Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

...

Current state: Under Discussion

Discussion thread:   here

JIRA: KAFKA-15348 

Please keep the discussion on the mailing list rather than commenting on the wiki (wiki discussions get unwieldy fast).

Motivation

The main goal is supporting to support interactive queries in the presence of versioned state stores (KIP-889) in AK. This KIP is the successor of KIP-960 and KIP-969968
For this KIP, the following query types are considered to be implemented.  :

Range Queries

  1. key-range latest-value query
  2. key-range with lower bound latest-value query
  3. key-range with upper bound latest-value query
  4. all-keys (no bound) latest-value query
  5. key-range query with timestamp (upper) bound
  6. key-range with lower bound with timestamp (upper) bound 
  7. key-range with upper bound with timestamp (upper) bound
  8. all-keys (no bound) with timestamp (upper) bound
  9. key-range query with timestamp range
  10. key-range query with lower bound with timestamp range
  11. key-range query with upper bound with timestamp range
  12. all-keys (no bound) with timestamp range
  13. key-range query all-versions
  14. key-range query with lower bound all-versions
  15. key-range query with upper bond all-versions
  16. all-keys query (no bound) all-versions (entire store)

...

In this KIP we propose the public classes, MultiVersionedRangeQuery that will be described in the next section. More over a method will be added to the VersionedKeyValueStore interface.

Proposed Changes

For supporting range queries, MultiVersionedRangeQuery class is used.

  • The methods are composable. Therefore, the meaningless combinations such as withRange(k1, k2).asOf(t1).allVersions() end up throwing a RunTimeException (for example NotSupportedException). The fromTime and toTime methods specify the time range while the withLowerKeyBound and withUpperKeyBound methods specify the key bounds. 
    • If a user applies the same time limit multiple times such as MultiVersionedRangeQuery.withLowerKeyBound(k1).from(t1).from(t2), then the last one wins (it will be translated to MultiVersionedRangeQuery.withLowerKeyBound(k1).from(t2)).

    • Defining a query with time range (empty, t1] will be translated
    • Defining a query with time range (empty, t1] will be translated into [0, t1]
    • Defining a query with time range (t1, empty) will be translated into [t1, MAX)
    • A query with no specified time range will be interpreted as a normal range query that returns the records with the latest timestamptranslated into [0, MAX). It means that the query will return all the versions of all the records with specified key range.
  • As explained in the javadocs, the query returns all valid records within the specified time range.
    • The fromTimestampfromTime specifies the starting point. There can be records which have been inserted before the fromTimestampfromTime and are valid in the time range. 
    • The asOfTimestamptoTime specifies the ending point. Records that have been inserted at asOfTimestamptoTime are returned by the query as well.
  • The overall order of the returned records is by Key. The method orderByTimestamp() can make the overall order by timestamp.
    • The order for both key and timestamp is by default ascending. They can be changed by the methods withDescendingKeys() and withDescendingTimestamps() respectively.
  • No ordering is guaranteed for the return records.
Code Block
languagejava
titleMultiVersionedRangeQuery
linenumbers
Code Block
languagejava
titleMultiVersionedRangeQuery
linenumberstrue
package org.apache.kafka.streams.query;


/**
 * Interactive query for retrieving a set of records with keys within a specified key range and time
 * range.
 */

@Evolving
public final class MultiVersionedRangeQuery<K, V> implements
    Query<KeyValueIterator<K, VersionedRecord<V>>> {

  private final Optional<K> lower;
  private final Optional<K> upper;
  private final Optional<Instant> fromTimestampfromTime;

  private final Optional<Instant> asOfTimestamptoTime;

  private finalMultiVersionedRangeQuery(
  boolean isKeyAscending;
  private final booleanOptional<K> isTimeAscending;
lower,
     private final boolean isOrderedByKey;
Optional<K> upper,
  private MultiVersionedRangeQuery(
      final Optional<K>Optional<Instant> lowerfromTime,
      final Optional<Instant> Optional<K>toTime) upper,{
    this.lower  final Optional<Instant> fromTimestamp,= lower;
    this.upper  final Optional<Instant> asOfTimestamp,= upper;
    this.fromTime  final boolean isOrderedByKey,= fromTime;
    this.toTime = finaltoTime;
 boolean isKeyAscending,}

   /** 
  final boolean* isTimeAscending)Interactive {
range query using a this.lower and upper =bound lower;
to filter the keys this.upper = upper;returned. * For each 
   * this.fromTimestampkey =the fromTimestamp;
records valid within the this.asOfTimestampspecified =time asOfTimestamp;
range are   thisreturned.isOrderedByKey =* isOrderedByKey;
In case the time this.isKeyAscendingrange =is isKeyAscending;
   * this.isTimeAscendingnot =specified isTimeAscending;
just the }

  /**latest record for each key is returned. 
   * Interactive@param rangelower queryThe usingkey athat lowerspecifies andthe upperlower bound of tothe filterrange the
 keys returned. * For@param each
upper The key *that keyspecifies the recordsupper validbound withinof the specifiedrange time
 range are returned. * In@param case<K> theThe timekey rangetype is
   * not@param specified<V> justThe thevalue latesttype record
 for each key is returned.*/ 
   *public @paramstatic lower<K, TheV> keyMultiVersionedRangeQuery<K, thatV> specifieswithKeyRange(final theK lower, final bound of the rangeK upper);


   /**
   * @paramInteractive upperrange Thequery keyusing thata specifieslower thebound upperto boundfilter of the range
keys returned.  * @param <K>   TheFor each key typethe
   * @paramrecords <V>valid within the Thespecified valuetime type
range are returned. */
 In publiccase staticthe <K,time V>range MultiVersionedRangeQuery<K,is V>not
 withRange(final K lower,* finalspecified K upper);

  /**
   * Interactive range query using a lower bound to filter the keys returned. * For each key thejust the latest record for each key is returned.
   * @param lower The key that specifies the lower bound of the range
   * records@param valid<K> within the specifiedThe timekey rangetype
 are returned. * In@param case<V> the time rangeThe isvalue nottype
   */
 specified justpublic thestatic latest<K, recordV> forMultiVersionedRangeQuery<K, eachV> keywithLowerKeyBound(final is returned.K lower);

  /**
 * @param lower* TheInteractive keyrange thatquery specifiesusing thea lower bound to offilter the range
keys returned.  * @paramFor <K>  each The key typethe
   * @paramrecords <V>valid within the Thespecified valuetime type
range are returned. */
 In publiccase staticthe <K,time V>range MultiVersionedRangeQuery<K,is V>not
 withLowerBound(final K lower);

*  /**specified just the latest record for each key is returned.
   * @param upper InteractiveThe rangekey querythat usingspecifies athe lower bound toof filterthe therange
 keys returned. * For each@param <K>   The key thetype
   * records@param valid<V> within the specifiedThe timevalue rangetype
 are returned. * In case the time range is not
   * specified just the latest record for each key is returned.
   * @param upper The key that specifies the lower bound of the range
   * @param <K>   The key type/
  public static <K, V> MultiVersionedRangeQuery<K, V> withUpperKeyBound(final K upper);

  /**
   * Interactive scan query that returns all records in the store. * For each key the records valid
   * within the specified time range are returned. * In case the time range is not specified just
   * @paramthe <V>latest record for The valueeach key is returned.
   * @param <K> The key type
   * @param <V> The value type
   */
  public static <K, V> MultiVersionedRangeQuery<K, V> withUpperBoundallKeys(final K upper);

  /**
   * InteractiveSpecifies scanthe querystarting thattime returnspoint allfor recordsthe inkey the storequery. The *range Forquery eachreturns keyall the records valid
   * within the specified time range that are returned.valid *in In case the time range starting from isthe nottimestamp specified just{@code fromTime}.
   * the@param latestfromTime recordThe forstarting each key is returned.
   * @param <K> The key typetime point
   * @param <V> The value type
   */
  public static <K, V> MultiVersionedRangeQuery<K, V> withNoBoundsfromTime(Instant fromTime);

  /**
   * Specifies the startingending time point for the key query. The range query returns all the records that
   * thathave aretimestamp valid in the time range starting from the timestamp <= {@code fromTimestamptoTime}.
   * @param fromTimestamptoTime The startingending time point
   */
  public MultiVersionedRangeQuery<K, V> fromtoTime(Instant fromTimestamptoTime);

  /**
   * SpecifiesThe thelower endingbound timeof pointthe forquery, theif key queryspecified.
 The range query*/
 returns allpublic the records thatOptional<K> lowerKeyBound();

   /**
 have timestamp <=* {@code asOfTimestamp}.
   * @param asOfTimestamp The ending time pointThe upper bound of the query, if specified
   */
  public MultiVersionedRangeQuery<K, V> asOf(Instant asOfTimestampOptional<K> upperKeyBound();

  /**
   * Specifies theThe starting andtime endingpoint points ofof the range query, asif MINspecified
 and MAX respectively.*/
  public * Therefore, the query returns all the existing records in the state store with keys within theOptional<Instant> fromTime();

  /**
   * The ending time point of the query, if specified
   * specified key range./
  public * @throws RuntimeException if {@code fromTimestamp} or {@code asOfTimestamp} have been already
   *       Optional<Instant> toTime();

}


Examples

The following example illustrates the use of the VersionedKeyQuery class to query a versioned state store.

Imagine we have the following records 

put(1, 1, time=2023-01-01T10:00:00.00Z)

put(1, null, time=2023-01-05T10:00:00.00Z)

put(2, 20, time=2023-01-10T10:00:00.00Z)

put(3, 30, time=2023-01-12T10:00:00.00Z)

put(1, 2, time=2023-01-15T10:00:00.00Z)

put(1, 3, time=2023-01-20T10:00:00.00Z)

put(2, 30, time=2023-01-25T10:00:00.00Z)

Code Block
languagejava
linenumberstrue
// example 1: MultiVersionedRangeQuery without specifying any time bound will be interpreted as all versions
final MultiVersionedRangeQuery<Integer, Integer> query1 =
        MultiVersionedRangeQuery.withKeyRange(1, 2);
final StateQueryRequest<KeyValueIterator<Integer, VersionedRecord<Integer>>> request1 =
        inStore("my_store").withQuery(query1);
 
final StateQueryResult<KeyValueIterator<Integer, VersionedRecord<Integer>>> versionedRangeResult1 = kafkaStreams.query(request1);
 
// Get the results from all partitions.
final Map<Integer, QueryResult<KeyValueIterator<Integer, VersionedRecord<Integer>>>> partitionResults = versionedRangeResult.getPartitionResults();
for (final Entry<Integer, QueryResult<KeyValueIterator<Integer, VersionedRecord<Integer>>>> entry : partitionResults.entrySet()) {
    try (final KeyValueIterator<Integer, VersionedRecord<Integer>> iterator = entry.getValue().getResult()) {
        while (iterator.hasNext()) {
          final KeyValue<Integer, VersionedRecord<Integer>> record = iterator.next();
          Integer key = record.key;
      specified.
   */
   publicInteger MultiVersionedRangeQuery<K,value V> allVersions= record.value.value();

  /**
   * Specifies the overall order ofLong returnedtimestamp records by timestamp
   */
  public MultiVersionedRangeQuery<K, V> orderByTimestamp= record.value.timestamp();
		  Long validTo = record.value.validTo();

	   /**
   * Specifies the order of keys as descending.
   */
  public MultiVersionedRangeQuery<K, V> withDescendingKeys();

  /**    
   * Specifies the order of the timestamps as descending.
   */
  public VersionedRangeQuery<K, V> withDescendingTimestamps();

  /**
   * The lower bound of the query, if specified.
   */
  public Optional<K> getLowerKeyBound();

  /**
   * The upper bound of the query, if specified
   */
  public Optional<K> getUpperKeyBound();

  /**
   * The starting time point of the query, if specified
   */
  public Optional<Instant> getFromTimestamp();

  /**
   * The ending time point of the query, if specified
   */
  public Optional<Instant> getAsOfTimestamp();

  /**
   * @return true if the query orders the returned records by key 
   */
  public boolean isOrderedByKey();

  /**
   * @return true if the query returns records in ascending order of keys
   */
  public boolean isKeyAscending();

  /**
   * @return true if the query returns records in ascending order of timestamps
   */
  public boolean isRangeAscending();

}

Examples

The following example illustrates the use of the VersionedKeyQuery class to query a versioned state store.

Code Block
languagejava
linenumberstrue
final MultiVersionedRangeQuery<Integer, Integer> query =
        MultiVersionedRangeQuery.withKeyRange(1, 2).from(Instant.parse(2023-08-03T10:37:30.00Z)).asOf(Instant.parse(2023-09-04T10:37:30.00Z));
 
final StateQueryRequest<KeyValueIterator<Integer, VersionedRecord<Integer>>> request =
        inStore("my_store").withQuery(query);
 
final StateQueryResult<KeyValueIterator<Integer, VersionedRecord<Integer>>> versionedRangeResult = kafkaStreams.query(request);
 
// Get the results from all partitions.
final Map<Integer, QueryResult<KeyValueIterator<Integer, VersionedRecord<Integer>>>> partitionResults = versionedRangeResult.getPartitionResults();
for (final Entry<Integer, QueryResult<KeyValueIterator<Integer, VersionedRecord<Integer>>>> entry : partitionResults.entrySet()) {
    try (final KeyValueIterator<Integer, VersionedRecord<Integer>> iterator = entry.getValue().getResult()) {
        while (iterator.hasNext()) {
          final KeyValue<Integer, VersionedRecord<Integer>> record = iterator.next();
          Integer key = record.key;
          Integer value = record.value.value();
          Long timestamp = record.value.timestamp();
        }
     }
}System.out.println ("key,value: " + key + "," +value + ", timestamp: " + Instant.ofEpochSecond(timestamp)+ ", valid till: " + Instant.ofEpochSecond(validTo));  	
        }
     }
}
/* the printed output will be
    key,value: 1,1, timestamp: 2023-01-01T10:00:00.00Z, valid till: 2023-01-05T10:00:00.00Z  
    key,value: 1,2, timestamp: 2023-01-15T10:00:00.00Z, valid till: 2023-01-20T10:00:00.00Z 
    key,value: 1,3, timestamp: 2023-01-20T10:00:00.00Z, valid till: now 
    key,value: 2,20, timestamp: 2023-01-10T10:00:00.00Z, valid till: 2023-01-25T10:00:00.00Z 
    key,value: 2,30, timestamp: 2023-01-25T10:00:00.00Z, valid till: now 
*/

// example 2: The value of the records with key range (1,2) from 2023-01-17 Time: 10:00:00.00Z till 2023-01-30 T10:00:00.00Z


MultiVersionedRangeQuery<Integer, Integer> query2 = MultiVersionedRangeQuery.withKeyRange(1, 2);
query2 = query2.fromTime(Instant.parse(2023-01-17T10:00:00.00Z)).toTime(Instant.parse(2023-01-30T10:00:00.00Z));
 
final StateQueryRequest<KeyValueIterator<Integer, VersionedRecord<Integer>>> request2 =
        inStore("my_store").withQuery(query2);
 
final StateQueryResult<KeyValueIterator<Integer, VersionedRecord<Integer>>> versionedRangeResult2 = kafkaStreams.query(request2);
 
// Get the results from all partitions.
final Map<Integer, QueryResult<KeyValueIterator<Integer, VersionedRecord<Integer>>>> partitionResults2 = versionedRangeResult2.getPartitionResults();
for (final Entry<Integer, QueryResult<KeyValueIterator<Integer, VersionedRecord<Integer>>>> entry : partitionResults.entrySet()) {
    try (final KeyValueIterator<Integer, VersionedRecord<Integer>> iterator = entry.getValue().getResult()) {
        while (iterator.hasNext()) {
          final KeyValue<Integer, VersionedRecord<Integer>> record = iterator.next();
          Integer key = record.key;
          Integer value = record.value.value();
          Long timestamp = record.value.timestamp();
 		  Long validTo = record.value.validTo();
	      System.out.println ("key,value: " + key + "," +value + ", timestamp: " + Instant.ofEpochSecond(timestamp)+ ", valid till: " + Instant.ofEpochSecond(validTo));           
		}
     }
}

/* the printed output will be   
    key, value: 2,30, timestamp: 2023-01-25T10:00:00.00Z, valid till: now 
    key, value: 1,3, timestamp: 2023-01-20T10:00:00.00Z, valid till: now
    key, value: 1,2, timestamp: 2023-01-15T10:00:00.00Z, valid till: 2023-01-20T10:00:00.00Z
    key, value: 2,20, timestamp: 2023-01-10T10:00:00.00Z, valid till: 2023-01-25T10:00:00.00Z 
*/


Compatibility, Deprecation, and Migration Plan

  • Since this is a completely new set of APIs, no backward compatibility concerns are anticipated. 
  • Since nothing is deprecated in this KIP, no backward compatibility concerns are anticipated. Since nothing is deprecated in this KIP, users have no need to migrate unless they want tousers have no need to migrate unless they want to.

Test Plan

The range interactive queries will be tested in versioned stored IQv2 integration test (like non-versioned range queries). Moreover , there will be unit tests where ever needed. 

Rejected Alternatives

The initial plan was to provide ordering based on key and/or timestamp, which is removed from the KIP and may be provided by subsequent KIPs based on user demand.