English 中文(简体)
B. 结 论
原标题:Resolve Flink SQL windows tumble performance drop

我有一个连接板块应用,把Kafka的数据整理成实时的数据库。 这些数据按年数汇总:[1、5、10、30、60、每日],并写到相应的非行表格。

我将水mark功能用于综合数据,如数据。 由TUMBLE(ts,InterVAL 5 MINUTES)组成的小组

table.exec.state.ttl = 24 h
table.exec.sink.not-null-enforcer = DROP
table.exec.emit.early-fire.enabled = true
table.exec.emit.early-fire.delay = 500 ms
table.optimizer.agg-phase-strategy= TWO_PHASE
table.exec.source.idle-timeout = 5000 ms

我以4个平行主义和2个平行的科索沃警察部队(总共2个科索沃邮联)为对象。 卡夫卡专题分四部分。

最初1.5分期的申请如期进行,数据延迟增加,一些分类结果缺失,这种情况在日常重复出现,数据数量和数据也相同。

从这些衡量标准中,我发现,电离线输入信息的趋势与卡夫卡在运行1.5hr(11)之后的输入信息不一致。

flink input vs kafka input flink input vs kafka input

flink input vs task resources flink input vs task resources

task status when output normally task status when output normally

Task:
1) Input from kafka
2) tumble into 1 min
3) Left join extra info to the 1 min result
4) tumble into 1, 5 min
5) Group and write to DB into 1,5,10,15,30,60 min

task status when output unexpected enter image description here

我改变了下文,局势仍然存在。

  1. Change a better instance type of database
  2. Change a better instance type of Kafka cluster
  3. Set 4 parallelism and 1 parallelism per KPU(4 KPU in total)

我很想知道造成这种情况的荒谬原因,以及解决这一问题的建议,感谢你。

问题回答

你们应当使用温饱电视基金,而不是假装的遗产小组Wow Aggregations。 为了提高业绩,这是优化的。 你们可以找到有关docs子和组合的细节,载于 docs . ?

*** https://weeklies.apache.org/flink/flink-docs-master/docs/dev/table/sql/queries/window-tvf/





相关问题
How to Process Events from now() to -10 minutes in Flink

I want to create a Stream Flink app that will process the events that falls within the latest event timestamp to different time windows, e.g. 10 minutes, 1 day and 1 month. So lets say I got an event ...

Flink Batch vs Stream how they process real time data

I have read the documentation of streaming mode and batch mode. I assume that if I have an unbounded stream and I apply windows (like tumbling) on it it becomes a bounded stream? Please correct and ...

Flink failed initials checkpoints

I have a flink job deployed on a local kind cluster, it saves checkpoints to AWS S3. The following error kept occurring in job manager log at the initial stage: 2023-07-07 19:33:48,657 INFO org....

RocksDB state backend - WhenTableAPI used

We are using TableAPI to execute few SQL queries to achieve the desired results. Whehn creating Windows out of the streams, we observed OutOfMenmory issues. We enabled RocksDB as the state backend in ...

flink on k8s How to set the minimum and maximum CPUs

I am using flink on k8s, and now there is a problem, the dynamic resource feature of k8s cannot be utilized, one of my flink tasks only occupies a small amount of CPU resources when it runs normally, ...

热门标签