Health Monitors and Server Maintenance
Monitor layers
Health monitors test a member and set its state. Layer 3/4 monitors (ICMP, TCP connect, UDP) only prove reachability; a TCP monitor can pass while the application is broken. Layer 7 monitors (HTTP, HTTPS, DNS, and similar) send a request and expect a response code or string, which is what you want for real availability.
Parameters
Each monitor defines an interval, a timeout, and retries, and often separate up and down retries. Detection time is roughly interval multiplied by down-retries, plus the timeout of the final probe, so a 5-second interval with 3 retries takes at least 15 seconds to mark a server down. Monitors attach to a service group and apply to every member unless overridden per server.
Server maintenance
To remove a server without cutting users, disable it gracefully: the ADC stops new connections while existing sessions drain, after which you can disable forcefully or shut the host down. Disabling is not deleting, and by itself it does not terminate established flows. A server with no monitor is assumed up, which is convenient in a lab and dangerous in production.
Traps
Expect questions where a monitor is up but users see errors (L4-only check), where retries are confused with interval, or where disabling a server is expected to kill sessions immediately. Binding a monitor to the wrong port is another common miss.
Rule of thumb
Monitor what users actually use: an HTTPS VIP deserves an application-aware check, not just a ping. Worked example: a web farm flaps every few seconds; raising retries and the interval stops false failovers caused by brief latency spikes.