Radiator Server Documentation — v10.34.2

stats Context Variables

The stats namespace provides read-only access to process and system statistics including memory usage, CPU time, and load averages. These can be used in configuration expressions, logging, and attribute assignments.

Process Statistics

stats.process_rss_bytes

Process resident set size (memory) in bytes.

Type: Integer (unsigned)

Example:

logging {
    aaa {
        logger "METRICS" {
            console;
        }
    }
}

aaa {
    @pipeline "LOG_MEMORY" {
        log "METRICS" {
            format "rss_bytes=%{stats.process_rss_bytes}";
        }
    }
}

stats.process_cpu_millis

Total CPU time used by the process in milliseconds (user + system).

Type: Integer (unsigned)

System Memory

stats.system_mem_total_bytes

Total system memory in bytes.

Type: Integer (unsigned)

stats.system_mem_available_bytes

Memory in bytes that the system can give to new allocations without swapping. This includes page cache and other memory that the kernel can reclaim. The value is MemAvailable from /proc/meminfo. Kernels that do not report MemAvailable report unused memory instead.

Type: Integer (unsigned)

stats.system_swap_total_bytes

Total swap space in bytes.

Type: Integer (unsigned)

stats.system_swap_used_bytes

Used swap space in bytes.

Type: Integer (unsigned)

stats.swap_pages_in

Cumulative count of pages swapped in from disk.

Type: Integer (unsigned)

stats.swap_pages_out

Cumulative count of pages swapped out to disk.

Type: Integer (unsigned)

System CPU

stats.system_cpu_count

Number of CPUs available.

Type: Integer (unsigned)

stats.system_cpu_active_millis

System-wide active (non-idle) CPU time in milliseconds.

Type: Integer (unsigned)

stats.system_cpu_total_millis

System-wide total CPU time in milliseconds.

Type: Integer (unsigned)

Calculating CPU utilization: To calculate CPU utilization percentage, compare stats.system_cpu_active_millis to stats.system_cpu_total_millis:

CPU % = (stats.system_cpu_active_millis / stats.system_cpu_total_millis) * 100

For interval-based measurements, take two snapshots and compute the delta:

CPU % = ((active2 - active1) / (total2 - total1)) * 100

Load Averages

Load averages are multiplied by 100 to preserve precision as integers. Divide by 100 to get the actual load average value.

stats.load_1m_x100

1-minute load average multiplied by 100.

Type: Integer (unsigned)

stats.load_5m_x100

5-minute load average multiplied by 100.

Type: Integer (unsigned)

stats.load_15m_x100

15-minute load average multiplied by 100.

Type: Integer (unsigned)

Network Queues

These values help detect overload.

stats.udp_receive_drops

Datagrams that the kernel dropped because the receive buffers of Radiator's open UDP sockets were full. Each socket counts from the time it was opened, so the value can decrease when a socket is closed.

Type: Integer (unsigned)

stats.tcp_listen_overflows

Number of times a TCP listen queue on the host overflowed since the host started. The value covers all applications on the host. Only available on Linux.

Type: Integer (unsigned)

stats.tcp_listen_drops

Connection attempts dropped by TCP listeners on the host since the host started, including listen queue overflows. The value covers all applications on the host.

Type: Integer (unsigned)

Example Usage

Logging Memory Usage

scripts {
    lua "LOG_STATS" {
        content """
            local context, previous = ...
            local rss = context.stats.process_rss_bytes
            if rss then
                print(string.format("Memory: %.2f MB", rss / 1024 / 1024))
            end
            return previous
        """;
    }
}

aaa {
    policy "RADIUS" {
        conditions all {
            aaa.protocol == protocol.RADIUS;
        }

        handler "AUTH" {
            @authentication {
                script "LOG_STATS";
            }
        }
    }
}