Forum OpenACS Development: Re: NaviServer "breaks" under high load
as a reference, we haven on learn@wu on the year average ~10k different users per day.
The "hang" occurred during re-spawning threads
Probably the server was not "hanging", but just hopelessly overloaded; if there is a single connection pool in use, a fresh request has to wait, until all waiting jobs are processed. In such situations, many users press eagerly reload, worsening the situation even more, and nobody sees finishing pages; even worse, with a single connection pool, even formerly fast requests (also static files) have to wait for the slow requests - so multiple connection pools are highly recommended for sites with high load and slow requests. The message in the error log said that the server would like to create more threads, but it was already at max.
You will need more db-connections: when using OpenACS most likely every active connection thread will require at least one DB connection, same is true for background processing. So, with 32 connection threads, you would need 32+ db connections for pool1...
just hopelessly overloaded
Nope! I had top and atop running exactly at that moment, and the load was pretty normal (some 150% CPU, out of 800% available), and atop didn't show too many I/O operations neither.
There was also no noticeable slow-down of the server in the 30-60 seconds before the "hang". I was monitoring performance using your excellent /request-monitor/...
I'll let you know how things are going next Friday!
Thanks a lot!