I had some pretty similar issues about a week ago. It would really kick off when I was receiving notifications. But sometimes would happen without.
Same solution here, I had to update some configs in my php-fpm to allow for more resources. Below is what I ended up setting. It's been rock solid since I made the change:
'pm.max_children = 60''pm = dynamic''pm.start_servers = 8''pm.min_spare_servers = 8''pm.max_spare_servers = 15''pm.max_requests = 500'