[
https://issues.apache.org/jira/browse/HBASE-16264?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15398495#comment-15398495
]
stack commented on HBASE-16264:
-------------------------------
See [~ghelmling] comment up in the parent issue:
https://issues.apache.org/jira/browse/HBASE-15638?focusedCommentId=15386820&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-15386820
Let me repeat it here:
{code}
For endpoints, the user will have:
# Created a .proto file with the service definition for their endpoint
# Run their own version of protoc to do the code gen for the service
# The generated PB messages for request/response, and, as you note, the
RpcController and RpcCallback in the service method signatures will be in the
com.google.protobuf package
# The endpoint jar is loaded in HBase
# The ServerRpcController we pass through for the endpoint invocation will
extend our shaded RpcController
So the only way I see this working is if we force endpoint implementors to run
a preprocessor step to shade the endpoint jar prior to step #4. Maybe that
would work, but it seems pretty cumbersome. Or else we somehow avoid shading
ServerRpcController.
{code}
Chatting w/ [~mbertozzi] on this issue, it is a painful one complicated by the
fact that we'll have a pb2.5 jar on our CLASSPATH (because of HDFS and in
particular because new AsyncWAL needs to pass HDFS HDFS-compat PBs). This pb2.5
jar occupies the com.google.protobuf slot on serverside at least. Non-shaded
refs to c.g.protobug will be 2.5pb until HDFS does shading.
On client side, the problem is in our published public API in Table. It'd be
ugly if these public APIs of a sudden were shaded, or at least, not w/o a means
of transitioning current EPCPs (Endpoint Coprocessors) -- Gary's preprocessor
step. Chatting w/ Matteo, let me try and leave the published APIs as unshaded
and similar on Endpoint CP side. Let me see how far I get w/ this. Will
probably crash and burn going from non-shaded to shaded but let me see. Could
add new API to Table that wasn't PB-based for going forward. We'd probably need
this anyways given c.g.pb is occupied by 2.5pb.
Fun, fun, fun!
> Figure how to deal with endpoints and shaded pb
> -----------------------------------------------
>
> Key: HBASE-16264
> URL: https://issues.apache.org/jira/browse/HBASE-16264
> Project: HBase
> Issue Type: Sub-task
> Components: Coprocessors, Protobufs
> Reporter: stack
> Assignee: stack
>
> Come up w/ a migration plan for coprocessor endpoints when our pb is shaded.
> Would be sweet if could make it so all just worked. At worst, come up w/ a
> prescription for how to migrate existing CPs.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
