[
https://issues.apache.org/jira/browse/DRILL-5657?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16129847#comment-16129847
]
ASF GitHub Bot commented on DRILL-5657:
---------------------------------------
Github user paul-rogers commented on a diff in the pull request:
https://github.com/apache/drill/pull/866#discussion_r133618354
--- Diff:
exec/java-exec/src/main/java/org/apache/drill/exec/physical/rowSet/impl/ResultVectorCache.java
---
@@ -0,0 +1,181 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+package org.apache.drill.exec.physical.rowSet.impl;
+
+import java.util.ArrayList;
+import java.util.HashMap;
+import java.util.List;
+import java.util.Map;
+
+import org.apache.drill.common.types.TypeProtos.MajorType;
+import org.apache.drill.exec.expr.TypeHelper;
+import org.apache.drill.exec.memory.BufferAllocator;
+import org.apache.drill.exec.record.MaterializedField;
+import org.apache.drill.exec.vector.ValueVector;
+
+/**
+ * Manages an inventory of value vectors used across row batch readers.
+ * Drill semantics for batches is complex. Each operator logically returns
+ * a batch of records on each call of the Drill Volcano iterator protocol
+ * <tt>next()</tt> operation. However, the batches "returned" are not
+ * separate objects. Instead, Drill enforces the following semantics:
+ * <ul>
+ * <li>If a <tt>next()</tt> call returns <tt>OK</tt> then the set of
vectors
+ * in the "returned" batch must be identical to those in the prior batch.
Not
+ * just the same type; they must be the same <tt>ValueVector</tt> objects.
+ * (The buffers within the vectors will be different.)</li>
+ * <li>If the set of vectors changes in any way (add a vector, remove a
+ * vector, change the type of a vector), then the <tt>next()</tt> call
+ * <b>must</b> return <tt>OK_NEW_SCHEMA</tt>.</ul>
+ * </ul>
+ * These rules create interesting constraints for the scan operator.
+ * Conceptually, each batch is distinct. But, it must share vectors. The
+ * {@link ResultSetLoader} class handles this by managing the set of
vectors
+ * used by a single reader.
+ * <p>
+ * Readers are independent: each may read a distinct schema (as in JSON.)
+ * Yet, the Drill protocol requires minimizing spurious
<tt>OK_NEW_SCHEMA</tt>
+ * events. As a result, two readers run by the same scan operator must
+ * share the same set of vectors, despite the fact that they may have
+ * different schemas and thus different <tt>ResultSetLoader</tt>s.
+ * <p>
+ * The purpose of this inventory is to persist vectors across readers, even
+ * when, say, reader B does not use a vector that reader A created.
+ * <p>
+ * The semantics supported by this class include:
+ * <ul>
+ * <li>Ability to "pre-declare" columns based on columns that appear in
+ * an explicit select list. This ensures that the columns are known (but
+ * not their types).</li>
+ * <li>Ability to reuse a vector across readers if the column retains the
same
+ * name and type (minor type and mode.)</li>
+ * <li>Ability to flush unused vectors for readers with changing schemas
+ * if a schema change occurs.</li>
+ * <li>Support schema "hysteresis"; that is, the a "sticky" schema that
+ * minimizes spurious changes. Once a vector is declared, it can be
included
+ * in all subsequent batches (provided the column is nullable or an
array.)</li>
+ * </ul>
+ */
+public class ResultVectorCache {
+
+ /**
+ * State of a projected vector. At first all we have is a name.
+ * Later, we'll discover the type.
+ */
+
+ private static class VectorState {
+ protected final String name;
+ protected ValueVector vector;
+ protected boolean touched;
+
+ public VectorState(String name) {
+ this.name = name;
+ }
+
+ public boolean satisfies(MaterializedField colSchema) {
+ if (vector == null) {
+ return false;
+ }
+ MaterializedField vectorSchema = vector.getField();
+ return vectorSchema.getType().equals(colSchema.getType());
+ }
+ }
+
+ private final BufferAllocator allocator;
+ private final Map<String, VectorState> vectors = new HashMap<>();
+
+ public ResultVectorCache(BufferAllocator allocator) {
+ this.allocator = allocator;
+ }
+
+ public void predefine(List<String> selected) {
+ for (String colName : selected) {
+ addVector(colName);
+ }
+ }
+
+ private VectorState addVector(String colName) {
+ VectorState vs = new VectorState(colName);
+ vectors.put(vs.name, vs);
+ return vs;
+ }
+
+ public void newBatch() {
+ for (VectorState vs : vectors.values()) {
+ vs.touched = false;
+ }
+ }
+
+ public void trimUnused() {
--- End diff --
Was one attempt to handle changing schemas by removing vectors that where
created in one batch, but then not used in a later batch. The original design
of some of Drill's code enforced the idea that required vectors are required.
So, if a required vector is added, but not data written to it, what to do? The
thought here was to remove the vector.
Later, it became clear that the only workable solution is to support
"fill-empties" even for required vectors: we must fill unused vectors with
zeros. It then becomes up to the client to determine if this makes sense; the
underlying framework no longer has an opinion.
> Implement size-aware result set loader
> --------------------------------------
>
> Key: DRILL-5657
> URL: https://issues.apache.org/jira/browse/DRILL-5657
> Project: Apache Drill
> Issue Type: Improvement
> Affects Versions: Future
> Reporter: Paul Rogers
> Assignee: Paul Rogers
> Fix For: Future
>
>
> A recent extension to Drill's set of test tools created a "row set"
> abstraction to allow us to create, and verify, record batches with very few
> lines of code. Part of this work involved creating a set of "column
> accessors" in the vector subsystem. Column readers provide a uniform API to
> obtain data from columns (vectors), while column writers provide a uniform
> writing interface.
> DRILL-5211 discusses a set of changes to limit value vectors to 16 MB in size
> (to avoid memory fragmentation due to Drill's two memory allocators.) The
> column accessors have proven to be so useful that they will be the basis for
> the new, size-aware writers used by Drill's record readers.
> A step in that direction is to retrofit the column writers to use the
> size-aware {{setScalar()}} and {{setArray()}} methods introduced in
> DRILL-5517.
> Since the test framework row set classes are (at present) the only consumer
> of the accessors, those classes must also be updated with the changes.
> This then allows us to add a new "row mutator" class that handles size-aware
> vector writing, including the case in which a vector fills in the middle of a
> row.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
