Skip to content

I0923 — Cold start and endpoint cost are benchmark targets like everything else

Measure the cold start of the container and the cost of the cheap endpoints through the benchmark infrastructure the repository already has, and record them as baselines.

BLOCKED wave 6 · p1 · implementation profile · parallel safe

Part of fly-deployment — A deployment is a projection of the repository model, and a reachable URL is its evidence.

Blocked. This issue cannot start until I0930 is done. The status is derived from that, not declared.

Objective

Measure the cold start of the container and the cost of the cheap endpoints through the benchmark infrastructure the repository already has, and record them as baselines.

Why

A profile with no minimum running machines means every idle period ends with a cold start that a user pays for. A number nobody measured cannot be improved and cannot be defended.

Current state

The benchmark command derives its targets from the command registry and holds a committed baseline. Nothing measures a container or an HTTP endpoint under deployment conditions.

Desired state

Process start to listening, process start to readiness and container start to readiness are recorded, the cheap endpoints have latency distributions under the constrained profile, and both are baselines the regression check knows about.

Scope

  • .ai/repo/benchmarks/rust
  • apps/majordomus-cli/src/bench

Out of scope

  • A load-testing framework
  • A throughput claim derived from one machine
  • Promising a cold-start guarantee from an unrepeatable measurement

Dependencies

What waits on this

Acceptance criteria

  • Process start to listening, process start to readiness and container start to readiness are each measured separately
  • The cheap endpoints are benchmarked under the constrained resource profile with distributions rather than a single number
  • The measurements are recorded as artifacts through the existing benchmark infrastructure with the environment they were taken on
  • No cold-start guarantee is stated that the repeatability of the measurement does not support

Validation

  • majordomus bench --check

Evidence required

  • coldstart_measured
  • endpoint_bench
  • baseline_recorded

Evidence

None recorded. Every token above needs a command or an artifact behind it before this issue can be completed; narrative is refused.

Risk

Container timings on a laptop and on a hosted machine differ enough that one cannot be a baseline for the other. The environment is part of the record.

Timeline

started
—
verified
—
completed
—

Those three fields, the evidence above and the state of the dependencies are all the status is made of. There is no status field to disagree with them.

Canonical record: .ai/repo/project/issues/I0923.yaml. Read it back with majordomus plan show I0923.