rd_refcnt_sub0: Assertion `!*"refcnt sub-zero"' failed - when the subscribed topic is deleted #2963
Milestone
Comments
|
#0 0x00007fb188359387 in __GI_raise (sig=sig@entry=6) at ../nptl/sysdeps/unix/sysv/linux/raise.c:55 |
|
Any updates on this @edenhill ? Is there anything else I can provide to you that would help? |
|
Make sure that you are destroying each Message and Topic object exactly once. |
|
I am creating TopicPartition using |
|
Can you share your code? |
|
I'll give you a minimal reproducer by tomorrow. |
|
I am able to reproduce this using th rdkafka_complex_example_cpp itself. After the topic gets deleted, I was getting the error From core: |
|
Is this with a completely unmodified v1.4.2 rdkafka_complex_consumer_example ? Can you try to reproduce on master? |
|
Yes, in master also its reproducible. rdkafka_complex_consumer_example_cpp with no modification. |
|
Can you reproduce this with ASAN? (build librdkafka with |
|
I don't have libasan installed. I'm restricted from installing it due to regulations. Are you unable to reproduce the error with the master ? Please let me know if I can do anything else instead. |
|
Hi, Any updates here ? |
|
Ping |
|
Sorry, but I'm not able to reproduce this and I don't have much else to go on if it can't be run in a debugger/valgrind/asan. |
|
If this is not reproducible with the same version of the rdkafka example, could it be because of different versions of the kafka broker ? Also, please note that I have given the output of the debugger (gdb) |
|
What is the broker version? |
|
Kafka broker version - 2.5.0. To make sure, I just had this checkout, started kafka brokers, topic and producer. The failure is not immediate. I was getting it after 4-5 minutes. |
|
Can't reproduce on master with AK 2.5.0. |
|
Ran with |
|
After doing Produced few messages Started consumer and left it running
From consumer output I ain't doing anything else. |
|
Before using the example kafka, did this Let me know if I missing anything. |
|
This is on librdkafka master with AK 2.5.0: No crash. |
|
Any thoughts on how to proceed ? Also, please see if I am using asan in right way. |
|
It looks like it does build with asan correctly (the weird compiler bug can be ignored since that is in a test). |
|
Do you think you could make a docker image with the client so that we're testing the exact same build? Please leave the source in the docker image |
|
I have uploaded the docker image, the docker file used and the consumer output here |
|
Hi, let me know if the docker was useful. |
|
Hi, Any updates here ? |
|
Ping |
|
Thank you! The docker build reproduces the issue and I'll start looking into it 👍 |
edenhill
added a commit
that referenced
this issue
Aug 20, 2020
…2963) .. for consumer errors that originate from light-weight topic objects.
edenhill
added a commit
that referenced
this issue
Aug 25, 2020
…2963) .. for consumer errors that originate from light-weight topic objects.
edenhill
added a commit
that referenced
this issue
Aug 25, 2020
…2963) .. for consumer errors that originate from light-weight topic objects.
|
Fixed on master. Thanks @dshivashankar1994 for putting in all the effort to provide a reproducible test case, very valuable! |
7 tasks
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
I am using a cpp version of librdkafka consumer (version 1.4.2)
I am subscribe to 2 topics that are existing. They got assigned fine and was receiving messages.
Now when I delete one of the topic, I was expecting 2 things
(1) consumer.consume to throw the UNKNOWN_TOPIC_OR_PART error
(2) Rebalance cb to revoke that partition
But both are not happening and instead I get the assertion error.
Logs in debug mode:
Topics are team.topic.test.abc-1 and team.topic.test.abc-2.
Deleted topic - team.topic.test.abc-1
The text was updated successfully, but these errors were encountered: