Your best bet might be to directly go to Snowflake and ask for its long running, or frequent running queries for the username Replicate is using. This will not only identify but likely give you input as to how to fix it.
To investigate through Replicate you need to enable LOGGING for TARGET_APPLY to TRACE (DEBUG)
Now run for a few hours until you 'caught' a slowdown window.
Lower Logging level, Roll the log, start investigating. What you are looking for is the lines with "from seq" which tells you Replicate is about to do the action on the target. On the left is the THREAD ID and 'start time'. Look for the next occurrence of that thread id which will be the end time, after the action. Subtract start from end to get a duration on the action.
Now this can become tedious so years ago i scripted this using PERL that this is readily available on Linux as well as Windows. Similar scripts can of course be done in Python or Powershell.but for me the Perl works so why bother.
Anyway, script attached run with -h for help and typically with -v and/or -t for verbose, per table outputs
Hein
