[infinispan-issues] [JBoss JIRA] (ISPN-4438) Entry is not properly unmarshalled in compatibility mode when L1 enabled

[ https://issues.jboss.org/browse/ISPN-4438?page=com.atlassian.jira.plugin.... ] Martin Gencur edited comment on ISPN-4438 at 6/26/14 3:27 AM: -------------------------------------------------------------- The root cause is following: When a HotRod client first reads a key from a node which is not the owner, the value is retrieved from another node. As part of this process, the value is marhalled on the owner node and sent back to the requester where it is stored in L1 as it is - marshalled. Then the entry is returned back to the client which unmarshalls the entry and has gets the result. When the key is requested second time the marshalled entry is taken from L1 cache but is marshalled AGAIN and returned to the client. The client unmarshalls it but gets another marshalled entry as a result because it was marshalled twice. This only happens when data is requested by HotRod client. It does not happen when requested by a local thread from an embedded cache or via REST. L1 cache does not make much sense when using HotRod client because the client knows which node is the owner and requests data from that node directly (no use of L1) so this is not a bug. was (Author: mgencur): The root cause is following: When a HotRod client first reads a key from a node which is not the owner, the value is retrieved from another node. As part of this process, the value is marhalled on the owner node and sent back to the requester where it is stored in L1 as it is - marshalled. Then a copy of the entry is unmarshalled and returned back to the client. When the key is requested second time the marshalled entry is taken from L1 cache but is marshalled AGAIN and returned to the client. This only happens when data is requested by HotRod client. It does not happen when requested by a local thread from an embedded cache or via REST. L1 cache does not make much sense when using HotRod client because the client knows which node is the owner and requests data from that node directly (no use of L1) so this is not a bug. -- This message was sent by Atlassian JIRA (v6.2.6#6264)