Close response body in SendEvent/HealthCheck to fix FD + memory leak (#15)

[michaeltoop]

Fixes #15.

Problem

SendEvent (swarmpit/client.go) discards the http.Post response and never closes resp.Body:

_, err := http.Post(arg.EventEndpoint, "application/json; charset=utf-8", buffer)

An unclosed/undrained response body keeps the underlying TCP connection out of the keep-alive pool, so each call leaks a connection + file descriptor. SendEvent runs on every stats tick, so the agent's open-FD count and memory grow without bound.

In production (7-node Swarm, agent as a global service) this shows up as:

• long-lived agents logging socket: too many open files — first on the event POST (Post http://swarmpit:8080/events), then collaterally on the docker.sock stats fetches — after which the agent stays alive but stops shipping events (UI graphs go blank);
• busy-node agents climbing to their memory limit and being OOM-killed (exit 137, OOMKilled=true) on a roughly hourly cadence.

The ContainerUsage stats path already does defer resp.Body.Close(); only SendEvent was missing it.

Fix

Drain and close the body in SendEvent (and the matching http.Get in HealthCheck, which had the same omission):

resp, err := http.Post(...)
if err != nil {
    log.Printf("ERROR: Event sending failed: %s", err)
    return
}
io.Copy(ioutil.Discard, resp.Body)
resp.Body.Close()

ioutil.Discard is used to stay compatible with the module's go 1.12 (io.Discard landed in 1.16). Builds clean with go build ./swarmpit/.

🤖 Generated with Claude Code