wangchao316 commented on pull request #2597: URL: https://github.com/apache/iotdb/pull/2597#issuecomment-771508916
> Hi, Current description of the PR is clear. > But IMO, the `regular` encoding is more suitable for the timestamp column... how do you think? (is there some cases that the value column obeys the pattern?) > > > sample: 1000，1100，1800，1400，1500... > > current algorithm be do not use... > > 1800 is a error point, we should identity error point, revise data. > > revise data should be : 1000，1100，1300，1400，1500 > > This is ... dangerous, as we modify the raw data directly and do not tell the user in time. @jixuan1989 Thanks review. （ 是的，定频编码适合时间戳。 目前因为一个时间戳数据错误就导致整个ts flush失败，从而开启只读模式。 我觉得我们应该做一些事保证一个服务的稳定性。 因为 1. 我们不能确定用户指定了定频编码，就能按照时间戳数据顺序插入。 2.如果插入这种数据，需要有容错能力，最起码不能影响同一个flush中的其他序列。有以下几个方法： （2.1） 要么修改这个数据。 （2.2） 要么删除这个数据。 （2.3） 或者报错，但是此时数据已经写了wal,和内存，只是在flush的时候出错，所以是否需要清理该数据wal. 我目前的解决办法是： 1. 尝试修复定频数据，并打印警告日志。 2.如果不能修复，输入的不是定频数据，那么我们抛出警告异常，然后删除错误数据，并且在日志中打印出来。） Yes, constant frequency coding is suitable for timestamps. However, the entire TS flush cannot fail because of a timestamp data error. In this case, the read-only mode cannot be enabled. Because 1. If you cannot determine whether the code is specified by the user, the data can be inserted in the sequence of timestamp data. 2. If such data is inserted, the error tolerance capability is required. (2.1) Modify the data. (2.2) Delete the data. (2.3) An error is reported. However, the data has been written to the wal and memory, and an error occurs only during flushing. Therefore, do you need to clear the wal? My current solution is: 1. Restore the constant-frequency data. 2. If the fault cannot be rectified and the input data is not constant-frequency data, a warning exception is thrown, the error data is deleted, and the error data is printed in the log. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
