zhangbutao created HIVE-27440:
---------------------------------
Summary: Improve data connector cache
Key: HIVE-27440
URL: https://issues.apache.org/jira/browse/HIVE-27440
Project: Hive
Issue Type: Sub-task
Reporter: zhangbutao
_*DataConnectorProviderFactory*_ uses HashMap to cache data connector
instances, and there is no way to invalidate the cache unless you restart the
MetaStore.
What is more serious is that if you drop or alter the dataconnector, the cache
will not change, and you maybe use a invalid dataconnector next time.
I think we can improve the dataconnector cache from the two aspects:
* Use Caffeine with a *maxmumsize* e.g. 100 to cache data connector instead
of HashMap, and set a *expire time* after the last accessing. And we also
should close the underlying datasource connection using {*}Caffeine
RemovalListener{*}.
* After executing Drop or Alter DDL on a dataConnector, we should *update
cache* to clean the dataConnector to avoid using the invalid dataConnector next
time.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
