Would not be the case slightly decrease the shared buffer, lower your checkpoints_timeout for eg 5 minutes or decrease checkpoints_segments and set the checkpoint_completation_target to 0.5 to not mess up the next checkpoints?
What logs tell me is that a checkpoint occurs immediately to the other.
When you have a large shared_buffer and checkpoint_completation_target value near the maximum time between checkpoints, which I imagine is very spread given for a very long time interval . As your shared_buffer is relatively large, more data should be handled by the checkpoint. The checkpoints are being triggered by time (15 minutes) accumulating a good amount of data.
All this happens every time you load your CVS, and when the writing process of checkpoints reaches the end occurs immediately another.
It seems reasonable high disk activity performed by checkpoint and writing processes, performing theirs job duties amid a high load on the shared_buffer.
