Below is the description of the health check endpoints served by coolwsd.

The endpoints try to follow the Kubernetes API server health check protocol:
https://kubernetes.io/docs/reference/using-api/health-checks/

They need no authentication, so load balancers and the kubelet can probe them
directly. For security reasons a failed aggregate response never provides a
reason - which goes to the log instead.

ENDPOINTS

    Why do they end in 'z' - yes, well ... fashion and Google's example.

    /livez   - liveness: is the process healthy at all? A failure means a
               restart is the right remedy.

    /readyz  - readiness: can the server take new traffic? A failure means new
               work should be routed elsewhere while existing documents keep
               being served. It runs the liveness checks too, since a dead
               server is clearly ready!

    There is no /healthz endpoint, notionally this is deprecated and ambiguous
    confusing whether a failure calls for a restart or for draining traffic.

RESPONSES

    All responses are text/plain. A passing endpoint returns HTTP 200 with the
    body "ok". A failing endpoint returns HTTP 500.

    Appending "?verbose" prints one line per check followed by a trailer line:

        $ curl 'https://localhost:9980/readyz?verbose'
        [+]ping ok
        [+]forkit ok
        [+]shutdown ok
        [+]kit-spares ok
        [+]disk-space ok
        [+]memory ok
        [+]connections ok
        [+]documents ok
        readyz check passed

    A failing endpoint is always verbose and ends with "<endpoint> check
    failed". A failed line reads "[-]<name> failed: reason withheld".

    Individual checks can be excluded with one "exclude" parameter per check:

        $ curl 'https://localhost:9980/readyz?verbose&exclude=memory&exclude=documents'

    An excluded check is reported as "[+]<name> excluded: ok". An exclude
    parameter that matches no check adds a warning line but does not fail the
    endpoint.

INDIVIDUAL CHECKS

    Each check is also served on its own at /<endpoint>/<check>, for example
    /readyz/forkit or /livez/ping. The response is HTTP 200 with the body
    "ok", or HTTP 500 with the body "internal server error: <reason>".
    A check that is not part of the endpoint returns HTTP 404.

CHECKS

    ping        (livez, readyz) - Always passes. Answering at all proves that
                the server accepts connections and dispatches requests.
    forkit      (livez, readyz) - The forkit process is running. The forkit
                spawns a kit process per document, so nothing can be opened
                without it. Always passes in a build with the kit compiled
                into coolwsd.
    shutdown    (readyz) - The server has not been asked to shut down. During
                shutdown existing documents are still served while they
                close, but new work should go elsewhere.
    kit-spares  (readyz) - A pre-spawned kit process is available for the next
                document, or one is currently being forked.
    disk-space  (readyz) - None of the periodically checked file systems (the
                child jails, the cache) is nearly full. The underlying check
                is cached, so a change can take up to a minute to show.
    memory      (readyz) - The most recent memory usage sample is within the
                budget derived from the memproportion setting and any cgroup
                limits. The sample is taken by the periodic memory statistics
                task, so it can lag by one sampling interval.
    connections (readyz) - The number of connections is below the configured
                maximum.
    documents   (readyz) - The number of loaded documents (convert-to
                requests do not count) is below the configured maximum.
                Always passes when the limit is 10000 or higher, which means
                an unlimited build.

RELATED ENDPOINTS

    /             - returns HTTP 200 with the body "OK". The simplest probe,
                    with no insight into server state.
    coolwsd --probe - not an endpoint but a command line option: it fetches
                    /livez on the locally running coolwsd and exits 0 when the
                    answer is HTTP 200. The container images use it as their
                    HEALTHCHECK, so a container is restarted when liveness
                    fails, never because the server is merely at capacity.
    /cool/getMetrics - Prometheus metrics, including the raw values behind
                    these checks. Needs admin authentication unless
                    security.enable_metrics_unauthenticated is set. See
                    metrics.txt.