The simplest possible "live viewers" counter: an HTTP server you poll on an interval to register your presence and obtain the current viewer count.
Two options are accepted via environment variable:
PORT: Specify the port the http server will listen on.
INTERVAL_SECS: The interval (in seconds) between client polls.
ALLOWED_ORIGINS: Origin header values that views will be accepted from.
Each client should generate a random string to be used as its identifier (to dedupe multiple browser tabs open).
POST /<IDENTIFIER>: Responds with the current number of viewers in plaintext. The Cache-Control: max-age=SECS header will be present, where SECS is the number of seconds the client should wait before the next poll.
Clients are given a grace period of half the configured interval after their expected check-in before being considered inactive.
Identifiers must not be shorter than 8 characters or over 32 characters long. Any other request method than POST is rejected.
This implementation is intentionally trivial, and has obvious scaling limitations. When one process is no longer enough, I would suggest adding a redis server for coordination, running multiple tinycount processes behind a load balancer. On the configured interval, each process should write its total to the redis and fetch all of the other totals.