Deploying a trading system without stopping the marketZero-downtime deployment is a solved problem for web services and an unsolved one for anything holding a position. Here is how the patterns change when state and orders are on the line.
Latency is a budget, not a numberA single average latency figure hides everything that matters. Treating latency as an allocated budget across a path, measured at the tail, is what turns optimisation from guesswork into engineering.
Migrating a legacy trading stack without a big-bang cutoverThe weekend cutover is the most expensive migration strategy available and the most commonly chosen. The alternative is slower to start, far cheaper to finish, and reversible at every step.
The build pipeline as a risk controlRegulators and risk committees ask who can change a production trading system and how that change is evidenced. A well-designed pipeline answers both automatically, as a by-product of working normally.