Yury, thanks for the example I have initially investigated the issue filed a ticket - https://issues.apache.org/jira/browse/IGNITE-1811
Dmitry, we need to send request to primary if: 1. rebalancing is in process 2. user might have called local clear for the key 3. to properly load from store (if store is enabled) Now situation is a bit worse and there are a lot of calls that can be avoided. I will take a look and will try to push it to 1.5 --Yakov 2015-10-29 8:58 GMT+03:00 Dmitriy Setrakyan <[email protected]>: > > > On Wed, Oct 28, 2015 at 3:50 AM, Alexey Goncharuk < > [email protected]> wrote: > >> Even if swap and store is disabled, a miss on a backup node will result >> in a network trip to primary node because an entry could have been evicted >> from the cache. >> > > Alexey, we are talking about a REPLICATED cache. I do not understand why > do we ever need to fetch the data from remote nodes on a cache miss, if > evictions and expirations are disabled, like described here? It should be a > local lookup only in this case, no? > > >> >>>>> -- >>>>> View this message in context: >>>>> http://apache-ignite-users.70518.x6.nabble.com/Performance-issue-with-REPLICATED-cache-tp1710p1730.html >>>>> Sent from the Apache Ignite Users mailing list archive at Nabble.com. >>>>> >>>> >>>> >>> >> >
