Time, Clocks and Ordering — Distributed Systems for Backend Engineers

Wall clocks (NTP-synchronized) drift and can jump backwards, so they can't order events across machines reliably. Monotonic clocks measure durations on one machine. Logical clocks (Lamport timestamps) order causally related events; vector clocks detect concurrency; hybrid logical clocks (HLC, used by CockroachDB and others) combine physical time with logical counters. Google Spanner's TrueTime exposes clock uncertainty bounds and waits them out to provide external consistency.

Clocks

Clock	Gives you	Use for
Wall clock (NTP)	Approximate real time	Timestamps for humans, TTLs (with slack)
Monotonic clock	Elapsed time on one host	Timeouts, latency measurement
Lamport clock	Total order consistent with causality	Ordering messages
Vector clock	Detects concurrent updates	Conflict detection (Dynamo-style)
Hybrid logical clock	Causality + close to physical time	Distributed databases, MVCC

class LamportClock:
    def __init__(self):
        self.time = 0

    def tick(self) -> int:                 # local event or send
        self.time += 1
        return self.time

    def receive(self, remote_time: int) -> int:
        self.time = max(self.time, remote_time) + 1
        return self.time

start := time.Now()                    // contains a monotonic reading
doWork()
elapsed := time.Since(start)           // uses the monotonic clock: safe across NTP adjustments

Common pitfalls

Last-write-wins by wall-clock timestamps across nodes (silent data loss).
Using wall time for timeouts.
Assuming event timestamps from different services are comparable to the millisecond.

Key points

Wall clocks drift and jump
Monotonic for durations
Logical/hybrid clocks for ordering

Further reading: Lamport: Time, Clocks, and the Ordering of Events

Practice questions

Why is last-write-wins using wall-clock timestamps dangerous?

Answer

Clock skew means a later write on a node with a slow clock can get an earlier timestamp and be discarded, silently losing data.

What does a vector clock let you detect that a Lamport clock can't?

Answer

Whether two events are concurrent (neither causally precedes the other), which indicates a conflict.