[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17835595#comment-17835595
]
Chia-Ping Tsai commented on KAFKA-16310:
have updated KIP-734
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.8.0
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17835583#comment-17835583
]
Chia-Ping Tsai commented on KAFKA-16310:
I will update KIP-734 later since the web site has no response
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.8.0
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17834131#comment-17834131
]
Chia-Ping Tsai commented on KAFKA-16310:
{quote}
Will this bug not be fixed in 3.6.2 ?
{quote}
yep. As we discussed above, all paths do NOT use same way to find the offset of
max timestamp, so it is not a regression of both 3.6 and 3.7. Instead, it is a
issue which is existent for a while :)
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.8.0
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17834129#comment-17834129
]
HiroArai commented on KAFKA-16310:
--
[~chia7712] [~omkreddy]
> [~ijuma] Are you worry about the 3.6.2? If so, the fixes are reverted
>already. see
[https://home.apache.org/~manikumar/kafka-3.6.2-rc2/RELEASE_NOTES.html]
This is my first time commenting on kafka's jira. I apologize if this is an
unnecessary question.
The fix for this bug is not mentioned in the release notes of 3.6.2-rc2.
Will this bug not be fixed in 3.6.2 ?
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.8.0
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17832228#comment-17832228
]
Chia-Ping Tsai commented on KAFKA-16310:
{quote}
I got more context and also feel that we can transfer the workload to
listMaxTimestamp when clients fetch this since it's rare. What do you think?
{quote}
yep, we are on the same page. I have filed a PR according to the solution.
Please take a look at https://github.com/apache/kafka/pull/15621
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.8.0
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17832226#comment-17832226
]
Johnny Hsu commented on KAFKA-16310:
thanks [~junrao] for pointing this out and sharing the solution, and thanks
[~chia7712] [~showuon] for those discussions and revert it for 3.6.
{quote}Since this is a rare operation, paying the decompression overhead is
fine.
Adding a new field in the batch requires record format change, which is a much
bigger effort. For now, the easiest thing is to add a method in Batch to find
out offsetOfMaxTimestanp by iterating all records.
Regarding the optimization on the leader side by caching offsetOfMaxTimestanp,
we could do it. However, my understanding is that listMaxTimestamp is rare and
I am not sure if it's worth the additional complexity.
{quote}
have go through the comments and had a offline discussion with Chia-Ping, I got
more context and also feel that we can transfer the workload to
listMaxTimestamp when clients fetch this since it's rare. What do you think?
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.8.0
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17831966#comment-17831966
]
Chia-Ping Tsai commented on KAFKA-16310:
see the following link to check the revert from 3.7
https://github.com/apache/kafka/commit/bd5989dd195d42c1608582316367a03b2c78cb11
https://github.com/apache/kafka/commit/fc646f920701b792eb683bacd513a1f20909f6bc
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.8.0
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17831965#comment-17831965
]
Chia-Ping Tsai commented on KAFKA-16310:
{quote}
Since the follower only maintains offsetForMaxTimestamp at the batch level, the
listMaxTimestamp API was never implemented correctly. So, technically, there
was no regression for listMaxTimestamp. We could just fix this issue in trunk
without backporting to the old branch.
{quote}
I will revert KAFKA-16341 and KAFKA-16342 from 3.7
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.8.0, 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17831959#comment-17831959
]
Jun Rao commented on KAFKA-16310:
-
Since the follower only maintains offsetForMaxTimestamp at the batch level, the
listMaxTimestamp API was never implemented correctly. So, technically, there
was no regression for listMaxTimestamp. We could just fix this issue in trunk
without backporting to the old branch.
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.8.0, 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17831848#comment-17831848
]
Ismael Juma commented on KAFKA-16310:
-
Perfect, thanks!
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.8.0, 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17831845#comment-17831845
]
Chia-Ping Tsai commented on KAFKA-16310:
{quote}
Team, there are serious issues that are fixed by 3.6.2. We should not delay it
due to a long standing issue. As I said elsewhere, if it's not a regression or
a security issue, it should be avoided. Precisely to avoid the issue that
happened here. Let's revert and move on please.
{quote}
[~ijuma] Are you worry about the 3.6.2? If so, the fixes are reverted already.
see
https://github.com/apache/kafka/commit/da1ee97f11c1c828a93db37023122b647a5a271e
and
https://github.com/apache/kafka/commit/64f7a0a300c70b00c0cc357d0a936ea8b42b69fb
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.8.0, 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17831837#comment-17831837
]
Ismael Juma commented on KAFKA-16310:
-
Team, there are serious issues that are fixed by 3.6.2. We should not delay it
due to a long standing issue. As I said elsewhere, if it's not a regression or
a security issue, it should be avoided. Precisely to avoid the issue that
happened here. Let's revert and move on please.
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.8.0, 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17831834#comment-17831834
]
Jun Rao commented on KAFKA-16310:
-
[~chia7712] :
{quote}Did you mean that we should change the schema to have a new field to
keep `offsetOfMaxTimestamp`? or just add a new method to Batch to iterate all
records to find out offsetOfMaxTimestamp?
{quote}
Adding a new field in the batch requires record format change, which is a much
bigger effort. For now, the easiest thing is to add a method in Batch to find
out offsetOfMaxTimestanp by iterating all records.
Regarding the optimization on the leader side by caching offsetOfMaxTimestanp,
we could do it. However, my understanding is that listMaxTimestamp is rare and
I am not sure if it's worth the additional complexity.
[~showuon] : Regarding the fix, we could fix forward if we could provide the
fix soon. Otherwise, we probably want to revert the fixes to avoid new code
depending on the new semantic.
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.8.0, 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17831747#comment-17831747
]
Chia-Ping Tsai commented on KAFKA-16310:
> Yes, I agree with this method. But the question is, should we revert the fix
> in this JIRA? I think no because the fix doesn't change anything (no
> additional overhead at all) if we will iterate the batch when returning
> listOffset for MaxTimestamp in the end.WDYT?
We can do a little optimization if current fixes are kept. We do iterate
batches to find the q only if we don’t cache it. i.e the path of leader can
return the cached offsetOfMaxTimestanp. The other paths do not cache the
offsetOfMaxTimestanpq
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.8.0, 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17831741#comment-17831741
]
Luke Chen commented on KAFKA-16310:
---
{quote}Did you mean that we should change the schema to have a new field to
keep `offsetOfMaxTimestamp`? or just add a new method to Batch to iterate all
records to find out offsetOfMaxTimestamp?
{quote}
I think it's the former, otherwise, we're still iterating all records, and the
overhead for follower is unnecessary.
{quote}When serving the listMaxTimestamp request, we iterate the batch
containing the maxTimestamp to find the exact record offset with maxTimestamp.
Since this is a rare operation, paying the decompression overhead is fine.
{quote}
Yes, I agree with this method. But the question is, should we revert the fix in
this JIRA? I think no because the fix doesn't change anything if we will
iterate the batch when returning listOffset for MaxTimestamp. WDYT?
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.8.0, 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17831683#comment-17831683
]
Chia-Ping Tsai commented on KAFKA-16310:
{quote}I think the easiest way to fix the listMaxTimestamp issue is probably to
still maintain offsetOfMaxTimestamp at the record batch level so that it can be
derived consistently at both the leader and the follower.
{quote}
Did you mean that we should change the schema to have a new field to keep
`offsetOfMaxTimestamp`? or just add a new method to Batch to iterate all
records to find out offsetOfMaxTimestamp?
{quote}When serving the listMaxTimestamp request, we iterate the batch
containing the maxTimestamp to find the exact record offset with maxTimestamp.
Since this is a rare operation, paying the decompression overhead is fine.
{quote}
I gree to this solution. Furthermore, there are other works we can do for this
solution.
# catch maxTimestamp instead of maxTimestampAndOffsetSoFar
([https://github.com/apache/kafka/blob/trunk/storage/src/main/java/org/apache/kafka/storage/internals/log/LogSegment.java#L93])
since we do iterate all records to find out the offsetOfMaxTimestamp
# update time index by (max timestamp, last offset) even though the mapping is
not accurate. It should be fine since:
** it does not violate spec of time index
([https://github.com/apache/kafka/blob/trunk/storage/src/main/java/org/apache/kafka/storage/internals/log/TimeIndex.java#L36])
`Any message whose timestamp is greater than TIMESTAMP must come after OFFSET`
** the index returned from time index is used to look up the position of
batch, so it makes sense we append lastOffset to both time index and offset
index
In short, we do NOT trace offsetOfMaxTimestamp for all paths. The side effect
is that request to offsetOfMaxTimestamp can get slower ...
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.8.0, 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17831658#comment-17831658
]
Chia-Ping Tsai commented on KAFKA-16310:
{quote}
I think the easiest way to fix the listMaxTimestamp issue is probably to still
maintain offsetOfMaxTimestamp at the record batch level so that it can be
derived consistently at both the leader and the follower. When serving the
listMaxTimestamp request, we iterate the batch containing the maxTimestamp to
find the exact record offset with maxTimestamp. Since this is a rare operation,
paying the decompression overhead is fine. What do you think?
{quote}
I'm still digging in :)
BTW, it seems the recovering the segments has similar issue.
{code}
// The max timestamp is exposed at the batch level, so no need
to iterate the records
if (batch.maxTimestamp() > maxTimestampSoFar()) {
maxTimestampAndOffsetSoFar = new
TimestampOffset(batch.maxTimestamp(), batch.lastOffset());
}
{code}
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.8.0, 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17831631#comment-17831631
]
Manikumar commented on KAFKA-16310:
---
Thanks, I have reverted KAFKA-16341 and KAFKA-16342. commits from 3.6 branch
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.8.0, 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17831629#comment-17831629
]
Chia-Ping Tsai commented on KAFKA-16310:
{quote}
Can we revert the relavent changes from 3.6 branch as this not regression in
3.6.x releases and looks like complete fix requires few more changes. This is
to unblock 3.6.2 release.
{quote}
sure, please feel free to revert KAFKA-16341 and KAFKA-16342. Sorry for that
incomplete fix and it blocks the 3.6.2 :(
I will update KIP-734 to remind users that behavior of maxTimestamp offset
could be changed in the future release.
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.6.2, 3.8.0, 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17831624#comment-17831624
]
Manikumar commented on KAFKA-16310:
---
[~chia7712] [~showuon] Can we revert the relavent changes from 3.6 branch as
this not regression in 3.6.x releases and looks fix requires few more changes.
This is to unblock 3.6.2 release.
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.6.2, 3.8.0, 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17831553#comment-17831553
]
Jun Rao commented on KAFKA-16310:
-
[~johnnyhsu] , [~showuon] and [~chia7712] : Sorry, but I just realized one
issue with the fix. The problem is that we only fixed offsetForMaxTimestamp
during leader append. The follower append still uses the lastOffset in the
batch.
{code:java}
UnifiedLog.analyzeAndValidateRecords()
lastOffset = batch.lastOffset
...
if (batch.maxTimestamp > maxTimestamp) {
maxTimestamp = batch.maxTimestamp
offsetOfMaxTimestamp = lastOffset
} {code}
We optimize the follower code to avoid decompressing a batch. So, it's kind of
hard to get the exact record offset for maxTimestamp in the batch.
I think the easiest way to fix the listMaxTimestamp issue is probably to still
maintain offsetOfMaxTimestamp at the record batch level so that it can be
derived consistently at both the leader and the follower. When serving the
listMaxTimestamp request, we iterate the batch containing the maxTimestamp to
find the exact record offset with maxTimestamp. Since this is a rare operation,
paying the decompression overhead is fine. What do you think?
If we want to do the above, we probably need to revert the changes in 3.6.2,
which is being voted now. cc [~omkreddy]
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.6.2, 3.8.0, 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17831464#comment-17831464
]
Johnny Hsu commented on KAFKA-16310:
{quote}[~chia7712] thanks for the help!
This returns the offset and timestamp corresponding to the record with the
highest timestamp on the partition. Noted that we should choose the offset of
the earliest record if the timestamp of the records are the same.
This sounds good to me, thanks!
{quote}
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.6.2, 3.8.0, 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17831463#comment-17831463
]
Chia-Ping Tsai commented on KAFKA-16310:
@johnnyhsu thanks for offers the description. I add following statement to
KIP-734 according to your comments.
{quote}
This returns the offset and timestamp corresponding to the record with the
highest timestamp on the partition. Noted that we should choose the offset of
the earliest record if the timestamp of the records are the same.
{quote}
WDYT?
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.6.2, 3.8.0, 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17831460#comment-17831460
]
Johnny Hsu commented on KAFKA-16310:
The update is in below section:
## When the TimestampType is LOG_APPEND_TIME
When the TimestampType is LOG_APPEND_TIME, the timestamp of the records are the
same. In this case, we should choose the offset of the first record. [This
path|https://github.com/apache/kafka/blob/6f38fe5e0a6e2fe85fec7cb9adc379061d35ce45/storage/src/main/java/org/apache/kafka/storage/internals/log/LogValidator.java#L294]
in LogValidator was added to handle this case for non-compressed type, while
[this
path|https://github.com/apache/kafka/blob/6f38fe5e0a6e2fe85fec7cb9adc379061d35ce45/storage/src/main/java/org/apache/kafka/storage/internals/log/LogValidator.java#L421]
in LogValidator was added to handle this case for compressed type.
I don't have the Confluence account yet, [~chia7712] would you please help
update the KIP in the wiki? I will send this update to the dev thread for
visibility. Thanks!
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.6.2, 3.8.0, 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17830026#comment-17830026
]
Chia-Ping Tsai commented on KAFKA-16310:
will resolve this jira after [~johnnyhsu] update KIP-734
(https://cwiki.apache.org/confluence/display/KAFKA/KIP-734:+Improve+AdminClient.listOffsets+to+return+timestamp+and+offset+for+the+record+with+the+largest+timestamp)
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Chia-Ping Tsai
>Priority: Blocker
> Fix For: 3.6.2, 3.8.0, 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17823460#comment-17823460
]
Luke Chen commented on KAFKA-16310:
---
Created 2 sub-tasks for compressed and un-compressed records.
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Priority: Blocker
> Fix For: 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17823286#comment-17823286
]
Chia-Ping Tsai commented on KAFKA-16310:
{quote}
[https://github.com/apache/kafka/commit/e905ef1edfb92d8771e8c2a9a668f32210ad7e07]
is indeed the only relevant regression then. We should take the chance to fix
the regression and the previously existing issue.
{quote}
[~ijuma] oh, I pointed to incorrect ticket. Thanks for correcting, and sorry :-
{quote}
Thanks for the analysis. Ironically, my original approach was to fix the
compressed path to match the uncompressed path so that we passed along the
actual offset of max timestamp. I ended up doing the opposite though. If either
you or Luke Chen have time to submit a fix, I am happy to review.
{quote}
hi [~hachikuji] I have offline discussion with [~showuon], so we have time to
submit/backport the fix.
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Priority: Blocker
> Fix For: 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17823280#comment-17823280
]
Jason Gustafson commented on KAFKA-16310:
-
[~chia7712] Thanks for the analysis. Ironically, my original approach was to
fix the compressed path to match the uncompressed path so that we passed along
the actual offset of max timestamp. I ended up doing the opposite though. If
either you or [~showuon] have time to submit a fix, I am happy to review.
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Priority: Blocker
> Fix For: 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17823266#comment-17823266
]
Ismael Juma commented on KAFKA-16310:
-
Thanks for confirming that.
[https://github.com/apache/kafka/commit/e905ef1edfb92d8771e8c2a9a668f32210ad7e07]
is indeed the only relevant regression then. We should take the chance to fix
both.
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Priority: Blocker
> Fix For: 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17823238#comment-17823238
]
Chia-Ping Tsai commented on KAFKA-16310:
{quote}
It looks like the compressed case had the issue already, but we didn't go
further back to see if that has always been the case or not.
{quote}
I think the bug is already there before. I write the tests for both cases
(compression and non-compression), and both cases fails at trunk. By contrast,
only compression get failure at 3.4
see:
https://github.com/chia7712/kafka/blob/KAFKA-16310-3.4/core/src/test/scala/integration/kafka/api/OffsetOfMaxTimestampTest.scala#L52
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Priority: Blocker
> Fix For: 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17823208#comment-17823208
]
Ismael Juma commented on KAFKA-16310:
-
More recent changes to LogValidator:
[https://github.com/apache/kafka/commit/e905ef1edfb92d8771e8c2a9a668f32210ad7e07]
My understanding is that the following change caused the change to the non
compressed case. It looks like the compressed case had the issue already, but
we didn't go further back to see if that has always been the case or not.
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Priority: Blocker
> Fix For: 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17823205#comment-17823205
]
Ismael Juma commented on KAFKA-16310:
-
By the way, the bug attribution doesn't seem completely right. There were more
recent changes to LogValidator after the conversion to Java.
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Priority: Blocker
> Fix For: 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17823174#comment-17823174
]
Chia-Ping Tsai commented on KAFKA-16310:
[~ijuma] thanks for feedback. good to know we are on the same page :)
{quote}
I think one of them was going to submit a fix soon.
{quote}
Please feel free to assign this JIRA to us if you guys have no free cycle. It
seems the fix is straightforward as we can follow the behavior of Scala code.
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Priority: Blocker
> Fix For: 3.7.1
>
>
> Updated: This is confirmed a regression issue in v3.7.0.
> The impact of this issue is that when there is a batch containing records
> with timestamp not in order, the offset of the timestamp will be wrong.(ex:
> the timestamp for t0 should be mapping to offset 10, but will get offset 12..
> etc). It'll cause the time index is putting the wrong offset, so the result
> will be unexpected.
> ===
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17823151#comment-17823151
]
Luke Chen commented on KAFKA-16310:
---
Thanks! Please let us know when the PR is open.
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Cheng-Kai, Zhang
>Priority: Minor
>
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17822991#comment-17822991
]
Ismael Juma commented on KAFKA-16310:
-
Yes, [~jolshan] and [~hachikuji] looked into this and arrived to a similar
conclusion. I think one of them was going to submit a fix soon.
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Cheng-Kai, Zhang
>Priority: Minor
>
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-16310) ListOffsets doesn't report the offset with maxTimestamp anymore
[
https://issues.apache.org/jira/browse/KAFKA-16310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17822981#comment-17822981
]
Chia-Ping Tsai commented on KAFKA-16310:
the bug is related to https://issues.apache.org/jira/browse/KAFKA-16310
the code of updating `offsetOfMaxTimestamp` was:
{code:scala}
if (timestampType == TimestampType.LOG_APPEND_TIME) {
maxTimestamp = now
if (magic >= RecordBatch.MAGIC_VALUE_V2)
offsetOfMaxTimestamp = offsetCounter.value - 1
else
offsetOfMaxTimestamp = initialOffset
}
{code}
and now we always update `offsetOfMaxTimestamp` to `offsetCounter.value - 1`
even though it is TimestampType.LOG_APPEND_TIME (see
https://github.com/apache/kafka/blob/trunk/storage/src/main/java/org/apache/kafka/storage/internals/log/LogValidator.java#L300)
{code:java}
if (timestampType == TimestampType.LOG_APPEND_TIME) {
maxTimestamp = now;
offsetOfMaxTimestamp = initialOffset;
}
if (toMagic >= RecordBatch.MAGIC_VALUE_V2) {
offsetOfMaxTimestamp = offsetCounter.value - 1;
}
{code}
Furthermore, the path of compressed data has similar issue. see
https://github.com/apache/kafka/blob/trunk/storage/src/main/java/org/apache/kafka/storage/internals/log/LogValidator.java#L417
[~kevinztw] [~ijuma] Please double check the root cause, thanks!
> ListOffsets doesn't report the offset with maxTimestamp anymore
> ---
>
> Key: KAFKA-16310
> URL: https://issues.apache.org/jira/browse/KAFKA-16310
> Project: Kafka
> Issue Type: Bug
>Affects Versions: 3.7.0
>Reporter: Emanuele Sabellico
>Assignee: Cheng-Kai, Zhang
>Priority: Minor
>
> The last offset is reported instead.
> A test in librdkafka (0081/do_test_ListOffsets) is failing an it's checking
> that the offset with the max timestamp is the middle one and not the last
> one. The tests is passing with 3.6.0 and previous versions
> This is the test:
> [https://github.com/confluentinc/librdkafka/blob/a6d85bdbc1023b1a5477b8befe516242c3e182f6/tests/0081-admin.c#L4989]
>
> there are three messages, with timestamps:
> {noformat}
> t0 + 100
> t0 + 400
> t0 + 250{noformat}
> and indices 0,1,2.
> then a ListOffsets with RD_KAFKA_OFFSET_SPEC_MAX_TIMESTAMP is done.
> it should return offset 1 but in 3.7.0 and trunk is returning offset 2
> Even after 5 seconds from producing it's still returning 2 as the offset with
> max timestamp.
> ProduceRequest and ListOffsets were sent to the same broker (2), the leader
> didn't change.
> {code:java}
> %7|1709134230.019|SEND|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ProduceRequest (v7,
> 206 bytes @ 0, CorrId 2) %7|1709134230.020|RECV|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received ProduceResponse
> (v7, 95 bytes, CorrId 2, rtt 1.18ms)
> %7|1709134230.020|MSGSET|0081_admin#producer-3|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2:
> rdkafkatest_rnd22e8d8ec45b53f98_do_test_ListOffsets [0]: MessageSet with 3
> message(s) (MsgId 0, BaseSeq -1) delivered {code}
> {code:java}
> %7|1709134235.021|SEND|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Sent ListOffsetsRequest
> (v7, 103 bytes @ 0, CorrId 7) %7|1709134235.022|RECV|0081_admin#producer-2|
> [thrd:localhost:39951/bootstrap]: localhost:39951/2: Received
> ListOffsetsResponse (v7, 88 bytes, CorrId 7, rtt 0.54ms){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
