You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
THERE IS NO OBJECT LEAK - program to re-create the issue attached here.
RdKafka::Consumer destructor thread hangs due to broker thread rkb->rkb_refcnt being NOT zero.
This happens during kafka cluster re-starts.
Operations like reading the last msg, query_watermark_offsets done, while Kafka cluster is being killed & re-started.
Set environment variables KAFKA_INC_PATH(point to librdkafka header files containing directory - both C and C++) and KAFKA_LINK_PATH(point to librdkafka shared libs containing directory)
type 'cmake ..'
type 'make'
This will build the binary LibrdKafkaIssueTest.
How to Run
Create 2 Kafka topics with topic partition count=4 for each.(3 broker cluster with replication factor=2 used for my re-creation)
Keep one topic empty and write a single message to each of the partition in other topic
Use bash command to run Kafka cluster kill-restart loop : 'while true; do [stop_the_kafka_cluster]; sleep 10; [re-start-kafka-cluster-WITHOUT-cleaning-data]; sleep 180; done'
Invoke LibrdKafkaIssueTest as :
'LibrdKafkaIssueTest 175 [kafka boost trap servers] [empty topic] [topic with 1 msg per partition] 0'
Run tool for about 10 minutes.
When issue is re-created it will show msg (in Color RED), 'Thread Index=[index] is stuck in Consumer Destroy for [stuck time]s'.
Checklist
librdkafka version (release number or git tag): 1.9.2 (issue present in 1.6 too)
Apache Kafka version: 2.7.0
librdkafka client configuration:All default values used
Operating system: Red Hat 8.4 x64 bit (Issue re-created on Red Hat Linux 7.9 too)
[] Provide broker log excerpts : Log sizes are large, hence NOT attached. Please let me know if this is essential, I will re-create the issue and attach all logs
Critical issue : No
The text was updated successfully, but these errors were encountered:
aKumara123
changed the title
(WITHOUT object leak) RdKafka::Consumer destructor hangs forever with kafka cluster re-starts taking place
(WITHOUT object leak) RdKafka::Consumer destructor hangs forever (when kafka cluster re-starts taking place)
Jan 3, 2023
Description
THERE IS NO OBJECT LEAK - program to re-create the issue attached here.
RdKafka::Consumer destructor thread hangs due to broker thread rkb->rkb_refcnt being NOT zero.
This happens during kafka cluster re-starts.
Operations like reading the last msg, query_watermark_offsets done, while Kafka cluster is being killed & re-started.
How to reproduce
Sample program attached.
How to Compile
In a Linux environment :
This will build the binary LibrdKafkaIssueTest.
How to Run
Create 2 Kafka topics with topic partition count=4 for each.(3 broker cluster with replication factor=2 used for my re-creation)
Keep one topic empty and write a single message to each of the partition in other topic
Use bash command to run Kafka cluster kill-restart loop : 'while true; do [stop_the_kafka_cluster]; sleep 10; [re-start-kafka-cluster-WITHOUT-cleaning-data]; sleep 180; done'
Invoke LibrdKafkaIssueTest as :
'LibrdKafkaIssueTest 175 [kafka boost trap servers] [empty topic] [topic with 1 msg per partition] 0'
Run tool for about 10 minutes.
When issue is re-created it will show msg (in Color RED), 'Thread Index=[index] is stuck in Consumer Destroy for [stuck time]s'.
Checklist
librdkafka version (release number or git tag):
1.9.2
(issue present in 1.6 too)Apache Kafka version: 2.7.0
librdkafka client configuration:All default values used
Operating system:
Red Hat 8.4 x64 bit
(Issue re-created on Red Hat Linux 7.9 too)Provide logs : attached
-- gdb.txt
-- librd kafka log output librdkafka_output.zip - Kindly Please note that IP and Port names are substituted.
-- stack stack_3323083.zip
-- stderr output : NOTE librdkafka was built with ENABLE_REFCNT_DEBUG stderr_LibrdKafkaIssueTest_1.zip
[] Provide broker log excerpts : Log sizes are large, hence NOT attached. Please let me know if this is essential, I will re-create the issue and attach all logs
Critical issue : No
The text was updated successfully, but these errors were encountered: