What does Single-Host Hot Migration and the Ten-Second Contract conclude, in numbers, about “Expand-Contract Migration: the Rollbackability Graph and Pushing Irreversible Steps to the End”?
Answer: Chapter 2 draws migration as a rollbackability graph: with a 24 hour bake, expand-contract has a rollback failure rate 10.39 times lower than a one-shot migration, and with two coexisting containers only 8 of 560 interleavings are safe.
Scope of the numbers: Examples use fictional names, parameters are illustrative, and all calculations come from the book's programs, which connect to no server. Marked boxes separate published evidence, runnable Python calculations (22 programs), classical sources and the author's hypotheses.
Data table
The numbers below are printed verbatim by the chapter's Python program and can be reproduced by running it. standard-library Python onlyprinted values that match the answer: 10.39, 24
expand-contract: bake B (hours) before each contract step
B P(fast) P(ok) E[time] s
0 0.0000 0.8802 3909.0
1 0.3645 0.9364 1999.7
6 0.7086 0.9741 763.0
24 0.8930 0.9904 242.8
72 0.9634 0.9957 78.5
168 0.9866 0.9972 31.7
720 0.9984 0.9979 11.3
big-bang: every rollback is a restore of 3600 s
restore success 0.50 -> P(ok) 0.5000
restore success 0.70 -> P(ok) 0.7000
restore success 0.90 -> P(ok) 0.9000
restore success 0.99 -> P(ok) 0.9900
B = 24 h against big-bang at 0.90: failure 0.0096 vs 0.1000, ratio 10.39Show the full program
# Chapter 2. p22: rollback success against the bake period before the
# contract steps (a dose-response sweep).
# Model: a migration carries a latent fault. Once the new read path is
# live, the fault is noticed after a time X, lognormal with median 2 h
# and spread 2 (most faults show within hours, a few after weeks).
# Expand-contract waits a bake period B before step F (stop old
# write) and another B before step G (drop old column). The rollback
# cost then depends on where X falls (times and success from p21):
# X < B fast undo, 9 s, success 0.998
# B <= X < 2B undo with reverse backfill, 309 s, success 0.978
# X >= 2B restore from backup, 3909 s, success 0.8802
# Big-bang is irreversible from the start: restore with success r.
# All figures are illustrative.
import math
MED_H = 2.0
SPREAD = 2.0
TIERS = ((9.0, 0.998), (309.0, 0.978), (3909.0, 0.8802))
BAKES = (0.0, 1.0, 6.0, 24.0, 72.0, 168.0, 720.0)
def cdf(x_h):
if x_h <= 0.0:
return 0.0
z = (math.log(x_h) - math.log(MED_H)) / SPREAD
return 0.5 * (1.0 + math.erf(z / math.sqrt(2.0)))
def expand_contract(bake):
p1 = cdf(bake)
p2 = cdf(2.0 * bake) - p1
p3 = 1.0 - p1 - p2
w = (p1, p2, p3)
succ = math.fsum(wi * s for wi, (_, s) in zip(w, TIERS))
time = math.fsum(wi * t for wi, (t, _) in zip(w, TIERS))
return p1, succ, time
print("expand-contract: bake B (hours) before each contract step")
print("%8s %10s %10s %12s" % ("B", "P(fast)", "P(ok)", "E[time] s"))
for b in BAKES:
p1, s, t = expand_contract(b)
print("%8.0f %10.4f %10.4f %12.1f" % (b, p1, s, t))
print("big-bang: every rollback is a restore of %.0f s" % 3600.0)
for r in (0.5, 0.7, 0.9, 0.99):
print(" restore success %.2f -> P(ok) %.4f" % (r, r))
_, s24, _ = expand_contract(24.0)
print("B = 24 h against big-bang at 0.90: failure %.4f vs %.4f,"
" ratio %.2f" % (1.0 - s24, 0.10, 0.10 / (1.0 - s24)))Single source and citation
Fan, Y. (2026). Zero-Downtime Hot Migration and Rollback on Multi-Container Single Hosts: Lossless Schema Evolution behind a Shared Tunnel and a Ten-Second Stop-the-Bleeding Contract (Version 1.0.0), Chapter 2: Expand-Contract Migration: the Rollbackability Graph and Pushing Irreversible Steps to the End. 5D Library. DOI pending
DOI pending · monograph page goes live after publication · 5DL-B-269 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026269,
author = {Fan, Yafu},
title = {Zero-Downtime Hot Migration and Rollback on Multi-Container Single Hosts: Lossless Schema Evolution behind a Shared Tunnel and a Ten-Second Stop-the-Bleeding Contract},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}Other chapter answers in this book
- Chapter 1 · Blue-Green Is Not Zero Downtime: the Interruption Window Set by Health Checks, Failure Thresholds and Connection Drain
- Chapter 3 · The Availability Markov Chain: How Repair Time Eats the Yearly Budget
- Chapter 4 · Many Containers on One Host behind a Shared Tunnel: Common-Cause Failure, Reconnect Backoff and Memory Headroom for Dual Running
- Chapter 5 · Lossless Evolution of a Production Database: Backfill Pace, Dual-Write Verification and the Write-Freeze Window
- Chapter 6 · The Ten-Second Stop-the-Bleeding Contract: Convolving the Time Budget, False and Missed Alarms, and the Rot of Rollback Paths
- Chapter 7 · The Classical Sources Examined: "When a Course Runs Out, Change" and the Ship of Theseus
- Chapter 8 · The Empirical Loop: a Local Drill Checklist, the Hot-Migration Interruption Window Calculator and a Pre-Registered Test