[infinispan-issues] [JBoss JIRA] (ISPN-8244) ejb client-mappings cache is missing entries

[ https://issues.jboss.org/browse/ISPN-8244?page=com.atlassian.jira.plugin.... ] Janine Eichler updated ISPN-8244: --------------------------------- Description: In JBoss EAP 7.0 the client-mappings is missing some entries which can result in incomplete cluster topologies sent to the client. [~rachmato] already analysed (and fixed, at least locally) the problem: {quote}A little more detail. In the EAP config, there is a cache container called "ejb" and a defaultt cache called "dist" which is used to store SFSB sessions when they are created and make them available across the cluster. This cache has, by default, cache mode DIST. The client-mappings cache, which stores node -> IP:port information for the cluster nodes, is created internally but is based on the configuration of the "ejb"/"dist" cache but is set to cache mode REPL. There seems to be a problem with the configuration of the internally-created client-mappings cache which is causing the problems (in particular, the consistent hash for the Infinispan cache is *possibly* not being set correctly).{quote} {quote} [...] the CHF setting is the problem. It should be either set to null or ReplicatedConsistentHashFactory. What was happening was that using DefaultConsistentHashFactory (aimed for DIST caches) with a REPL cache only writes entries to numOwners nodes, but the reader thinks that the entries will be available at all nodes, resulting in null values. {quote} was: In JBoss EAP 7.0 the client-mappings is missing some entries which can result in incomplete cluster topologies sent to the client. [~rachmato] already analysed (and fixed, at least locally) the problem: {code} A little more detail. In the EAP config, there is a cache container called "ejb" and a defaultt cache called "dist" which is used to store SFSB sessions when they are created and make them available across the cluster. This cache has, by default, cache mode DIST. The client-mappings cache, which stores node -> IP:port information for the cluster nodes, is created internally but is based on the configuration of the "ejb"/"dist" cache but is set to cache mode REPL. There seems to be a problem with the configuration of the internally-created client-mappings cache which is causing the problems (in particular, the consistent hash for the Infinispan cache is *possibly* not being set correctly). {code} {code} [...] the CHF setting is the problem. It should be either set to null or ReplicatedConsistentHashFactory. What was happening was that using DefaultConsistentHashFactory (aimed for DIST caches) with a REPL cache only writes entries to numOwners nodes, but the reader thinks that the entries will be available at all nodes, resulting in null values. {code} -- This message was sent by Atlassian JIRA (v7.2.3#72005)