Change getStreams to use multiple clients#19
Change getStreams to use multiple clients#19vfraga wants to merge 19 commits intoflight-jdbc-driverfrom
Conversation
There was a problem hiding this comment.
Nit: let's avoid default super calls.
There was a problem hiding this comment.
I am not sure about closing the parent allocator here... I don't see a call to close() being removed which means prior to this patch we were either leaking an allocator or we are double-closing this.
There was a problem hiding this comment.
The casting of the factory to call a close method on it is weird.
Instead, you can make this class own the allocator:
- Have the allocator be the parameter to the constructor.
- call super(new KeyedFlightSql...()) and have that take in the allocator
- override close() in this class, and move all the allocator cleanup there.
There was a problem hiding this comment.
I think you should actually call super.close() first to clean up the clients which are depending on the allocator, then clean up the allocator. (Normally the right thing to do is do child class clean-up, then base class clean-up but I think the logic needs to be different in this scenario).
There was a problem hiding this comment.
Don't use String.format() with loggers. This executes the work to generate a string, even if logging is off.
Use the logger directly. Note that placeholders are curly braces in SLF4J eg:
logger.info("Getting client for location {}.", location);
please fix this throughout the PR.
There was a problem hiding this comment.
I would think this happens automatically rather than needing try/catch.
There was a problem hiding this comment.
We cannot return the client to the pool until we are done getting the stream. Otherwise it allows another query being executed to re-use the client while it's still being used.
There was a problem hiding this comment.
Maybe we can write an auto-closeable wrapper on top of FlightStream that returns to the pool on closure.
vfraga
force-pushed
the
flight-jdbc-driver
branch
from
532d91f to
59148df
Compare
* Remove scope from 'hamcrest' dependency on java/pom.xml * Use flight top-level module on parent pom.xml instead of declaring each one * Avoid using getStatement inside StatementContext methods * Make StatementContext.getQuery() return String * Minor fixes on pom.xml * Move 'os-maven-plugin' to parent pom.xml * Update protobuf generation on pom.xml files * Use ClassLoader#getResource to get network.properties on TestFlightSql * Bind to any ephemeral port on TestFlightSql * Move JDBC-Arrow type default conversion from JdbcToArrowConfig to JdbcToArrowUtils * Micro-optimization: initialize ArrayList with the right size * Fix null-check on PreparedStatement#setParameters * Avoid wrapping vector into a ImmutableList and then into an ArrayList on FlightSqlExample#getTablesRoot * Remove null-check on VectorSchemaRoot on FlightSqlClient#setParameters() * Remove the need of separate cache for ResultSets * Add missing 'final' modifiers
…les and query execution methods. (apache#226) This add an auxiliary class FlightSqlColumnMetadata meant to read and write known metadata for Arrow schema fields, such as CATALOG_NAME, SCHEMA_NAME, TABLE_NAME, PRECISION, SCALE, IS_AUTO_INCREMENT, IS_CASE_SENSITIVE, IS_READ_ONLY and IS_SEARCHABLE.
vfraga
force-pushed
the
change-get-streams
branch
from
8db5a48 to
a8d1d53
Compare
5 participants
This is currently just a draft but it tries to address the issue of only using a single client to grab multiple streams.
It uses Apache Pools to pool the clients keyed to their location correspondingly.