[
https://issues.apache.org/jira/browse/KAFKA-6446?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16343150#comment-16343150
]
Yu LIU edited comment on KAFKA-6446 at 1/29/18 9:59 AM:
--------------------------------------------------------
Hello,
I have a look at this and it appears that TRANSACTION_TIMEOUT_CONFIG is defined
as the time to wait when *a producer tries to get the producer id from
transaction coordinator by specifying transaction id*, more concretely it's for
[InitProducerIdRequest|https://github.com/apache/kafka/blob/trunk/clients/src/main/java/org/apache/kafka/common/requests/InitProducerIdRequest.java].
But where we hang here is before InitProducerIdRequest, we need to find the
transaction coordinator by sending a
[FindCoordinatorRequest|https://github.com/apache/kafka/blob/trunk/clients/src/main/java/org/apache/kafka/common/requests/FindCoordinatorRequest.java],
in this phase we didn't deal with timeout so it leads to the [Sender thread
endless retrying to send
FindCoordinatorRequest|https://github.com/apache/kafka/blob/trunk/clients/src/main/java/org/apache/kafka/clients/producer/internals/Sender.java#L161-L167].
I think anyway the producer shoud not be stuck in initTransactions() when it
can't connect to the kafka cluster ? We should define a new TIMEOUT for
connecting to the cluster (FindCoordinatorRequest) or applying the
TRANSACTION_TIMEOUT_CONFIG to it directly ?
was (Author: pnpie):
Hello,
I have a look at this and it appears that TRANSACTION_TIMEOUT_CONFIG is defined
as the time to wait when *a producer tries to get the producer id from
transaction coordinator by specifying transaction id*, more concretely it's for
[InitProducerIdRequest|https://github.com/apache/kafka/blob/trunk/clients/src/main/java/org/apache/kafka/common/requests/InitProducerIdRequest.java].
But where we hang here is before InitProducerIdRequest, we need to find the
transaction coordinator by sending a
[FindCoordinatorRequest|https://github.com/apache/kafka/blob/trunk/clients/src/main/java/org/apache/kafka/common/requests/FindCoordinatorRequest.java],
for this phase we didn't treat timeout case so it leads to the [Sender thread
endless retrying to send
FindCoordinatorRequest|https://github.com/apache/kafka/blob/trunk/clients/src/main/java/org/apache/kafka/clients/producer/internals/Sender.java#L161-L167].
So I think anyway the producer shoud not be stuck in initTransactions() when it
can't connect to the kafka cluster ? We should to define a new TIMEOUT for
connecting to the cluster (FindCoordinatorRequest) or applying the
TRANSACTION_TIMEOUT_CONFIG to it directly ?
> KafkaProducer with transactionId endless waits when bootstrap server is down
> ----------------------------------------------------------------------------
>
> Key: KAFKA-6446
> URL: https://issues.apache.org/jira/browse/KAFKA-6446
> Project: Kafka
> Issue Type: Bug
> Components: clients, producer
> Affects Versions: 0.11.0.0, 1.0.0
> Reporter: Eduardo Sciullo
> Priority: Critical
> Attachments: Test.java
>
>
> When bootstrap server is down, a KafkaProducer with transactionId endless
> waits on initTransactions.
> The timeouts don't apply to that operation: don't honor the
> {{TRANSACTION_TIMEOUT_CONFIG.}}
> Attached an example of my code to reproduce the scenario.
>
> I opened this issue as suggested by [Gary
> Russell|https://stackoverflow.com/users/1240763/gary-russell]
> [https://stackoverflow.com/questions/48226546/defaultkafkaproducerfactory-with-transactionidprefix-endless-waits-when-bootstra]
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
