Notes on where HAProxy handles the "observe" check

haproxy_observe_check.md

Various HTTP states along the pipline are grouped into HANA_STATUS_* codes, which are defined in checks.h. Then health_adjust() is called with that status code, which ends up calling __health_adjust() if the observe config parameter is enabled.

Example from proxy_http.c:

/* Adjust server's health based on status code. Note: status codes 501
 * and 505 are triggered on demand by client request, so we must not
 * count them as server failures.
 */
if (objt_server(s->target)) {
	if (txn->status >= 100 && (txn->status < 500 || txn->status == 501 || txn->status == 505))
		health_adjust(objt_server(s->target), HANA_STATUS_HTTP_OK);
	else
		health_adjust(objt_server(s->target), HANA_STATUS_HTTP_STS);
}

Inside __health_adjust() (in checks.c)

• lookup the status code in struct analyze_status (which is defined in checks.h)
• if it's determined that we are in an error state then one of the HANA_ONERR_* actions is taken