[
https://issues.apache.org/jira/browse/FLINK-20370?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17455065#comment-17455065
]
Jingsong Lee edited comment on FLINK-20370 at 12/10/21, 2:11 AM:
-----------------------------------------------------------------
Fixed via:
part1: Fix wrong results when sink primary key is not the same with query
result's changelog upsert key
master: f8f6935adc841529ecdc0636174650cffbf73719
release-1.14: 7c5ddbd201005e55ab68b4db7ee74c7cbeb13400
release-1.13: 54aaffc7d9b2b06726294cf636d1b9acf74c5d49
part2: introduce 'table.exec.sink.keyed-shuffle' option to auto keyby on sink's
pk if parallelism are not the same for insertOnly input
master: 9e76585f1fa110288604913a73d86ac2f1542777
part2 does not cherry-pick to 1.14 because it may affect the normal plan and
lead to incompatibility.
was (Author: lzljs3620320):
Fixed via:
part1: Fix wrong results when sink primary key is not the same with query
result's changelog upsert key
master: f8f6935adc841529ecdc0636174650cffbf73719
release-1.14: 7c5ddbd201005e55ab68b4db7ee74c7cbeb13400
part2: introduce 'table.exec.sink.keyed-shuffle' option to auto keyby on sink's
pk if parallelism are not the same for insertOnly input
master: 9e76585f1fa110288604913a73d86ac2f1542777
part2 does not cherry-pick to 1.14 because it may affect the normal plan and
lead to incompatibility.
> Result is wrong when sink primary key is not the same with query
> ----------------------------------------------------------------
>
> Key: FLINK-20370
> URL: https://issues.apache.org/jira/browse/FLINK-20370
> Project: Flink
> Issue Type: Bug
> Components: Table SQL / Planner
> Affects Versions: 1.12.0
> Reporter: Jark Wu
> Assignee: lincoln lee
> Priority: Critical
> Labels: pull-request-available
> Fix For: 1.15.0, 1.14.1, 1.13.4
>
>
> Both sources are upsert-kafka which synchronizes the changes from MySQL
> tables (source_city, source_customer). The sink is another MySQL table which
> is in upsert mode with "city_name" primary key. The join key is "city_id".
> In this case, the result will be wrong when updating
> {{source_city.city_name}} column in MySQL, as the UPDATE_BEFORE is ignored
> and the old city_name is retained in the sink table.
> {code}
> Sink(table=[default_catalog.default_database.sink_kafka_count_city],
> fields=[city_name, count_customer, sum_gender], changelogMode=[NONE])
> +- Calc(select=[city_name, CAST(count_customer) AS count_customer,
> CAST(sum_gender) AS sum_gender], changelogMode=[I,UA,D])
> +- Join(joinType=[InnerJoin], where=[=(city_id, id)], select=[city_id,
> count_customer, sum_gender, id, city_name],
> leftInputSpec=[JoinKeyContainsUniqueKey],
> rightInputSpec=[JoinKeyContainsUniqueKey], changelogMode=[I,UA,D])
> :- Exchange(distribution=[hash[city_id]], changelogMode=[I,UA,D])
> : +- GlobalGroupAggregate(groupBy=[city_id], select=[city_id,
> COUNT_RETRACT(count1$0) AS count_customer, SUM_RETRACT((sum$1, count$2)) AS
> sum_gender], changelogMode=[I,UA,D])
> : +- Exchange(distribution=[hash[city_id]], changelogMode=[I])
> : +- LocalGroupAggregate(groupBy=[city_id], select=[city_id,
> COUNT_RETRACT(*) AS count1$0, SUM_RETRACT(gender) AS (sum$1, count$2)],
> changelogMode=[I])
> : +- Calc(select=[city_id, gender], changelogMode=[I,UB,UA,D])
> : +- ChangelogNormalize(key=[customer_id],
> changelogMode=[I,UB,UA,D])
> : +- Exchange(distribution=[hash[customer_id]],
> changelogMode=[UA,D])
> : +- MiniBatchAssigner(interval=[3000ms],
> mode=[ProcTime], changelogMode=[UA,D])
> : +- TableSourceScan(table=[[default_catalog,
> default_database, source_customer]], fields=[customer_id, city_id, age,
> gender, update_time], changelogMode=[UA,D])
> +- Exchange(distribution=[hash[id]], changelogMode=[I,UA,D])
> +- ChangelogNormalize(key=[id], changelogMode=[I,UA,D])
> +- Exchange(distribution=[hash[id]], changelogMode=[UA,D])
> +- MiniBatchAssigner(interval=[3000ms], mode=[ProcTime],
> changelogMode=[UA,D])
> +- TableSourceScan(table=[[default_catalog,
> default_database, source_city]], fields=[id, city_name], changelogMode=[UA,D])
> {code}
> We have suggested users to use the same key of the query as the primary key
> on sink in the documentation:
> https://ci.apache.org/projects/flink/flink-docs-master/dev/table/sql/queries.html#deduplication.
> We should make this attention to be more highlight in CREATE TABLE page.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
