PH22784: IBM WEBSPHERE APPLICATION SERVER LOAD BALANCER FOR IPV4 AND IPV6 USERS

APAR status

• Closed as program error.

Error description

• A crash can occur when running the load balancer on AIX
  

Local fix

Problem summary

• ****************************************************************
  * USERS AFFECTED:  IBM WebSphere Application Server Load       *
  *                  Balancer for IPv4 and IPv6 users            *
  ****************************************************************
  * PROBLEM DESCRIPTION: This issue can only occur on the Load   *
  *                      Balaner for IPv4 and IPv6 on AIX. A     *
  *                      hang or crash occurs when LB is         *
  *                      sending an ARP packet.                  *
  ****************************************************************
  * RECOMMENDATION:                                              *
  ****************************************************************
  A crash can occur with a stack similar to :
  e_block_thread+000280 ()
  soesend+00069C
  sosend+000014
  .kernel_add_gate_cstack+000030 ()
  os_packetxmit+000E1C
  lbarp_send+00099C
  _lbneigh_reply+0008C4
  lbneigh_pbuf_entry+0003CC
  lbarp_entry+000B44
  aix_recv+000234
  Offsets may differ depending on code levels.  This crash is
  more likely with multiple interfaces.
  The crash occur when LB is sending an ARP packet and an
  interrupt occurs, most likely from a packet received off a
  different interface.  The interrupt processing triggers a
  context switch within the operating system kernel and the
  transmit API LB was using is not interrupt safe.
  

Problem conclusion

• LB code was modified to change the API used to send ARP
  packets. The previous API (sosend) was not thread safe. Code
  was modified to use the interface output function which is
  thread safe.
  
  Fix packs containing fix:
     8.5.5.18
     9.0.5.4
  

Temporary fix

• Minimize ARP as much as possible to avoid the problem. Use
  static ARP entries for HA partner, reach targets, MAC-based
  servers, etc. Gratuitous ARP on takeovers could still trigger
  issue even with static ARP entries.
  

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