Replies: 10 comments
-
感觉这个问题挺有挑战性,我也遇到类似的场景在想办法解决。 @yuyii 你的这种情况,开启了GTID没有,用了GTID的启动模式吗? |
Beta Was this translation helpful? Give feedback.
-
有解决方案了,踢一下我 |
Beta Was this translation helpful? Give feedback.
-
增加线程数也不可以吗,是不是遇到了带宽瓶颈? |
Beta Was this translation helpful? Give feedback.
-
这是来自QQ邮箱的自动回复邮件。你好,邮件正常收到,谢谢! -----万玉坤
|
Beta Was this translation helpful? Give feedback.
-
感觉是带宽不够,可以观察一下流量 |
Beta Was this translation helpful? Give feedback.
-
这是来自QQ邮箱的自动回复邮件。你好,邮件正常收到,谢谢! -----万玉坤
|
Beta Was this translation helpful? Give feedback.
-
增量同步只能一个进程, |
Beta Was this translation helpful? Give feedback.
-
这是来自QQ邮箱的自动回复邮件。你好,邮件正常收到,谢谢! -----万玉坤
|
Beta Was this translation helpful? Give feedback.
-
应该是带宽不够 cdc太消耗带宽了 不知道有没解决方案解决 |
Beta Was this translation helpful? Give feedback.
-
这是来自QQ邮箱的自动回复邮件。你好,邮件正常收到,谢谢! -----万玉坤
|
Beta Was this translation helpful? Give feedback.
-
在公司生产环境中使用flink-cdc 3.0.1,单机模式跑从mysql到starrocks的同步,不考虑全量的情况,做增量的场景发现无论怎么设置增量同步都会延迟,binlog的增长高峰期大约在100~150w/s以上(有个日志埋点库也在一起),开的debug模式查的日志发现flink的source处理binlog最高只能达到60w/s,这里只是根据binlog的增长值来看的。如果在半夜binlog产生量小的情况,可以正常同步,但到白天基本就会疯狂延迟,可以到8小时以上。配置上做了表过滤,数据库过滤等,真实数据大概每秒不到100条的。尝试了如下各种设置:
schema-change.enabled: false
debezium.database.include.list: "test_db"
debezium.snapshot.mode: "never"
debezium.max.batch.size: 404096000
debezium.max.queue.size: 1001920000
debezium.include.schema.changes: false
flink 的task memory给到了20G,实际测下来发现和上述设置并没关系,只是一度维持在60w/s的binlog处理。flink的进程资源毫无压力,都是0。我想请教如何能加快速度来消费,看了debezium的文档也没找到合适的配置。或者有办法开启多个进程或者线程同时同步么?目前3.0的版本是只有一个进程作为source来获取binlog的,也没找到可以增加的地方。
Beta Was this translation helpful? Give feedback.
All reactions