Read Apache busy and idle worker metrics

Apache's busy workers are occupied worker slots; idle workers are available to take work. The metrics help identify pressure on request-processing capacity, but they are not a direct count of connected clients. Interpret them alongside the active multi-processing module (MPM), traffic, and latency.

Read a restricted status endpoint

With mod_status loaded, expose its handler only to trusted monitoring clients:

<Location "/server-status">
    SetHandler server-status
    Require ip 127.0.0.1 ::1 192.0.2.0/24
</Location>

Replace the documentation subnet with the actual management range; do not leave a public status endpoint. Apply the location in the intended server or virtual-host context and check proxy access paths too. Detailed status can expose request URLs, client addresses, and virtual-host information.

For machine-readable metrics:

curl -fsS 'http://127.0.0.1/server-status?auto'

A sample includes:

BusyWorkers: 47
IdleWorkers: 28

The scoreboard adds state detail: _ indicates waiting for a connection, R reading a request, and W sending a reply. Do not classify every symbol other than _ as an actively executing request; empty slots, starting workers, and graceful shutdown states also exist.

Look for sustained pressure

Exhausted workers can lead to queued connections and eventually timeouts or refused work as other limits are reached. They do not guarantee an immediate 502 response. High traffic and slower service time can both increase concurrency; the metrics alone do not prove which caused the rise.

Relate the numbers to the MPM

With prefork, workers are processes. With worker and event MPMs, request workers are threads. Event MPM can manage eligible keep-alive and other asynchronous connection states without occupying a worker thread for the entire connection lifetime. Consequently, open connections can exceed the busy-worker count.

An illustrative event-MPM configuration is:

<IfModule mpm_event_module>
    StartServers 3
    MinSpareThreads 25
    MaxSpareThreads 75
    ThreadsPerChild 25
    MaxRequestWorkers 400
</IfModule>

These are example values, not a sizing recommendation. ServerLimit, ThreadLimit, memory, and backend capacity constrain the valid and useful settings. MaxRequestWorkers limits simultaneous request service; it is not a promise that every status snapshot's busy-plus-idle count equals that value. Graceful process generations and asynchronous connections require additional interpretation.

Before increasing capacity, find where requests spend their time. Restarting Apache may temporarily reduce symptoms without identifying whether the underlying issue is a slow dependency, a resource leak, or ordinary traffic growth.

Guance's Apache collector can read a restricted server-status?auto endpoint. Compare its initial readings with the endpoint and correlate worker usage with request latency and error logs.

References: mod_status, event MPM, and MPM directives.