[AMBARI-18296] Database Consistency Check Fails With NPE With Missing Service From Stack

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 2.4.0
• Fix Version/s: trunk, 2.4.1, 2.5.0, 2.4.2
• Component/s: ambari-server
• Assignee: Vitaly Brodetskyi
• Reporter: Jonathan Hurley
• Created: 01/Sep/16 15:00
• Updated: 07/Sep/16 23:07
• Resolved: 03/Sep/16 01:04

Description

The database consistency checker can fail if a service configuration exists for a service which is no longer on the current stack:

2016-08-31 21:52:28,082 INFO - ******************************* Check database started ******************************* 2016-08-31 21:52:31,647 INFO - Checking for configs not mapped to any cluster 2016-08-31 21:52:31,653 INFO - Checking for configs selected more than once 2016-08-31 21:52:31,655 INFO - Checking for hosts without state 2016-08-31 21:52:31,657 INFO - Checking host component states count equals host component desired states count 2016-08-31 21:52:31,660 INFO - Checking services and their configs 2016-08-31 21:52:33,669 ERROR - Unexpected error, database check failed java.lang.NullPointerException at org.apache.ambari.server.checks.DatabaseConsistencyCheckHelper.checkServiceConfigs(DatabaseConsistencyCheckHelper.java:543) at org.apache.ambari.server.checks.DatabaseConsistencyChecker.main(DatabaseConsistencyChecker.java:115)

It seems like what happens is this query returns a service which is not defined in the current stack:

SELECT
  c.cluster_name,
  cs.service_name,
  cc.type_name,
  sc.version
FROM clusterservices cs
JOIN serviceconfig sc
  ON cs.service_name = sc.service_name
  AND cs.cluster_id = sc.cluster_id
JOIN serviceconfigmapping scm
  ON sc.service_config_id = scm.service_config_id
JOIN clusterconfig cc
  ON scm.config_id = cc.config_id
  AND sc.cluster_id = cc.cluster_id
JOIN clusters c
  ON cc.cluster_id = c.cluster_id
  AND sc.stack_id = c.desired_stack_id
WHERE sc.group_id IS NULL
AND sc.service_config_id = (SELECT
  MAX(service_config_id)
FROM serviceconfig sc2
WHERE sc2.service_name = sc.service_name
AND sc2.cluster_id = sc.cluster_id)
GROUP BY c.cluster_name,
         cs.service_name,
         cc.type_name,
         sc.version

Problem area of code:

serviceInfo is null

        for (String serviceName : serviceNames) {
          ServiceInfo serviceInfo = serviceInfoMap.get(serviceName);
          Set<String> configTypes = serviceInfo.getConfigTypeAttributes().keySet();
          for (String configType : configTypes) {
            stackServiceConfigs.put(serviceName, configType);
          }
        }

Issue Links

• links to: review request