[IOTDB-829] Accelerate delete timeseries

[Wei Shao <ko...@163.com>]

Hi, 


I am currently working on JIRA issue: [1]. 


When deleting data from multiple timeseries, a successful deletion will leave a record in mods file for each timeseries.  
Suppose we executing "delete from root.sg1.d1.*" or "delete timeseries root.sg1.d1.* " when d1 contains N measurements.
We would write at most N lines of records in each mods file, however, this may cause performance issue if N is very large.


My optimization plan is:
In Mods files, we record exact the same path string that user specified. For example, "delete from root.sg1.d1.sd1.*" will write "root.sg1.d1.sd1.*" into mods file as path. 
In this way, for each timeseries path in delete statement, we access each Mods file once at most.
As a tradeoff, a path string matching is needed during query, to check if the timeseries has any modifications in mods files.


Any suggestions are welcome.


Thanks,


[1] https://issues.apache.org/jira/projects/IOTDB/issues/IOTDB-829?


---------------------
Wei Shao