jinshuangxian created FLINK-28250:
-------------------------------------
Summary: exactly-once sink kafka cause out of memory
Key: FLINK-28250
URL: https://issues.apache.org/jira/browse/FLINK-28250
Project: Flink
Issue Type: Bug
Affects Versions: 1.15.0
Environment: *flink version: flink-1.15.0*
*tm: 8* parallelism, 1 slot, 2g
centos7
Reporter: jinshuangxian
Attachments: image-2022-06-25-22-07-35-686.png,
image-2022-06-25-22-07-54-649.png, image-2022-06-25-22-08-04-891.png,
image-2022-06-25-22-08-15-024.png
*my sql code:*
CREATE TABLE sourceTable (
data bytes
)WITH(
'connector'='kafka',
'topic'='topic1',
'properties.bootstrap.servers'='host1',
'properties.group.id'='gorup1',
'scan.startup.mode'='latest-offset',
'format'='raw'
);
CREATE TABLE sinkTable (
data bytes
)
WITH (
'connector'='kafka',
'topic'='topic2',
'properties.bootstrap.servers'='host2',
'properties.transaction.timeout.ms'='30000',
'sink.semantic'='exactly-once',
'sink.transactional-id-prefix'='xx-kafka-sink-a',
'format'='raw'
);
insert into sinkTable
select data
from sourceTable;
*problem:*
After the program runs online for about half an hour, full gc frequently appears
{*}Troubleshoot{*}：
I use command 'jmap -dump:live,format=b,file=/tmp/dump2.hprof' dump the problem
tm memory. It is found that there are 115 FlinkKafkaInternalProducers, which is
not normal.
!image-2022-06-25-22-07-54-649.png!!image-2022-06-25-22-07-35-686.png!
After reading the code of KafkaCommitter, it is found that after the commit is
successful, the producer is not recycled, only abnormal situations are recycled.
!image-2022-06-25-22-08-04-891.png!
I added a few lines of code. After the online test, the program works normally,
and the problem of oom memory is solved.
!image-2022-06-25-22-08-15-024.png!
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
