Table of Contents |
---|
Status
Current state: Under Discussion Accepted
Discussion thread: here [Change the link from the KIP proposal email archive to your own email thread]
Voting thread: https://lists.apache.org/thread/xxyb5yyqrsdxsyxxbjhvnlxw5fl8xd0c
JIRA:
Jira | ||||||
---|---|---|---|---|---|---|
|
Please keep the discussion on the mailing list rather than commenting on the wiki (wiki discussions get unwieldy fast).
Motivation
Provide an implementation of the Query interface, introduced in KIP-796: Interactive Query v2 , to support reverseRange and reverseAll
Proposed Changes
The concepts of reverseRange
and reverseAll
are not tied to any specific method or class. Rather, they represent functionalities we wish to achieve. Currently, with RangeQuery
, we can use methods like withRange()
, withLowerBound()
, withUpperBound()
, and withNoBounds()
.
Utilizing these, the query results are ordered based on the serialized byte[] of the keys, not the 'logical' key order.
Take IQv2
StoreIntegrationTest
as an example: we have two partitions with four key-value pairs:
<0,0>
in Partition0<1,1>
in Partition1<2,2>
in Partition0<3,3>
in Partition1
When we use RangeQuery.withRange(1,3)
, the returned result is:
- Partition0:
[2]
- Partition1:
[1, 3]
To achieve the functionalities of reverseRange
and reverseAll
, we can introduce a method named withDescendingKeys()
for reversed queries. For example, by using RangeQuery.withRange(1,3).withDescendingKeys()
, the expected result would be:
- Partition0:
[2]
- Partition1:
[3, 1]
This means the results are in the reverse order of their keys.
To ensure that we can achieve this functionality, the keys in both RocksDB
and InMemoryKeyValueStore
should be sorted. We know that RocksDB
keys are inherently sorted. After investigation, we found that InMemoryKeyValueStore
uses a TreeMap
, implying its keys are also sorted. Therefore, performing the aforementioned queries is feasible.
Proposed Changes
According to KIP-968, this KIP introduces the public enum ResultOrder to determine whether keys are sorted in ascending or descending or unordered order. Order is based on the serialized byte[] of the keys, not the 'logical' key order. employs the withDescendingKeys() and withAscendingKeys()
methods to specify that the keys should be ordered in descending or ascending or unordered sequence, and the resultOrder() method to retrieve the value of enum value in ResultOrder. I've incorporated these variables and methods into the RangeQuery
class and modified some method inputs. As a result, we can now use withDescendingKeys()
to obtain results in reverse order and use withAscendingKeys to obtain the result in ascending order.
Code Block |
---|
/**
* Interactive query for issuing range queries and scans over KeyValue stores.
* <p>
* A range query retrieves a set of records, specified using an upper and/or lower bound on the keys.
* <p>
* A scan query retrieves all records contained in the store.
* <p>
*/
@Evolving
public final class RangeQuery<K, V> implements Query<KeyValueIterator<K, V>> {
...
/**
* Determines if the serialized byte[] of the keys in ascending or descending or unordered order.
* Order is based on the serialized byte[] of the keys, not the 'logical' key order.
* @return return the order of returned records based on the serialized byte[] of the keys (can be unordered, or in ascending or in descending order).
*/
public ResultOrder resultOrder()
/**
* Set the query to return the serialized byte[] of the keys in descending order.
* Order is based on the serialized byte[] of the keys, not the 'logical' key order.
* @return a new RangeQuery instance with descending flag set.
*/
public RangeQuery<K, V> withDescendingKeys()
/**
* Set the query to return the serialized byte[] of the keys in ascending order.
* Order is based on the serialized byte[] of the keys, not the 'logical' key order.
* @return a new RangeQuery instance with ascending flag set.
*/
public RangeQuery<K, V> withAscendingKeys()
...
} |
According to KIP-968, we introduce a public enum ResultOrder.
ResultOrder enum
It helps with specifying the order of the returned results by the query.
Code Block | ||||
---|---|---|---|---|
| ||||
package org.apache.kafka.streams.query;
public enum ResultOrder {
ANY,
ASCENDING,
DESCENDING
} |
Test Plan
This time, our goal is to implement reverseRange
and reverseAll
functionalities. While these terms are used for clarity, in practice, they correspond to RangeQuery.withRange().withDescendingKeys()
and RangeQuery.withNoBounds().withDescendingKeys()
, respectively. To ensure the accurate retrieval of results for both functionalities, adjustments to IQv2StoreIntegrationTest
are required. In our previous approach, we stored query results in a set, which doesn't maintain order. I've transitioned to using a list for storing query results, enabling us to distinguish between rangeQuery
and reverseQuery
. Here, rangeQuery
refers to standard queries (those not using withDescendingKeys()
) such as withRange()
, withLowerBound()
, withUpperBound()
, and withNoBounds()
. In contrast, reverseQuery
denotes queries that employ the withDescendingKeys()
method.
We've transitioned the expectedValue
from a Set
to a List
and arranged the partition numbers in order. This organization assists us in predicting the results. If the partition numbers were random, predicting the outcome would be challenging. Ultimately, this enables us to obtain and store the answer in the expectedValue
. Consequently, the results between rangeQuery
and reverseQuery
will differThe ReverseRangeQuery
class will be used for both reverseRange and reverseAll queries. A reverseAll is performed when no lower and no upper bound is specified. A reverseRange query retrieves a set of keys, specified using an upper and/or lower bound, from the underlying KV store. A reverseAll, on the other hand, retrieves all keys contained in the KV store.
Code Block | ||||
---|---|---|---|---|
| ||||
public class IQv2StoreIntegrationTest {
...
@SuppressWarnings("unchecked")
public <V> void shouldHandleRangeQuery(
final Optional<Integer> lower,
final Optional<Integer> upper,
final boolean isKeyAscending,
final Function<V, Integer> valueExtactor,
final List<Integer> expectedValue) {
final RangeQuery<Integer, V> query;
if (isKeyAscending) {
query = RangeQuery.withRange(lower.orElse(null), upper.orElse(null));
} else {
query = (RangeQuery<Integer, V>) RangeQuery.withRange(lower.orElse(null), upper.orElse(null)).withDescendingKeys();
}
...
} else {
final List<Integer> actualValue = new ArrayList<>();
...
final List<Integer> partitions = new ArrayList<>(queryResult.keySet());
partitions.sort(null);
for (final int partition : partitions) {
...
}
...
}
|
Compatibility, Deprecation, and Migration Plan
- Utilizing the existing
RangeQuery
class, we can make some modifications to realize the concepts ofreverseRange
andreverseAll
. To reiterate,reverseRange
andreverseAll
are not classes or methods but merely concepts. - Since nothing is deprecated in this KIP, users have no need to migrate unless they want to.
Rejected Alternatives
After initial plans to create a ReverseRangeQuery
from the ground up, we opted to leverage existing code from the RangeQuery
class following further discussions.
Code Block | ||||
---|---|---|---|---|
| ||||
@Evolving public final class ReverseRangeQuery<K, V> implements Query<KeyValueIterator<K, V>> { private final Optional<K> lower; private final Optional<K> upper; private ReverseRangeQuery(final Optional<K> lower, final Optional<K> upper) { this.lower = lower; this.upper = upper; } /** * Interactive range query using a lower and upper bound to filter the keys returned. * @param lower The key that specifies the lower bound of the range * @param upper The key that specifies the upper bound of the range * @param <K> The key type * @param <V> The value type */ public static <K, V> ReverseRangeQuery<K, V> withRange(final K lower, final K upper) { return new ReverseRangeQuery<>(Optional.ofNullable(lower), Optional.ofNullable(upper)); } /** * Interactive range query using an upper bound to filter the keys returned. * If both <K,V> are null, RangQuery returns a full range scan. * @param upper The key that specifies the upper bound of the range * @param <K> The key type * @param <V> The value type */ public static <K, V> ReverseRangeQuery<K, V> withUpperBound(final K upper) { return new ReverseRangeQuery<>(Optional.empty(), Optional.of(upper)); } /** * Interactive range query using a lower bound to filter the keys returned. * @param lower The key that specifies the lower bound of the range * @param <K> The key type * @param <V> The value type */ public static <K, V> ReverseRangeQuery<K, V> withLowerBound(final K lower) { return new ReverseRangeQuery<>(Optional.of(lower), Optional.empty()); } /** * Interactive scan query that returns all records in the store. * @param <K> The key type * @param <V> The value type */ public static <K, V> ReverseRangeQuery<K, V> withNoBounds() { return new ReverseRangeQuery<>(Optional.empty(), Optional.empty()); } /** * The lower bound of the query, if specified. */ public Optional<K> getLowerBound() { return lower; } /** * The upper bound of the query, if specified */ public Optional<K> getUpperBound() { return upper; } } |
Compatibility, Deprecation, and Migration Plan
- Because we have already have RangeQuery, so we can achieve the ReverseRangeQuery following RangeQuery.
- Since nothing is deprecated in this KIP, users have no need to migrate unless they want to.
Rejected Alternatives
...