PH50064: CONNECTION BROKEN WHEN TAKEOVER OCCURS

APAR status

• Closed as program error.

Error description

• Long running connection broken after highavailability
  takeover. Replication enabled for the connection.
  

Local fix

• N/A
  

Problem summary

• ****************************************************************
  * USERS AFFECTED:  IBM WebSphere Application Server Network    *
  *                  Deployment Load Balancer for IPv4 and IPv6  *
  *                  users                                       *
  ****************************************************************
  * PROBLEM DESCRIPTION: A connection may be broken during a     *
  *                      takeover when Load Balancer is          *
  *                      configured with highavailability        *
  *                      replication.                            *
  ****************************************************************
  * RECOMMENDATION:                                              *
  ****************************************************************
  Existing connections should not be broken when a high
  availability takeover occurs and connection replication is
  defined on the Load Balancer.
  This problem was observed on AIX with LB_PROMISICUOUS enabled.
  Clusters were aliased to the ethernet interface using LB's
  go scripts.
  When a highavailability takeover occurs, the Load Balancer's
  executor transitions from active forwarding to backup mode
  immedidately. Any cluster traffic received by the Load
  Balancer after this state change is not forwarded. Cluster
  traffic is not discarded but packets are passed to the
  operating system for processing to allow for collocated
  servers.
  The Load Balancer server code is responsible for invoking the
  go scripts when a state change is detected. The server code
  polls the executor code every two seconds to obtain state
  information. The operating system will respond to any cluster
  traffic received when the cluster is aliased on the ethernet
  interface. If the cluster is not aliased on the ethernet
  interface, the operating system discards the traffic.
  When cluster traffic is received and the executor is running
  in the backup state, if the cluster is still aliased on the
  operating system interface, the operating system will respond
  with a RST packet if the traffic is not for a collocated
  server. The RST packet will terminate the existing connection.
  

Problem conclusion

• This problem will only occur when clusters are aliased to the
  systems ethernet interfaces. Only AIX users who have
  configured LB to run in promisicuous mode and Linux users on
  IBM Z running in layer 3 mode should alias cluster addresses
  to the ethernet interfaces.
  
  To eliminate the problem, the Load Balancer will forward
  existing traffic when running in the backup state. The backup
  LB will not forward new connections. It will not answer ARP
  requests for the cluster either. Any new connections and ARP
  requests are passed to the operating system for processing.
  
  Fix levels:
     8.5.5.23
     9.0.5.15
  

Temporary fix

Comments

APAR Information

• APAR is sysrouted FROM one or more of the following:

• APAR is sysrouted TO one or more of the following:

Fix information

• Fixed component name

  WS EDGE LB IPV4

• Fixed component ID

  5724H8812

Applicable component levels