Remove Inner SQL Connection and Session Pool Caching

[hnwyllmm]

Task Description

Solution Description

Passed Regressions

Upgrade Compatibility

Other Information

Release Note

[hnwyllmm]

The mapping Dima issue is about removing ObInnerSQLConnectionPool and using it only as needed.

[footka]

src/sql/hybrid_search/ob_hybrid_search_executor.cpp:142

Why is reference counting still needed?

[footka]

src/observer/ob_inner_sql_connection.cpp:1704

What's special about this release, and why is it called release_inner_sql_connection_for_proxy? What's its relationship with the proxy?

[hnwyllmm]

src/observer/ob_inner_sql_connection.cpp:1704

This provides a Proxy for Oblib to resolve the WEAK_SYMBOL function dependencies from upper layers. There's no special cleanup logic, just an unref, used to isolate the dependencies between oblib and observer.

[hnwyllmm]

src/sql/hybrid_search/ob_hybrid_search_executor.cpp:142

Reference counting is unrelated to the connection pool; it ensures that the Conn is not released prematurely while a ReadResult is using the Session inside that Conn. This is the same function originally used for release—only the call location has been moved.

[hnwyllmm]

src/observer/ob_inner_sql_connection.cpp:1704

The smart pointer release has been unified, and release_inner_sql_connection_for_proxy is no longer needed and has been removed.

[hnwyllmm]

src/sql/hybrid_search/ob_hybrid_search_executor.cpp:142

Changed to use ObSharedGuard smart pointer for allocation and deallocation.

[footka]

deps/oblib/src/common/mysqlclient/ob_single_connection_proxy.h:66

Using a guard in a local function scope is relatively safe, but using it as a class member can be dangerous.