As a platform lead, how would you choose the function platform your organisation standardises on, and when would you allow a second one?
answer
- data gravity decides more than features
- a default per workload class
- the paved road is the cost
- exceptions need an owner
basics
~20 sStandardise on the regional function platform of the cloud already holding your data, identity and network, with a paved road around it. Allow a second platform, such as an edge runtime, only for a workload class the default cannot serve, with an owner.
solid answer
~50 sI would start from **data gravity**: the function platform of the cloud that already holds the databases, identity and private network wins most ties, because cross-cloud calls add latency, egress cost and a second permission model. Then I would check the **workload mix** against each candidate: event glue between managed services favours AWS Lambda or Azure Functions in their own clouds, HTTP services that may later move favour a container-based platform such as Cloud Run, and latency-critical request handling with little data favours an edge runtime such as Cloudflare Workers. The standard is a **default per workload class** plus a paved road of templates, CI, telemetry and policy. A second platform is allowed when a real workload class cannot be served by the default, and only with an owner, paved-road investment and a review trigger, because every extra platform multiplies tooling, on-call knowledge and security review.
go deeper
Recall that functions sit next to databases and queues, so the cloud holding the data usually decides which function platform a company uses.
Explain how workload classes such as event glue, HTTP services and edge request handling map onto different function platforms.
Show how a paved road, policy guardrails and cost tagging make a platform genuinely supported rather than merely tolerated.
Own the trade-off: one default per workload class, the standing cost of each extra platform, explicit exception criteria and the triggers that reopen the decision.
## Why this is a lead's decision Choosing a **function platform** for one service is an engineering call. Choosing the one an organisation **standardises on** is a decision with a standing cost: it shapes templates, pipelines, security reviews, on-call skills and contracts for years. There is no single right answer, which is why it sits with a platform lead. The candidates usually include **AWS Lambda**, **Azure Functions**, **Cloud Run** and **Cloudflare Workers**, and the answer depends more on the organisation than on the platforms. ## The inputs that actually decide it 1. **Data gravity.** Functions spend their lives calling databases, queues and storage. The platform inside the cloud where those already live avoids cross-cloud latency, egress charges and a second identity model. This one input settles most real decisions. 2. **Workload mix.** Glue between managed services (react to an upload, a queue message, a schedule) suits the native function product of that cloud, because its triggers are wired for you. Long-lived HTTP services that might move suit a container-based platform. Request handling that must be close to users and needs little data suits an edge isolate runtime. 3. **Portability appetite.** A container-based platform with a plain HTTP contract keeps an exit open to Kubernetes or another cloud; a platform with rich bindings trades that exit for less glue code. 4. **Runtime constraints.** Native libraries, large memory or long-running requests rule out isolate runtimes for those workloads; very tight latency budgets near users rule in the edge. 5. **Cost shape of the dominant workloads.** Wall-clock, instance-time and CPU-time billing reward different traffic patterns; model the organisation's real mix rather than one benchmark. 6. **Compliance and residency.** An edge runtime runs code in many countries by default; regulated data may need placement controls or a regional platform. 7. **Team skills and support contracts.** The platform people already operate well has a lower real cost than a theoretically better one. ## Shape of a good standard - **A default per workload class**, not one platform for everything: for example, the primary cloud's function product for event glue, its container platform for HTTP services, and an edge runtime as an allowed option for edge request handling. - **A paved road** for each allowed platform: a template, a pipeline, standard telemetry, security policy and cost tagging. A platform without a paved road is not really allowed; it is tolerated. - **Guardrails** expressed as policy: approved regions, required tags, mandatory per-function concurrency caps and budget alarms. ## When to allow a second platform | Signal | Allow it? | |---|---| | A workload class the default cannot serve, such as latency-critical edge logic | yes, with an owner and a paved road | | An acquired team already running production on another platform | yes, temporarily, with a convergence or keep decision and a date to make it | | A team prefers another platform's developer experience | usually no; fix the paved road instead | | A benchmark shows a small price difference on one workload | no; migration and tooling costs outweigh it | Every additional platform carries a **standing cost**: - duplicated templates, pipelines and policy; - a second permission model to review and audit; - on-call engineers who must know two sets of failure modes; - telemetry that has to be joined across two systems. ## Revisit triggers A standard is not permanent. Re-open it when: - the organisation's data moves to another cloud; - a new workload class appears that the defaults serve badly; - the cost shape of the dominant workloads shifts, for example from bursty to steady; - a platform changes its isolation model, limits or pricing shape enough to change a past decision. ## What interviewers listen for They want a lead who starts from where the data and people already are, defines defaults by workload class, prices the paved road and the standing cost of each extra platform, and names the evidence that would justify an exception or a change. Answers that pick a platform from a feature list alone, or allow every team its own choice with no paved road, miss the point of standardising at all.
- A product team wants Cloudflare Workers for a new API whose every request reads a regional database; how do you respond?Ask what the edge buys them. If each request crosses back to the regional database, users see little latency gain while the team takes on a second platform, permission model and runbook. Unless the data can be cached or replicated near the edge, the default regional platform serves them better; an edge front for caching and routing is a cheaper compromise.
- How do you stop the default from quietly becoming the only option for workloads it fits badly?Publish the exception criteria and the paved-road cost alongside the default, review exception requests on evidence such as latency, data locality or runtime needs, and schedule a periodic review of the workload mix so a new class of work can earn its own default.
- Pick the platform with the best feature list and move the data later.
- Standardising means one platform for every workload class.
- Letting each team choose its platform has no lasting cost.
- A small benchmark price difference justifies adopting a second platform.
- Once chosen, the standard never needs to be revisited.