03  HTTP/[e-mail]

Root cause is:
The livy smoke test launched by Ambari is run as livy user, but kinits as ambari-qa, and therefore messes up the credential cache for livy.

Issue Links

• links to: ReviewBoard