meridian-plugin-usage

Meridian V1's open-source Usage plugin provides immutable metering events, exact unit normalization, bounded queries, deterministic window aggregation, retention inputs, and OpenTelemetry/Evidence correlation over logical Meridian Resources.

This distribution is a Python library and Meridian plugin. It is not a service or a Usage Catalog. It owns no database, provisions no engine, accepts no engine credentials, calculates no prices, and uses only released mapping-first Meridian Expressions.

Install

python -m pip install meridian-plugin-usage==2.0.1

See decimal replay compatibility for the storage roundtrip fix, unchanged v1 event fingerprints and legacy verification.

Python 3.12, 3.13, and 3.14 are supported. Runtime Meridian dependencies are exactly pinned to the compatible releases listed in compatibility.

Record immutable usage

Platform IaC renders the Meridian runtime, logical Resource placements, identity/ACL, migrations, retention, and engine lifecycle. Application code only creates the ready runtime and uses this library:

from datetime import UTC, datetime
from decimal import Decimal

from meridian_storage.plugins.usage import (
    DimensionSpec,
    MeterV1,
    UnitTransform,
    Usage,
    UsageEvent,
    UsageScope,
    UsageWindow,
)

usage = Usage(ready_meridian)

meter = MeterV1(
    meter_id="api.requests",
    version=1,
    quantity="requests",
    canonical_unit="request",
    transforms={"kilorequest": UnitTransform(Decimal("1000"))},
    dimensions=(DimensionSpec("region", required=True),),
    scale=6,
    active_from=datetime(2026, 1, 1, tzinfo=UTC),
)
usage.repository.register_meter(meter)

event = UsageEvent(
    event_id="publisher-event-0001",
    scope=UsageScope({"tenant": "acme", "environment": "prod"}),
    subject_id="account-42",
    meter_id=meter.meter_id,
    meter_version=meter.version,
    window=UsageWindow(
        datetime(2026, 8, 26, 10, 0, tzinfo=UTC),
        datetime(2026, 8, 26, 10, 1, tzinfo=UTC),
    ),
    value=Decimal("2"),
    unit="kilorequest",
    dimensions={"region": "us-west"},
    source="publisher",
)
receipt = usage.recorder.record(event)
assert receipt.event.value == Decimal("2000.000000")

An identical replay returns RecordStatus.REPLAYED. The same scoped event_id with different normalized content raises the stable MERIDIAN_USAGE_IMMUTABLE_CONFLICT error. Corrections are new signed events referencing correction_of; existing events are never mutated.

Query and aggregate

page = usage.queries.events(
    event.scope,
    datetime(2026, 8, 26, 10, 0, tzinfo=UTC),
    datetime(2026, 8, 26, 11, 0, tzinfo=UTC),
    where={"meterId": meter.meter_id},
).page(limit=100).execute()

Every query is scope-bound, time-bounded, cursor-paginated, and exposed both as a mapping-first Expression and a serialized Query 1.0.0 logical plan. Aggregation is a library worker with deterministic claims, immutable aggregate versions, and compare-and-set watermark checkpoints. Retrying after a crash replays identical outputs; late events create a new version that names the version it supersedes.

Batch profiles

• partitioned records deterministic per-event outcomes and is portable to append-oriented placements such as ClickHouse.
• atomic opens one Meridian transaction on the events Resource and is accepted only when the rendered placement advertises the transaction capability.

Batch manifests and aggregation control records are logical structured Resources. They are not a private database.

Contracts and evidence

The package ships released Semantics schemas, plugin/schema-provider entry points, JSON Schema compatibility evidence, deterministic golden fingerprints, unit/integration/conformance tests, reproducible build checks, and Apache-2.0 LICENSE/NOTICE material.

See architecture, recording, aggregation, configuration, compatibility, and conformance.

Licensed under the Apache License, Version 2.0.