Physical plan cloning could lead to out of order connections
------------------------------------------------------------
Key: PIG-520
URL: https://issues.apache.org/jira/browse/PIG-520
Project: Pig
Issue Type: Bug
Affects Versions: types_branch
Reporter: Santhosh Srinivasan
Fix For: types_branch
In the PhysicalPlan clone method, the algorithm used is as follows:
1. Create an empty plan
2. For all the operators in the plan,
a. clone the operator
b. add it to the plan
3. For all the keys (from_node) in the map mFromEdges
a. For all the values (to_node) for this key
i. Connect the from_node to the to_node in the plan
There are no guarantees on the order in which the from_nodes in the mFromEdges
are processed, we could get out of order connections in the graph.
Example:
If we have UDF with two arguments like myUDF(a, b) in a plan, the order in
which the nodes are processed will determine the cloned plan. We could end up
with
myUDF(a, b)
OR
myUDF(b,. a)
depending on the order in which a and b appear in the mFromEdges look up table.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
