I don't know if this is still an issue or not in the 6.x branch, but this is occurring when using the Infinispan 5.2.6 provided with EAP6.1.1 (Yes, yes, I know you should not use the provided infinispan, but it's the easiest way to get it clustered on openshift.)
This is my configuration:
<cache-container name="redoculous" aliases="ha-partition" default-cache="default" jndi-name="java:jboss/infinispan/redoculous-cluster"> <transport/> <replicated-cache name="filesystem.metadata" mode="SYNC" start="LAZY"> <locking isolation="REPEATABLE_READ"/> <eviction strategy="LIRS" max-entries="1000"/> <file-store path="redoculous" preload="true" passivation="true"/> </replicated-cache> <distributed-cache name="filesystem.content" owners="2" mode="SYNC" start="LAZY"> <locking isolation="REPEATABLE_READ"/> <eviction strategy="LIRS" max-entries="10000"/> <file-store path="redoculous" preload="true" passivation="true"/> </distributed-cache> </cache-container>
This is revealed particularly when pushing zip files to the grid. After retrieving them, upon attempt to decompress the file, ZipEntry will typically throw Exceptions about invalid entry header information, and various other sporadic issues.
I believe this is an eviction issue because typically the first few files work, but things begin to fail more and more, the more files are placed into the grid filesystem.
Hi Lincoln,
there where lots of issues regarding data corruption when using CacheStores fixed during 6.0 development, I'm pretty sure it's fixed, but to narrow down the area:
could you try the same test without any CacheStore ? (Assuming you don't need it for this specific test)
If the issue disappears you're likely hitting one of the many issues about CacheStore inconsistency and it's then likely already fixed in WildFly (Infinispan 6.0),
otherwise you might be hitting a different problem.