It looks to me like it’s triggered when requests are queued for too long - in other words, requests coming in faster than they are serviced. One might wonder why so many requests, or why such slow service. At the Discourse level, there are tunables which are already discussed in this thread and also, for example, in Extreme load error.
At the Linux level, I would check
uptime
free
vmstat 5 5
ps auxrc