Concept Contrast Fable

Rollback and model rollback

Conclusion first:Fallback is a general design pattern—switching to a backup path when the primary path fails. Model fallback is a specific scenario of fallback—switching to a backup model when the primary model is unavailable. The former can apply to any system component; the latter targets only the model-calling chain.

Fallback is a general design pattern—switching to a backup path when the primary path fails. Model fallback is a specific scenario of fallback—switching to a backup model when the primary model is unavailable. The former can apply to any system component; the latter targets only the model-calling chain.

Fable

Backup plans and backup drivers

Old Cheng's logistics fleet had fifty trucks running long interprovincial routes. Each truck had a primary driver, but the company required confirming a backup driver before departure; if the primary driver became ill on the road or triggered a fatigue-driving alert, the backup driver would take over.

That day dispatcher Xiao Hou received an urgent task: a cold-chain truck had to deliver vaccines to a place six hundred kilometers away within ten hours. He arranged the route, arranged the main driver, and confirmed the backup driver. Before departure he checked again: was there a fuel backup plan, a spare tire if a tire blew, a backup refrigeration unit if the cold machine failed, an alternative route if road signs were down.

The deputy dispatcher said, don't you already have backup drivers, why arrange so many more? Xiao Hou said, backup drivers only cover the one situation where the main driver has a problem. If fuel runs low, a tire blows, or the cooling unit breaks, swapping drivers won't help. Every step needs its own fallback.

Sure enough, trouble struck on the road—the refrigeration unit failed halfway. The backup driver couldn't help; he didn't know how to fix refrigeration units. But the partner cold storage that Xiao Hou had arranged in advance came into play: the convoy stopped for forty minutes, swapped in a working refrigeration unit, and continued on. The vaccines were delivered on time.

In the after-action review, Xiao Hou drew a fleet diagram. Drivers, fuel, tires, refrigeration units, road signs, warehouses—every point could fail; every point had a different fallback. Some could swap people, some needed swapped goods, some had to detour, and some could only pause.

Later, a newcomer applied the backup driver's rules to the cold storage, and when the main cold storage lost power, the goods were sent to a small warehouse with unstable temperature, causing even greater losses. At the meeting, Old Cheng said that the words "backup route" sound the same, but in actual execution you have to see which link has failed.

The fleet manual later gained an extra page: first look at the failure point, then look at the switchover conditions, then look at what would be lost by switching over, and finally look at how to come back. Saying only "we have a backup plan" still doesn't count as actually having a plan.

Later, an even trickier job came along. On the surface it looked like just an old problem magnified, but in reality it forced out all the gaps hidden in the process: who takes over first, who can make changes, and how to return to the previous step when something goes wrong. Places that used to get by on acquaintances' tacit understanding now all had to be spelled out clearly.

Rollback

A general fault-tolerant design pattern that automatically switches to a backup path or component when the primary path or component is unavailable. Applicable at any layer of the system: network, storage, compute, authentication, model invocation, etc.

Model fallback

Specific application of fallback mode in model invocation scenarios: when the primary model is unavailable, times out, or returns quality below a threshold, automatically switch to a backup model to continue service. It needs to consider interface compatibility between models, latency budget and output format consistency.

Story Mapping

Backup driver: model fallback—failover only for the driver (model), a single component.
Fuel, tires, and cold machines each have alternatives: system-level fallback—every critical component needs its own fallback path.
The refrigerator broke but the backup driver was not used: fallbacks for different components do not cover each other, and each type of failure needs a corresponding plan.
Database rollback configuration applied to models: rollback strategies for different components cannot be reused directly, and the switch conditions differ.
Layered fallback handbook: general principles plus component-level differences, with each component refined under a unified framework.

In the project

Do not just do a "switch to backup when the model goes down" and call it done. Draw a fallback matrix: the horizontal axis is system components (model, database, cache, authentication, storage), the vertical axis is the four elements of fallback (trigger conditions, switch actions, data loss, recovery steps). Fill in every cell clearly, and fallback turns from a sentence into an operable plan. In review, if you only hear "we have fallback," ask the other party to open the matrix.

Rollback Model fallback