[
https://issues.apache.org/jira/browse/KAFKA-10148?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Bruno Cadonna updated KAFKA-10148:
----------------------------------
Description:
System test {{StreamsEosTest.test_failure_and_recovery}} for eos-beta exposes a
bug that results in wrong results in the output topic. The cause seems to be a
too low end offset during restoration of a state store.
Example:
The system test computes a minimum aggregate over records in an input topic and
writes the results to an output topic. The input topic partition {{data-1}}
contains the following records among others:
{code}
...
offset: 125 CreateTime: 1591690264681 keysize: 5 valuesize: 4 sequence: 125
headerKeys: [] key: 14920 payload: 9215
...
offset: 1611 CreateTime: 1591690297424 keysize: 5 valuesize: 4 sequence: 1611
headerKeys: [] key: 14920 payload: 1595
...
offset: 2104 CreateTime: 1591690308542 keysize: 5 valuesize: 4 sequence: 2104
headerKeys: [] key: 14920 payload: 9274
...
{code}
The output topic partition {{min-1}} contains:
{code}
...
offset: 125 CreateTime: 1591690264681 keysize: 5 valuesize: 4 sequence: 125
headerKeys: [] key: 14920 payload: 9215
...
offset: 1828 CreateTime: 1591690297424 keysize: 5 valuesize: 4 sequence: 1213
headerKeys: [] key: 14920 payload: 1595
...
offset: 2324 CreateTime: 1591690308542 keysize: 5 valuesize: 4 sequence: 10
headerKeys: [] key: 14920 payload: 9215
...
{code}
The last record is obviously wrong because 1595 is less than 9215.
To test the resilience to an unexpected failure of a Streams client, the system
tests aborts a Streams client, i.e., the client is closed in a dirty manner.
This dirty close causes the Streams client to restore its local state store
that maintains the minimum aggregate from the beginning of the changelog topic
partitions {{EosTest-KSTREAM-AGGREGATE-STATE-STORE-0000000003-changelog-1}}.
The partition {{EosTest-KSTREAM-AGGREGATE-STATE-STORE-0000000003-changelog-1}}
contains:
{code}
...
offset: 125 CreateTime: 1591690264681 keysize: 5 valuesize: 4 sequence: 125
headerKeys: [] key: 14920 payload: 9215
...
offset: 1828 CreateTime: 1591690297424 keysize: 5 valuesize: 4 sequence: 1213
headerKeys: [] key: 14920 payload: 1595
...
offset: 2324 CreateTime: 1591690308542 keysize: 5 valuesize: 4 sequence: 10
headerKeys: [] key: 14920 payload: 9215
...
{code}
Also here the last record is wrong.
was:
System test {{StreamsEosTest.test_failure_and_recovery}} for eos-beta exposes a
bug that results in wrong results in the output topic. The cause seems to be a
too low end offset during restoration of a state store.
Example:
The system test computes a minimum aggregate over records in an input topic and
writes the results to an output topic. The input topic partition {{data-1}}
contains the following records among others:
{code}
...
offset: 125 CreateTime: 1591690264681 keysize: 5 valuesize: 4 sequence: 125
headerKeys: [] key: 14920 payload: 9215
...
offset: 1611 CreateTime: 1591690297424 keysize: 5 valuesize: 4 sequence: 1611
headerKeys: [] key: 14920 payload: 1595
...
offset: 2104 CreateTime: 1591690308542 keysize: 5 valuesize: 4 sequence: 2104
headerKeys: [] key: 14920 payload: 9274
...
{code}
The output topic partition {{min-1}} contains:
{code}
...
offset: 125 CreateTime: 1591690264681 keysize: 5 valuesize: 4 sequence: 125
headerKeys: [] key: 14920 payload: 9215
...
offset: 1828 CreateTime: 1591690297424 keysize: 5 valuesize: 4 sequence: 1213
headerKeys: [] key: 14920 payload: 1595
...
offset: 2324 CreateTime: 1591690308542 keysize: 5 valuesize: 4 sequence: 10
headerKeys: [] key: 14920 payload: 9215
...
{code}
The last record is obviously wrong because 1595 is less than 9215.
> Kafka Streams Restores too few Records with eos-beta Enabled
> -------------------------------------------------------------
>
> Key: KAFKA-10148
> URL: https://issues.apache.org/jira/browse/KAFKA-10148
> Project: Kafka
> Issue Type: Bug
> Components: streams
> Reporter: Bruno Cadonna
> Assignee: Bruno Cadonna
> Priority: Blocker
> Fix For: 2.6.0
>
>
> System test {{StreamsEosTest.test_failure_and_recovery}} for eos-beta exposes
> a bug that results in wrong results in the output topic. The cause seems to
> be a too low end offset during restoration of a state store.
> Example:
> The system test computes a minimum aggregate over records in an input topic
> and writes the results to an output topic. The input topic partition
> {{data-1}} contains the following records among others:
> {code}
> ...
> offset: 125 CreateTime: 1591690264681 keysize: 5 valuesize: 4 sequence: 125
> headerKeys: [] key: 14920 payload: 9215
> ...
> offset: 1611 CreateTime: 1591690297424 keysize: 5 valuesize: 4 sequence: 1611
> headerKeys: [] key: 14920 payload: 1595
> ...
> offset: 2104 CreateTime: 1591690308542 keysize: 5 valuesize: 4 sequence: 2104
> headerKeys: [] key: 14920 payload: 9274
> ...
> {code}
> The output topic partition {{min-1}} contains:
> {code}
> ...
> offset: 125 CreateTime: 1591690264681 keysize: 5 valuesize: 4 sequence: 125
> headerKeys: [] key: 14920 payload: 9215
> ...
> offset: 1828 CreateTime: 1591690297424 keysize: 5 valuesize: 4 sequence: 1213
> headerKeys: [] key: 14920 payload: 1595
> ...
> offset: 2324 CreateTime: 1591690308542 keysize: 5 valuesize: 4 sequence: 10
> headerKeys: [] key: 14920 payload: 9215
> ...
> {code}
> The last record is obviously wrong because 1595 is less than 9215.
> To test the resilience to an unexpected failure of a Streams client, the
> system tests aborts a Streams client, i.e., the client is closed in a dirty
> manner. This dirty close causes the Streams client to restore its local state
> store that maintains the minimum aggregate from the beginning of the
> changelog topic partitions
> {{EosTest-KSTREAM-AGGREGATE-STATE-STORE-0000000003-changelog-1}}. The
> partition {{EosTest-KSTREAM-AGGREGATE-STATE-STORE-0000000003-changelog-1}}
> contains:
> {code}
> ...
> offset: 125 CreateTime: 1591690264681 keysize: 5 valuesize: 4 sequence: 125
> headerKeys: [] key: 14920 payload: 9215
> ...
> offset: 1828 CreateTime: 1591690297424 keysize: 5 valuesize: 4 sequence: 1213
> headerKeys: [] key: 14920 payload: 1595
> ...
> offset: 2324 CreateTime: 1591690308542 keysize: 5 valuesize: 4 sequence: 10
> headerKeys: [] key: 14920 payload: 9215
> ...
> {code}
> Also here the last record is wrong.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
