看下你的 flink 命令对不对,然后去 Flink Web UI Checkpoint 界面,看下是否从 Savepoint 恢复(下面有个 restore path). 之后再看下你的窗口时间类型用的是什么。
Best, LakeShen 王春浩 <wang.chun...@navercorp.com> 于2021年5月27日周四 上午9:26写道: > hi, 社区 > > 版本flink 1.7 > > > 我正在尝试从保存点(或检查点)还原flink作业,该作业的工作是从kafka读取->执行30分钟的窗口聚合(只是AggregationFunction,就像一个计数器)->下沉到kafka。 > > 我使用rocksdb和启用检查点。 > > 现在我尝试手动触发一个保存点。 每个汇总的期望值是30(1个数据/每分钟)。 但是,当我从保存点还原时(flink运行-d -s > {savepoint的url}),聚合值不是30(小于30,取决于我取消flink作业并还原的时间)。 但是当作业正常运行时,它将达到30。 > > 我不知道为什么有些数据似乎会丢失? > > 日志显示``No restore state for FlinkKafkaConsumer'' > > > > 四川省成都市高新区天府三街199号 太平洋保险金融大厦A区11楼 > 11th Floor,Tower A,Pacific insurance finance Building, No.199 TianFu 3rd > Street, GaoXin District, Chengdu, Sichuan Province > Mobile +86 15817382279 > Email wangchun...@navercorp.com > > NCloud > > -----Original Message----- > From: "王春浩"<wang.chun...@navercorp.com> > To: <u...@flink.apache.org>; > Cc: > Sent: 2021/5/26周三 17:03 (GMT+08:00) > Subject: inquire about restore from savepoint > > Hi Community, > > version flink 1.7 > im trying to make a flink job restore from a savepoint(or checkpoint), > what the job do is reading from kafka -> do a 30-minutes-window > aggregation(just AggregationFunction, acts like a counter) -> sink to kafka. > i use rocksdb and enabled checkpoint. > now i try to trigger a savepoint manually. the expected value of each > aggregated one is 30(1 data/per minute). but when i restore from a > savepoint(flink run -d -s {savepoint's url}), the aggregated value is not > 30(less than 30, depends on the time i cancel flink job and restore). but > when the job run normally, it gets 30. > i don't know why could some data seems to be lost? > and a log shows "No restore state for FlinkKafkaConsumer" > > > > 四川省成都市高新区天府三街199号 太平洋保险金融大厦A区11楼 > 11th Floor,Tower A,Pacific insurance finance Building, No.199 TianFu 3rd > Street, GaoXin District, Chengdu, Sichuan Province > Mobile +86 15817382279 > Email wangchun...@navercorp.com > > NCloud >