|
|
NOTE: we cant use Adabas Clustering in a production environments.
The issue is from the Natural client side when it connects to Adabas.
In Natparm the DB connection points to three nodes, A B and C.
When node A gets abruptly stopped the Adabas client will still attempt to connect to node A first as this is first in the list (alphabetical).
Because it has been killed off, there is nothing to tell the client this so will attempt to connect to it. DNS will still resolve A's IP address and so will attempt to connect to the Adabas TCP port. It times out after 30 seconds. At 20 seconds (this is a very sensible web timeout) the client disconnects and the NHA user just get a gateway timeout.
Improvement idea to fix the BUG:
Adabas client (Natural) will always try and connect to node A then B then C always no matter. The client should be cluster aware and try going to the last know, primary first instead of A all the time with will improve performance if the primary is not A.
introduce a connection timeout in the client so that is fails faster. At the moment we get a complete outage for all users if node A is down.
