We are having a cluster of 9 nodes. All connected fine in the same network.
There are 5 clients and 4 server nodes.
All the client nodes and 2 of 4 server nodes are in the same data center.
The other two server nodes are in different date center (but in the same
network). There can be some network latency between two data centres.
All the 4 server nodes has full_sync cache operations being executed. The
mode is replication for all caches.
We are finding delays in executions and some warning about some count down
latch on operations not being quick.
Can someone suggest a way how we can try to make it faster without changing
cache config? We need to connect the 2 servers in other DC for disaster
recovery purpose. Can we make the replication happen in only few servers
with fullsync and for other servers in a delayed manner? The clients will
only listen to same DC servers.
I guess that some of your operations were slow because of load balancer
work. Two ways of affinity calculation exist:
1)If readFromBackup = false then the node with primary partition will be
chosen. Primary can be on different nodes.
2)If readFromBackup = true then the random node will be chosen because of a
Ignite doesn't have data center awareness functionality.
As a WA you can use computes that will be started on some required node.
This compute can use localPeek mode for cache operations: