timerfd¶

timerfd converts timer expiration events into file descriptor events. This makes timer handling fit naturally into an epoll event loop.

Why timerfd?¶

Traditional timer approaches can be awkward in event-driven programs:

Approach	Limitation
`sleep()`	Blocks the thread
Signal-based timer	Requires async signal handling
Extra timer thread	Requires synchronization
`timerfd`	Integrates directly with `epoll`

timer expires
    ↓
timerfd becomes readable
    ↓
epoll_wait() returns
    ↓
read(timerfd) to consume expirations

Reading from a timerfd returns the number of expirations since the last read. It does not return a timestamp.

uint64_t expirations;
read(timer_fd, &expirations, sizeof(expirations));

Example:

timer interval = 1 second
read after 5 seconds → expirations = 5

A scalable idle timeout design often uses:

single global timerfd
    +
per-client last_active timestamp

Instead of creating one timerfd per client.

timerfd tick
    ↓
scan client list
    ↓
compare now - last_active
    ↓
close idle clients

Pitfall	Result
Forgetting to read the timerfd	epoll keeps reporting the timerfd as readable
Creating too many timerfds	More fd management complexity
Assuming read returns time	Incorrect timeout logic
Blocking inside timer handler	Delays all other event sources