Github user jinfengni commented on the issue:
https://github.com/apache/drill/pull/701
+1
I have one naive question. Before we dive into MVCC, or try to improve the
current ZK-based approach, have we consider adding a command "refresh function
registry", so that users enforce that function registry across the entire
cluster is consistent? I remember someone (probably @amansinha100 ) once
brought up this idea.
1. How often would user run into the issue caused by inconsistency of
functions in different drillbit? Sounds like we are adding non-negligible
overhead for other queries which do not use dynamic UDF or overloaded
functions, just for the sake of resolving issues of function inconsistency.
2. Drill's function resolution logic not only happens in planning but also
execution time (due to the fact schema may be known in execution time). This
means the function registry could be checked multiple times during the query
lifetime. Adding overhead to each check would slow down the entire query.
3. With the "refresh function registry" command, use has a way to ensure
function registry is consistent after issuing such command. We could tell user
that drillbit could run into various of problems, before run "refresh function
registry", therefore it's always recommend to run such command after user
creates a new UDF.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
