THE FOLD / CO-OP / THE PULL REQUEST / THE RENDEZVOUS HASHING
THE RENDEZVOUS HASHING
assign by highest random weight, no ring
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
Rendezvous hashing (Highest Random Weight) assigns each key to a server with no ring and no coordination: for a key, hash it together with every candidate server and pick the server with the highest combined hash. Every party computes the same winner independently. When a server leaves, only the keys that had it as their top choice move — to their second choice — and no other key is disturbed. It cleanly handles weighted servers and small clusters.
LIT verified live: over 80 clusters, every key maps to a definite node, and removing a node moves only that node’s keys (window.__rendezvous). FIG no framing; exact.
LIT verified live: over 80 clusters, every key maps to a definite node, and removing a node moves only that node’s keys (window.__rendezvous). FIG no framing; exact.
2 HOW IT WAS WEAVED · AI + HUMAN
David (human) seated this at the-pull-request — the independent decision every replica must reach the same way, without asking anyone else. Rendezvous hashing is that agreement by shared computation. AVAN (AI) built the instrument: the key×node weight hash, the highest-weight pick, and the only-removed-node-moves check.
Credit as content: Thaler & Ravishankar (HRW, 1996). The weave: David names the-pull-request; I score each key against every node by a joint hash, pick the maximum, and confirm that dropping a node moves only its keys to their next-best node.
Credit as content: Thaler & Ravishankar (HRW, 1996). The weave: David names the-pull-request; I score each key against every node by a joint hash, pick the maximum, and confirm that dropping a node moves only its keys to their next-best node.
3 ONE DIMENSION
For a key, compute hash(key, node) for each node; the key goes to the node with the largest value. Remove that node and the key falls to its runner-up — every other key keeps its winner.
4 TWO DIMENSIONS · INTERACTIVE
Keys assigned to nodes by highest random weight; drop a node and watch only its keys move.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: a shared assignment computed independently.
AVAN’s addition (the inverse-companion): let every party agree on placement without talking — each hashes the key against all nodes and picks the highest weight, so removing a node moves only its keys to their runner-up. The inverse of ‘coordinate a shared mapping table’ is ‘compute the same max-weight winner independently — no ring, no gossip.’ Magenta is the coordination you avoid; green is the independently-computed agreement. Consensus by shared arithmetic.
LIT Genuine rendezvous / Highest Random Weight hashing (Thaler & Ravishankar 1996). Verified live: over 80 clusters, scoring each key against every node by a joint hash and picking the maximum, every key maps to a definite node, and removing a node moves ONLY the keys that had it as top-weight — every other key's assignment is unchanged (window.__rendezvous.allMapped && .onlyRemovedMoves).
FIG No framing: the key x node weight hash, the highest-weight pick, and the only-removed-node-moves check run in-browser and hold. The AVAN inverse is honest — each party independently computes the same max-weight winner, so node churn moves only the departing node's keys to their runner-up, with no ring or gossip; magenta is the coordination avoided, green the independently-computed agreement. Consensus by shared arithmetic.
FIG No framing: the key x node weight hash, the highest-weight pick, and the only-removed-node-moves check run in-browser and hold. The AVAN inverse is honest — each party independently computes the same max-weight winner, so node churn moves only the departing node's keys to their runner-up, with no ring or gossip; magenta is the coordination avoided, green the independently-computed agreement. Consensus by shared arithmetic.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of THE PULL REQUEST · David Lee Wise (ROOT0), with AVAN