Lake Yunxi was perfectly round, and thousands of households lived scattered along its shore, their house numbers strewn about in no particular order — nobody could tell which house sat next to which. The town’s post office came up with a scheme: treat the entire shoreline as one continuous ring road. Every house number (really just name plus address) got run through a fixed hashing formula that placed it at some specific point on that ring — the larger the resulting number, the further “along” the ring it landed, and the largest number wrapped right back around to sit next to the smallest, closing the loop.

The post office had originally hired four carriers: Old Zhou, Old Chen, Ah Wu, and Xiao Lin. Each picked a spot on the ring and planted a flag there as his “post.” The rule was simple: walking clockwise from your own flag until you hit the next flag, every household along that stretch was yours to deliver to. Old Zhou’s flag sat near the ring’s “starting point,” so he covered everything clockwise from his flag until Old Chen’s flag; Old Chen covered from there to Ah Wu’s flag; and so on, until Xiao Lin’s stretch wrapped back around to meet Old Zhou. Four flags split the whole ring into four segments — no overlaps, no gaps.

As the town grew, the post office decided to hire a fifth carrier, Xiao Ma. Xiao Ma picked an open spot on the ring and planted his flag — right in the middle of what had been Old Chen’s stretch. That single move meant only the small stretch of households between Xiao Ma’s flag and Old Chen’s flag (counting backward) needed to be reassigned to Xiao Ma; everything Old Zhou, Ah Wu, and Xiao Lin had been covering stayed exactly as it was. Out of the town’s several thousand households, only the small cluster near Xiao Ma’s flag actually changed hands.

Later, when Old Zhou grew old and retired, pulling up his flag was just as simple: the stretch that had been his (immediately clockwise of his old spot) automatically merged into the next flag along — Old Chen picked it up, and nobody else was affected at all.

The post office also ran into trouble: where a flag landed was pure luck. Sometimes a carrier’s flag happened to land near a dense cluster of hundreds of households and he’d be run off his feet, while another carrier’s flag landed in a sparse stretch with barely any households, leaving him idle. The fix they eventually landed on: instead of planting just one flag, each carrier planted several small flags at different points scattered around the ring — all counted as his own posts, just spread out. That way, even if some region happened to be crowded, it would likely fall near someone’s small flag, and the workload got smoothed out much more evenly — no longer left to the bad luck of where a single flag happened to land.

The old postmaster still remembered an even older method: just take each house number, divide it by the number of carriers, and whoever matched the remainder got the delivery. Simple enough — but the moment the post office hired or let go even a single carrier, the divisor changed, and almost every household’s remainder changed with it. That meant re-sorting almost the entire town’s mail from scratch, and the handoff between old and new carriers turned into chaos. By comparison, the flag-and-segment scheme meant that adding or removing a carrier only ever touched the small stretch immediately next to that one flag.

——By now you’ve probably recognized it: this ring-shaped post office’s segmenting scheme is really describing consistent hashing, and the trick of planting several small flags to spread out the load is its optimization, virtual nodes.

The concept, and why it matters Consistent hashing is a mapping scheme widely used in distributed systems for data sharding and load balancing: both nodes (servers, cache instances, mail carriers) and data keys are run through the same hash function onto a shared ring-shaped numeric space, and each key belongs to the first node encountered going clockwise around the ring. Its biggest advantage — and the point of this fable — is that adding or removing a node only requires reassigning the small interval immediately adjacent to it; the vast majority of keys are completely unaffected. This stands in sharp contrast to traditional “key mod N” sharding, where changing the node count N at all forces nearly every key to be remapped — a very expensive operation. DynamoDB, Cassandra, distributed caching schemes like memcached, CDN edge-node routing, and even some Kubernetes load-balancing/session-affinity strategies all rely on this idea. As for the problem of a node landing an unfairly large or small interval purely by chance (data skew), that’s exactly what virtual nodes — giving each physical node multiple virtual positions on the ring — are designed to smooth out.

Metaphor mapping

• The ring road along Lake Yunxi’s shore → the hash ring in consistent hashing (typically a numeric space from 0 to 2^32-1)
• The position each house number hashes to → a data key’s position on the ring
• A carrier’s planted flag/post → a node’s position on the ring
• “Everything clockwise until the next flag is mine” → each node owns the interval up to the next node on the ring
• Xiao Ma’s new flag only affecting the adjacent stretch → adding a node only requires reassigning the adjacent interval’s keys
• Old Zhou’s flag pulled at retirement, his stretch auto-merging into the next → removing a node merges its interval into the next node clockwise
• One carrier planting several small flags at scattered points → virtual nodes, spreading load across multiple virtual positions to avoid skew
• The old “house number mod number of carriers” method → traditional key-mod-N sharding
• Nearly every household needing reassignment when the carrier count changes → the high cost of full remapping when N changes under mod-N sharding