There are multiple problems with disabling registration when load average reached a static magic number such as 5.
-
If the machine has 128 cores, it is perfectly normal to have a load average of 64. The installer should count the number of cores and may consider to use a multiple of that by default.
-
Virtual machines and commercial VPS can sometimes utilize containerization technology that shares the kernel with other services for improved efficiency, such as OpenVZ and LXC. In this case, you may sometimes see the loadavg for the host, not for your 1 or 2-core virtual CPU (unless you know they are using
lxcfs -l). -
Such servers can also be hardened in varied ways to provide a stable service level availability for each tenant, you could easily see the loadavg go up to the hundreds on a 16-core host even if your VPS seems quite responsive (while processed owned by the offending tenant would be crawling to a halt of course).
-
According to Wikipedia, Linux counts each thread individually of a multi-threaded process and includes processes which are running, ready to run and also those queued for uninterruptible sleep such as when doing I/O. Note that if one of your neighbor is executing buggy code, this could even happen when no active I/O is happening.
-
The operator should have an option to disable each of these checks. This is done in various ways, such as unassigning it, assigning it the empty string, infinity,
-1,0. No such outlier value is supported at present, so the only workaround is to assign it a very high “infinite” constant such as9999.
If they are reached, users can’t create new addresses on your relay:
max_load_1m = 5