Steps to reproduce:

1. create a logfile monitor item logrt，eg：logrt["/home/rkmoner/procmon//agent_procmon.log",\{$GD_REGEXP_PATTERN},\{$GD_CHARSET},1000,skip,,,mtime-noreread] ,interval 1m
2. exec sql select * from item_rddata to check current logsize and mtime, eg:SELECT FROM_UNIXTIME(t.mtime, '%Y-%m-%d %H:%i:%S'),t.* FROM `item_rtdata` t  where t.itemid=45509
3. vi agent_procmon.log keep it a few minutes，dont modify any things...
4. check item_rddata ，mtime and last file already incorrect.

Result:
 it lead to agent reread all log context and alert a lots alarm
Expected:
See screenshot....
See attached patch file...