Qiskit / qiskit-ibm-runtime / 37458629915

91%

main: 91%

push

github

Make local mode `job.metrics()` match service jobs (#3434)

### Summary

Make `LocalRuntimeJob.metrics()` (local testing mode) return the same
layout as the metrics of a job run on the service, following #2884: ISO
8601 UTC timestamp strings, and `usage` with `qpu_charge_time_seconds`
and `status`.

### Details and comments

`LocalRuntimeJob.metrics()` still returned the layout from before API
version 2026-04-15:

| | Service job (`BaseRuntimeJob.metrics()`) | Local mode, before |
Local mode, after |
|---|---|---|---|
| `timestamps` values | ISO 8601 strings, e.g.
`"2026-09-28T16:44:27.019265Z"` | naive local `datetime` objects | ISO
8601 UTC strings (`...Z`) |
| `usage` | `qpu_charge_time_seconds`, `status`, ... |
`quantum_seconds`, `seconds` | `{"qpu_charge_time_seconds": 0, "status":
"complete"}` |
| `bss` | not present | `{"seconds": 0}` | removed |

- The timestamps are converted with the existing `local_to_utc()`
helper. The job's stored times are unchanged, so `creation_date` and the
other job attributes behave as before.
- `usage.status` is `"complete"`, as returned by the service for a
finished job and as used in `test_jobs.py`.
- I didn't add `circuits_execution_time_ns` or `qiskit_version`, since
local jobs have no service-side execution time to report. Happy to add
them if you'd prefer full key parity.
- The new test is parameterized over `SamplerV2` and `EstimatorV2`. It
checks the `usage` layout, that `bss` is gone, and that the timestamps
are UTC strings that `utc_to_local()` converts back to the job's local
`creation_date`. It fails on `main` and passes with this change.
`test/unit/local_mode`, `test/unit/test_jobs.py` and
`test/unit/backwards_compatibility` pass locally.
- This changes what local mode `metrics()` returns, so code that read
`quantum_seconds` from a local job will see the new key, as it would on
a service job. The release note is
`release-notes/unreleased/3434.bug.rst`.

Fixes #3431

### AI/LLM disclosure

- [ ] I didn'... (continued)

1 of 1 new or added line in 1 file covered. (100.0%)

8669 of 9516 relevant lines covered (91.1%)

9.95 hits per line