Hi, all
I submitted a pr to solve the problem of when to delete the raft log 
(https://github.com/apache/incubator-iotdb/pull/1339)
issue: https://issues.apache.org/jira/browse/IOTDB-723
The main idea is as follows:
Combine time-based method and number-based method together. Each server 
periodically triggers the operation to check the deletion of the log. If the 
distance between the last deletion log index and  the current maximum committed 
index exceeds a certain threshold, it performs the deletion operation, 
otherwise do nothing.
The reason for keeping a certain length of log between two deletions is to 
avoid installing snapshot on slow replica.
please leave your opinions, thanks.

______________________________

Thanks,
Houliang Qi

Reply via email to