Scott Kurz (JIRA) wrote:
binding-ws-axis2 should use Axis2 ServiceClient in thread-safe manner
---------------------------------------------------------------------
Key: TUSCANY-3643
URL: https://issues.apache.org/jira/browse/TUSCANY-3643
Project: Tuscany
Issue Type: Bug
Affects Versions: Java-SCA-2.0-M5, Java-SCA-1.6
Reporter: Scott Kurz
Priority: Minor
In running a version of our 1.x code (the exact version doesn't so much matter
since the problem has existed for awhile), against the Axis2 trunk running
multiple threads at once, I hit the problem:
java.lang.NullPointerException
at
org.apache.axis2.client.ServiceClient.cleanupTransport(ServiceClient.java:860)
at
org.apache.axis2.client.ServiceClient.createClient(ServiceClient.java:665)
at
org.apache.tuscany.sca.binding.ws.axis2.Axis2BindingInvoker.createOperationClient(..)
at
org.apache.tuscany.sca.binding.ws.axis2.Axis2BindingInvoker.invokeTarget(...)
at
org.apache.tuscany.sca.binding.ws.axis2.Axis2BindingInvoker.invoke(....)
It seems Axis2 has introduced a new "autoOperationCleanup" function.
So one solution would be to disable this. However, it seems from discussions
such as this:
http://www.mail-archive.com/[email protected]/msg48406.html
that a better solution would be to stop using a single Axis2 ServiceClient
across threads, since apparently these are not intended to be thread-safe.
So it seems we should switch from using an instance variable to a local
variable to hold these refs.
Creating a new instance of ServiceClient for every request could add
a fair bit of overhead to the request path in the simple common case
where only a single thread is being used.
Another approach would be to use a pooling/caching approach where the
existing ServiceClient could be reused if there's no contention between
threads, and a new ServiceClient could be created if there is contention.
There's some code like this near the end of RuntimeWireImpl (look for
wireReserved and cachedWire).
Simon
