I have few tables with big amount of data (about 100 million records). So I can't store this data in memory but I would like to stream this result set using java.util.stream
class and pass this stream to another class. I read about Stream.of
and Stream.Builder
operators but they are buffered streams in memory. So is there any way to resolve this question? Thanks in advance.
UPDATE #1
Okay I googled and found jooq library. I'm not sure but looks like it could be applicable to my test case. To summarize I have few tables with big amount of data. I would like to stream my resultset and transfer this stream to another method. Something like this:
// why return Stream<String>? Because my result set has String type private Stream<Record> writeTableToStream(DataSource dataSource, String table) { Stream<Record> record = null; try (Connection connection = dataSource.getConnection()) { String sql = "select * from " + table; try (PreparedStatement pSt = connection.prepareStatement(sql)) { connection.setAutoCommit(false); pSt.setFetchSize(5000); ResultSet resultSet = pSt.executeQuery(); // record = DSL.using(connection) .fetch(resultSet).stream(); } } catch (SQLException sqlEx) { logger.error(sqlEx); } return record; }
Could please someone advise, am I on correct way? Thanks.
UPDATE #2
I made some experiment on jooq and could say now that above decision is not suitable for me. This code record = DSL.using(connection).fetch(resultSet).stream();
takes too much time
The first thing you have to understand is that code like
try (Connection connection = dataSource.getConnection()) { … try (PreparedStatement pSt = connection.prepareStatement(sql)) { … return stream; } }
does not work as by the time you leave the try
blocks, the resources are closed while the processing of the Stream
hasn’t even started.
The resource management construct “try with resources” works for resources used within a block scope inside a method but you are creating a factory method returning a resource. Therefore you have to ensure that the closing of the returned stream will close the resources and the caller is responsible for closing the Stream
.
Further, you need a function which produces an item out of a single line from the ResultSet
. Supposing, you have a method like
Record createRecord(ResultSet rs) { … }
you may create a Stream<Record>
basically like
Stream<Record> stream = StreamSupport.stream(new Spliterators.AbstractSpliterator<Record>( Long.MAX_VALUE,Spliterator.ORDERED) { @Override public boolean tryAdvance(Consumer<? super Record> action) { if(!resultSet.next()) return false; action.accept(createRecord(resultSet)); return true; } }, false);
But to do it correctly you have to incorporate the exception handling and closing of resources. You can use Stream.onClose
to register an action that will be performed when the Stream
gets closed, but it has to be a Runnable
which can not throw checked exceptions. Similarly the tryAdvance
method is not allowed to throw checked exceptions. And since we can’t simply nest try(…)
blocks here, the program logic of suppression exceptions thrown in close
, when there is already a pending exception, doesn’t come for free.
To help us here, we introduce a new type which can wrap closing operations which may throw checked exceptions and deliver them wrapped in an unchecked exception. By implementing AutoCloseable
itself, it can utilize the try(…)
construct to chain close operations safely:
interface UncheckedCloseable extends Runnable, AutoCloseable { default void run() { try { close(); } catch(Exception ex) { throw new RuntimeException(ex); } } static UncheckedCloseable wrap(AutoCloseable c) { return c::close; } default UncheckedCloseable nest(AutoCloseable c) { return ()->{ try(UncheckedCloseable c1=this) { c.close(); } }; } }
With this, the entire operation becomes:
private Stream<Record> tableAsStream(DataSource dataSource, String table) throws SQLException { UncheckedCloseable close=null; try { Connection connection = dataSource.getConnection(); close=UncheckedCloseable.wrap(connection); String sql = "select * from " + table; PreparedStatement pSt = connection.prepareStatement(sql); close=close.nest(pSt); connection.setAutoCommit(false); pSt.setFetchSize(5000); ResultSet resultSet = pSt.executeQuery(); close=close.nest(resultSet); return StreamSupport.stream(new Spliterators.AbstractSpliterator<Record>( Long.MAX_VALUE,Spliterator.ORDERED) { @Override public boolean tryAdvance(Consumer<? super Record> action) { try { if(!resultSet.next()) return false; action.accept(createRecord(resultSet)); return true; } catch(SQLException ex) { throw new RuntimeException(ex); } } }, false).onClose(close); } catch(SQLException sqlEx) { if(close!=null) try { close.close(); } catch(Exception ex) { sqlEx.addSuppressed(ex); } throw sqlEx; } }
This method wraps the necessary close operation for all resources, Connection
, Statement
and ResultSet
within one instance of the utility class described above. If an exception happens during the initialization, the close operation is performed immediately and the exception is delivered to the caller. If the stream construction succeeds, the close operation is registered via onClose
.
Therefore the caller has to ensure proper closing like
try(Stream<Record> s=tableAsStream(dataSource, table)) { // stream operation }
Note that also the delivery of an SQLException
via RuntimeException
has been added to the tryAdvance
method. Therefore you may now add throws SQLException
to the createRecord
method without problems.
I'm going to answer the jOOQ part of your question. As of jOOQ 3.8, there have now been quite a few additional features related to combining jOOQ with Stream. Other usages are also documented on this jOOQ page.
You tried this:
Stream<Record> stream = DSL.using(connection).fetch(resultSet).stream();
Indeed, this doesn't work well for large result sets because fetch(ResultSet)
fetches the entire result set into memory and then calls Collection.stream()
on it.
Instead, you could write this:
try (Stream<Record> stream = DSL.using(connection).fetchStream(resultSet)) { ... }
... which is essentially convenience for this:
try (Cursor<Record> cursor = DSL.using(connection).fetchLazy(resultSet)) { Stream<Record> stream = cursor.stream(); ... }
See also DSLContext.fetchStream(ResultSet)
Of course, you could also let jOOQ execute your SQL string, rather than wrestling with JDBC:
try (Stream<Record> stream = DSL.using(dataSource) .resultQuery("select * from {0}", DSL.name(table)) // Prevent SQL injection .fetchSize(5000) .fetchStream()) { ... }
SELECT *
As was criticised in the comments, their jOOQ usage seemed slow because of how jOOQ eagerly fetches LOB data into memory despite using fetchLazy()
. The word "lazy" corresponds to fetching records lazily (one by one), not fetching column data lazily. A record is completely fetched in one go, assuming you actually want to project the entire row.
If you don't need some heavy rows, don't project them! SELECT *
is almost always a bad idea in SQL. Drawbacks:
More info in this blog post here.
Do note that a Stream
produced by jOOQ is "resourceful", i.e. it contains a reference to an open ResultSet
(and PreparedStatement
). So, if you really want to return that stream outside of your method, make sure it is closed properly!
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With