feat(pricing): add AWS Price List Service support by ShubhamDX · Pull Request #821 · floci-io/floci

ShubhamDX · 2026-05-12T18:01:57Z

Summary

Adds the AWS Price List Service (pricing:* / AWSPriceListService.*) emulation — the stateless prerequisite for the Cost Explorer and CUR services proposed in #791.

Five operations: DescribeServices, GetAttributeValues, GetProducts, ListPriceLists, GetPriceListFileUrl
JSON 1.1 protocol, reuses existing AwsJson11Controller dispatch
Backed by a bundled classpath snapshot; override via FLOCI_SERVICES_PRICING_SNAPSHOT_PATH for custom datasets
Pagination with Base64 offset tokens (AWS-compatible shape)
PriceList returned as array-of-JSON-strings — matches real wire format so SDKs parse offers directly

Why

Per scope discussion in #791, this is PR 1 of 3. FinOps tooling (cost detection, budget guards, CUR parsers, anomaly detectors) can't currently run against Floci because the billing APIs are absent. Pricing is the smallest stateless piece and unblocks the next two PRs (ce:*, cur:*).

Design choices

Bundled snapshot (minimal). AmazonEC2, AmazonS3, AWSLambda in us-east-1 only. Keeps the image small; users wanting full coverage drop a Price List bulk download under FLOCI_SERVICES_PRICING_SNAPSHOT_PATH.
No new dependencies. Snapshot is plain JSON, parsed with the existing Jackson mapper.
No SPI. Cost-synthesis / pluggability deferred until a second consumer emerges — keeps scope minimal per the existing stateless pattern.
Path-safe snapshot loader. Segment-level validation on any input used to build a filesystem path (ServiceCode, AttributeName, regionCode), plus root-containment check in SnapshotLoader so the override directory can't be escaped.
EffectiveDate accepts both wire shapes. AWS JSON spec uses Unix epoch seconds (number); the handler also accepts ISO-8601 strings for SDK/CLI variance.

Test plan

./mvnw test -Dtest=PricingIntegrationTest — 26/26 green locally (Java 25)
Wire-format assertions for each operation via RestAssured against a booted @QuarkusTest instance
Validation-error paths (ValidationException, InvalidParameterException, UnknownOperationException)
Filter matching, filter input validation (missing Field/Value), pagination round-trip
EffectiveDate version selection (numeric epoch + ISO-8601, with multi-version AmazonEC2 us-east-1 fixture)
Path-traversal rejection on AttributeName and regionCode
CI to run the full suite

Scope boundaries (out of scope for this PR)

CE / CUR / BCM Data Exports — will land as separate PRs per the scope thread in Proposal: add AWS Cost Explorer / CUR / Pricing API surface #791.
Bulk download of full regional price lists (GetPriceListFileUrl returns a stub URL).
Savings Plans / Reserved Instance pricing terms beyond the bundled snapshot.

Refs: https://github.com/orgs/floci-io/discussions/791

ShubhamDX · 2026-05-12T18:14:17Z

For reviewer context — this is PR 1 of the 3-PR scope discussed in #791. Tracking issues for the remaining two:

feat: add Cost Explorer (ce:*) service #822 — Cost Explorer (ce:*) service (consumes this PR's Pricing snapshot)
feat: add CUR + BCM Data Exports services #823 — CUR + BCM Data Exports services (consumes the CE synthesis from feat: add Cost Explorer (ce:*) service #822)

Each PR is independently reviewable; this one does not depend on the others landing.

ShubhamDX · 2026-05-12T18:20:45Z

Marked this ready for review. CI workflows are sitting in action_required state - looks like the first-time-contributor approval gate. Waiting on a maintainer to click "Approve and run". Happy to address anything CI flags.

Implements the `AWSPriceListService.*` JSON 1.1 surface backed by a bundled static snapshot on the classpath. Covers the five public operations: `DescribeServices`, `GetAttributeValues`, `GetProducts`, `ListPriceLists`, and `GetPriceListFileUrl`, with pagination (Base64 offset tokens) and the array-of-JSON-strings `PriceList` shape the real AWS SDKs expect. Rationale: Floci today cannot back FinOps tooling (cost detection, budget guards, CUR parsers, etc.) because the billing APIs are absent. Pricing is the stateless prerequisite for the Cost Explorer and CUR services that follow — cost lines are synthesized as `resource state x pricing snapshot`, so the snapshot data path is the foundation. Snapshot layout on the classpath (and mirrored on the filesystem when `FLOCI_SERVICES_PRICING_SNAPSHOT_PATH` is set): pricing-snapshots/services.json pricing-snapshots/attribute-values/<ServiceCode>/<Attr>.json pricing-snapshots/products/<ServiceCode>/<Region>.json pricing-snapshots/price-lists/<ServiceCode>.json The bundled snapshot is intentionally minimal — `AmazonEC2`, `AmazonS3`, and `AWSLambda` in `us-east-1` — to exercise SDK parsing and filter logic without bloating the image. Users needing broader coverage point the override env var at a full snapshot generated from the AWS Price List bulk API. Wiring follows the existing JSON 1.1 stateless pattern (Textract, Secrets Manager): descriptor in `ResolvedServiceCatalog`, handler injected into `AwsJson11Controller`, service config on `EmulatorConfig.ServicesConfig`. No new protocol, no changes to existing services, no new runtime dependencies. Tests: `PricingIntegrationTest` covers all five operations via RestAssured JSON 1.1 wire-format assertions, including validation errors, unknown-service errors, filter matching, and pagination tokens (17 tests, all green). Refs: https://github.com/orgs/floci-io/discussions/791

hectorvent

Thanks @ShubhamDX

cfranzen · 2026-05-13T20:42:18Z

I tried to use this service from testcontainers and there seems to be an issue with the PricingService.load() method. It looks to me as if this method is never called when running Floci as container. I can't see any log output from the PricingService in the logs and the load() method should always log some information. When calling the pricing service via the API, there seem to be no service codes available.

ShubhamDX · 2026-05-14T03:20:49Z

Thanks for the report @cfranzen — confirmed: the bundled pricing-snapshots/** tree was missing from the native-image build, so the standard container had no JSON files to read. load() returns silently before logging because the classpath lookup returns null rather than going down the WARN path.

Hotfix up at #836: registers the tree under quarkus.native.resources.includes and switches to the class's own classloader (the thread-context loader can be the bootstrap loader in a native image and miss application resources). Should restore the service-code list in the container build.

ShubhamDX · 2026-05-14T06:12:36Z

@cfranzen — verified the fix locally against the native build (full log on #836). DescribeServices now returns the expected three services (AmazonEC2, AmazonS3, AWSLambda) and GetProducts returns rate data. The missing Pricing snapshot loaded log line is now present in the container output, confirming @PostConstruct runs as intended.

ShubhamDX · 2026-05-14T06:19:56Z

@cfranzen — thanks again for catching this. The original PR was working on the JVM image (and all the tests in PricingIntegrationTest pass under ./mvnw test, which boots Quarkus on the JVM), so the snapshot loader looked fine in CI. The bug was specific to the GraalVM native binary that ships in the standard floci/floci:latest image, which is what Testcontainers pulls by default.

Two GraalVM-only failures combined:

Resource bundling. Quarkus's static resource analysis only includes classpath resources it can prove are read. The pricing snapshot path is built dynamically ("pricing-snapshots/" + relativePath), so GraalVM stripped every JSON file out of the native image. On the JVM the files are still on the classpath, so the loader worked.
Classloader. Thread.currentThread().getContextClassLoader() returned the bootstrap loader inside the native image, which doesn't see application resources. On the JVM it returns the application loader, so even without (1) the read would have succeeded.

Fix in #836:

Register pricing-snapshots/** under quarkus.native.resources.includes so GraalVM bundles the tree.
Read via PricingService.class.getClassLoader() so the same code path works on JVM and native.

Verified locally with docker build -f docker/Dockerfile.native + container run — DescribeServices now returns the expected three services and the missing Pricing snapshot loaded log line shows up. Full evidence on #836.

Implements the `AWSInsightsIndexService.*` JSON 1.1 surface backed by synthesizing cost and usage from Floci's existing resource state multiplied by the bundled AWS Pricing snapshot (floci-io#821). Covers the nine Cost Explorer operations the FinOps tooling ecosystem most commonly exercises: - `GetCostAndUsage` / `GetCostAndUsageWithResources` — full `TimePeriod`, `Granularity` (DAILY / MONTHLY / HOURLY), `Metrics` (UnblendedCost / BlendedCost / AmortizedCost / NetUnblendedCost / NetAmortizedCost / UsageQuantity / NormalizedUsageAmount), `GroupBy` (DIMENSION / TAG / COST_CATEGORY), and the recursive `Filter` expression tree (`And` / `Or` / `Not` / `Dimensions` / `Tags` / `CostCategories`). - `GetDimensionValues` — returns the dimension values present in the synthesized data; honors filters and search strings. - `GetTags` — returns tag keys / values across enumerated resources. - `GetReservationCoverage` / `GetReservationUtilization` — zero-totalled stubs so callers that hit these endpoints during smoke tests don't fail on `UnknownOperationException`. - `GetSavingsPlansCoverage` / `GetSavingsPlansUtilization` — same stubbed shape. - `GetCostCategories` — empty list (cost-category management is out-of-scope for this PR). `GROUP_BY=RECORD_TYPE` distinguishes `Usage`, `Credit`, `Tax`, `Refund`, `DiscountedUsage`, and the four `SavingsPlan*` variants. `Tax` / `Refund` / `DiscountedUsage` / `SavingsPlan*` are reserved for follow-up PRs; `Credit` is emitted when `FLOCI_SERVICES_CE_CREDIT_USD_MONTHLY > 0`, capped at the synthesized monthly usage so net cost never goes below zero. ## Architecture A new SPI lives in `core/common/`: public interface ResourceUsageEnumerator { Stream<UsageLine> enumerate(Instant start, Instant end, String region); } CDI auto-discovers all `@ApplicationScoped` beans implementing it. Cost Explorer iterates `Instance<ResourceUsageEnumerator>` per request, so adding a new Floci service with cost data is a matter of dropping a new enumerator bean next to the service — zero changes to `CostExplorerService`. The same enumerators will feed the upcoming CUR and BCM Data Exports Parquet writer (floci-io#823) without duplication. The bundled enumerators are: - `Ec2UsageEnumerator` — `BoxUsage:<instanceType>` * hours per running instance, priced from the Pricing snapshot. - `S3UsageEnumerator` — `TimedStorage-ByteHrs` * GB-month per bucket, priced from the snapshot. - `LambdaUsageEnumerator` — catalog-only line per function (zero quantity until invocation tracking lands in a follow-up). - `UnpricedServicesEnumerator` — emits zero-quantity catalog rows for the remaining ~30 Floci services so they appear in `GetDimensionValues SERVICE` without contributing billed cost. Each priced enumerator additionally emits a zero-quantity catalog line so the service shows up in `GetDimensionValues SERVICE` even when no resources exist. ## Tests `CostExplorerIntegrationTest` covers all nine operations via RestAssured wire-format assertions: granularity bucketing, filter-expression evaluation, group-by combinations, dimension and tag listings, RI / SP stubs, validation errors. 20 tests, all green. The full Floci test suite stays green with the additions (4030 tests run, 0 failures, 0 errors, excluding the pre-existing local-env flakes documented during PR floci-io#821). Refs: https://github.com/orgs/floci-io/discussions/791 Refs: floci-io#822

Implements the `AWSInsightsIndexService.*` JSON 1.1 surface backed by synthesizing cost and usage from Floci's existing resource state multiplied by the bundled AWS Pricing snapshot (#821). Covers the nine Cost Explorer operations the FinOps tooling ecosystem most commonly exercises: - `GetCostAndUsage` / `GetCostAndUsageWithResources` — full `TimePeriod`, `Granularity` (DAILY / MONTHLY / HOURLY), `Metrics` (UnblendedCost / BlendedCost / AmortizedCost / NetUnblendedCost / NetAmortizedCost / UsageQuantity / NormalizedUsageAmount), `GroupBy` (DIMENSION / TAG / COST_CATEGORY), and the recursive `Filter` expression tree (`And` / `Or` / `Not` / `Dimensions` / `Tags` / `CostCategories`). - `GetDimensionValues` — returns the dimension values present in the synthesized data; honors filters and search strings. - `GetTags` — returns tag keys / values across enumerated resources. - `GetReservationCoverage` / `GetReservationUtilization` — zero-totalled stubs so callers that hit these endpoints during smoke tests don't fail on `UnknownOperationException`. - `GetSavingsPlansCoverage` / `GetSavingsPlansUtilization` — same stubbed shape. - `GetCostCategories` — empty list (cost-category management is out-of-scope for this PR). `GROUP_BY=RECORD_TYPE` distinguishes `Usage`, `Credit`, `Tax`, `Refund`, `DiscountedUsage`, and the four `SavingsPlan*` variants. `Tax` / `Refund` / `DiscountedUsage` / `SavingsPlan*` are reserved for follow-up PRs; `Credit` is emitted when `FLOCI_SERVICES_CE_CREDIT_USD_MONTHLY > 0`, capped at the synthesized monthly usage so net cost never goes below zero. ## Architecture A new SPI lives in `core/common/`: public interface ResourceUsageEnumerator { Stream<UsageLine> enumerate(Instant start, Instant end, String region); } CDI auto-discovers all `@ApplicationScoped` beans implementing it. Cost Explorer iterates `Instance<ResourceUsageEnumerator>` per request, so adding a new Floci service with cost data is a matter of dropping a new enumerator bean next to the service — zero changes to `CostExplorerService`. The same enumerators will feed the upcoming CUR and BCM Data Exports Parquet writer (#823) without duplication. The bundled enumerators are: - `Ec2UsageEnumerator` — `BoxUsage:<instanceType>` * hours per running instance, priced from the Pricing snapshot. - `S3UsageEnumerator` — `TimedStorage-ByteHrs` * GB-month per bucket, priced from the snapshot. - `LambdaUsageEnumerator` — catalog-only line per function (zero quantity until invocation tracking lands in a follow-up). - `UnpricedServicesEnumerator` — emits zero-quantity catalog rows for the remaining ~30 Floci services so they appear in `GetDimensionValues SERVICE` without contributing billed cost. Each priced enumerator additionally emits a zero-quantity catalog line so the service shows up in `GetDimensionValues SERVICE` even when no resources exist. ## Tests `CostExplorerIntegrationTest` covers all nine operations via RestAssured wire-format assertions: granularity bucketing, filter-expression evaluation, group-by combinations, dimension and tag listings, RI / SP stubs, validation errors. 20 tests, all green. The full Floci test suite stays green with the additions (4030 tests run, 0 failures, 0 errors, excluding the pre-existing local-env flakes documented during PR #821). Refs: https://github.com/orgs/floci-io/discussions/791 Refs: #822

@cfranzen

* fix(pricing): bundle snapshot resources into native image The bundled pricing-snapshots tree was missing from the GraalVM native-image build, so the standard Floci container (which ships the native binary) had no service codes available — the Pricing snapshot loader silently returned null because the JSON files weren't present at runtime. Two changes: 1. Register `pricing-snapshots/**` under `quarkus.native.resources.includes` so GraalVM bundles every snapshot file into the image. Quarkus's static analysis cannot detect resource paths constructed dynamically (the loader joins `pricing-snapshots/` with a per-call relative path), so an explicit include is required. 2. Read snapshot resources via the class's own `ClassLoader` rather than the thread context loader. In a native image the thread context loader can be the bootstrap loader, which sees no application resources, while the class's loader sees everything bundled into the image. Reported by @cfranzen against #821 — Pricing service returned no service codes when called via Testcontainers and `load()` produced no log output, indicating the snapshot files were absent rather than malformed. Refs: #821 (comment) * ci: retrigger after sdk-test-node flake on apigatewayv2 websocket disconnect

…ports Implements PR 3 of the 3-PR billing-API scope discussed in floci-io#791. Adds two service surfaces that share one underlying export pipeline: - `cur:*` — legacy Cost and Usage Reports management plane, target prefix `AWSOrigamiServiceGatewayService.*`. - `bcm-data-exports:*` — modern Billing and Cost Management Data Exports management plane, target prefix `AWSBillingAndCostManagementDataExports.*`. Both run as JSON 1.1 services dispatched through the existing `AwsJson11Controller`. ## Operations CUR (7): - `PutReportDefinition`, `ModifyReportDefinition`, `DescribeReportDefinitions`, `DeleteReportDefinition` - `TagResource` / `UntagResource` / `ListTagsForResource` (stubs) BCM Data Exports (7): - `CreateExport`, `GetExport`, `ListExports`, `UpdateExport`, `DeleteExport` - `ListExecutions`, `GetExecution` AWS-shaped error codes throughout: `DuplicateReportNameException`, `ReportLimitReachedException`, `ReportNotFoundException`, `ResourceNotFoundException`, `ValidationException`, `UnknownOperationException`. Storage keys are account-scoped (`<accountId>::<region>::<reportName>` for CUR; `<accountId>::<exportArn>` and `<accountId>::<exportArn>::<executionId>` for BCM) so both services participate in Floci's per-account isolation. ## Parquet emission The same `ResourceUsageEnumerator` SPI introduced in floci-io#832 (Cost Explorer) backs the Parquet artifact: 1. `EmissionEngine` collects `UsageLine` rows from every `@ApplicationScoped` enumerator bean. 2. `FocusRowProjector` projects them into FOCUS 1.2 / CUR 2.0 column shape, reusing floci-io#821's bundled Pricing snapshot for unit prices. 3. Rows are staged as newline-delimited JSON in `s3://floci-cur-staging/...` via Floci's S3 service — escape-safe for tags, descriptions, and resource identifiers. 4. The `floci-duck` sidecar runs `COPY (SELECT * FROM read_json_auto('s3://...')) TO 's3://...' (FORMAT PARQUET)` so Parquet writes happen inside DuckDB, not in the Java process. This avoids pulling Apache Parquet / Hadoop / Avro into Floci's classpath and keeps the native binary footprint unchanged. 5. Staging objects are removed in a best-effort `finally` block; each run carries a fresh `runId` (UUID) in both staging and destination keys so concurrent emissions cannot clobber each other. The emission engine is shared between CUR and BCM, but the management planes are otherwise independent — they each own their own validation, storage, and JSON dispatch. ## Run modes `FLOCI_SERVICES_CUR_EMIT_MODE` and `FLOCI_SERVICES_BCM_DATA_EXPORTS_EMIT_MODE`: - `synchronous` (default): emit on every report/export mutation - `daily`: emit once per 24h via a CUR-owned single-thread `ScheduledExecutorService` — separate from the EventBridge Scheduler dispatcher, so internal billing exports do not surface as user-visible Scheduler resources. - `off`: management plane only Emission failures never roll back the management mutation. Failures surface as `ReportStatus.LastStatus = ERROR` on CUR and `Execution.ExecutionStatus.StatusCode = DELIVERY_FAILURE` on BCM. ## Tests 42 new tests across 5 classes: - `CurIntegrationTest` (11) — full CRUD on report definitions, duplicate / not-found / validation paths - `BcmDataExportsIntegrationTest` (16) — full CRUD on exports + executions, all validation paths - `FocusRowProjectorTest` (9) — pure-unit projection: usage rows, USD-denominated record types (Credit / Refund / Tax), missing resource id, tags JSON, billing-period anchoring, region+account flow-through, rate cache reuse - `ParquetEmitterIntegrationTest` (3) — DuckDB readback, staging cleanup, concurrent run isolation - `CurEmissionIntegrationTest` (3) — end-to-end: PutReportDefinition -> emit -> Parquet readable through DuckDB; ReportStatus updates; CreateExport produces an execution record Full Floci suite stays green (4194/4194, excluding the pre-existing local-env flakes documented during PR floci-io#821 / floci-io#832 and the two duck-required tests added here that need the `floci-duck` sidecar and a fixed Floci HTTP port to verify cross-container S3 access). ## Documentation - `docs/services/cur.md` — operations, validation rules, emission pipeline, configuration, examples. - `docs/services/bcm-data-exports.md` — operations, validation, execution lifecycle, configuration, examples. - README service table updated with rows for both, plus a note that CUR / BCM Data Exports share one Parquet engine. Refs: https://github.com/orgs/floci-io/discussions/791 Refs: floci-io#823 Builds on: floci-io#821, floci-io#832

…ports (#850) Implements PR 3 of the 3-PR billing-API scope discussed in #791. Adds two service surfaces that share one underlying export pipeline: - `cur:*` — legacy Cost and Usage Reports management plane, target prefix `AWSOrigamiServiceGatewayService.*`. - `bcm-data-exports:*` — modern Billing and Cost Management Data Exports management plane, target prefix `AWSBillingAndCostManagementDataExports.*`. Both run as JSON 1.1 services dispatched through the existing `AwsJson11Controller`. ## Operations CUR (7): - `PutReportDefinition`, `ModifyReportDefinition`, `DescribeReportDefinitions`, `DeleteReportDefinition` - `TagResource` / `UntagResource` / `ListTagsForResource` (stubs) BCM Data Exports (7): - `CreateExport`, `GetExport`, `ListExports`, `UpdateExport`, `DeleteExport` - `ListExecutions`, `GetExecution` AWS-shaped error codes throughout: `DuplicateReportNameException`, `ReportLimitReachedException`, `ReportNotFoundException`, `ResourceNotFoundException`, `ValidationException`, `UnknownOperationException`. Storage keys are account-scoped (`<accountId>::<region>::<reportName>` for CUR; `<accountId>::<exportArn>` and `<accountId>::<exportArn>::<executionId>` for BCM) so both services participate in Floci's per-account isolation. ## Parquet emission The same `ResourceUsageEnumerator` SPI introduced in #832 (Cost Explorer) backs the Parquet artifact: 1. `EmissionEngine` collects `UsageLine` rows from every `@ApplicationScoped` enumerator bean. 2. `FocusRowProjector` projects them into FOCUS 1.2 / CUR 2.0 column shape, reusing #821's bundled Pricing snapshot for unit prices. 3. Rows are staged as newline-delimited JSON in `s3://floci-cur-staging/...` via Floci's S3 service — escape-safe for tags, descriptions, and resource identifiers. 4. The `floci-duck` sidecar runs `COPY (SELECT * FROM read_json_auto('s3://...')) TO 's3://...' (FORMAT PARQUET)` so Parquet writes happen inside DuckDB, not in the Java process. This avoids pulling Apache Parquet / Hadoop / Avro into Floci's classpath and keeps the native binary footprint unchanged. 5. Staging objects are removed in a best-effort `finally` block; each run carries a fresh `runId` (UUID) in both staging and destination keys so concurrent emissions cannot clobber each other. The emission engine is shared between CUR and BCM, but the management planes are otherwise independent — they each own their own validation, storage, and JSON dispatch. ## Run modes `FLOCI_SERVICES_CUR_EMIT_MODE` and `FLOCI_SERVICES_BCM_DATA_EXPORTS_EMIT_MODE`: - `synchronous` (default): emit on every report/export mutation - `daily`: emit once per 24h via a CUR-owned single-thread `ScheduledExecutorService` — separate from the EventBridge Scheduler dispatcher, so internal billing exports do not surface as user-visible Scheduler resources. - `off`: management plane only Emission failures never roll back the management mutation. Failures surface as `ReportStatus.LastStatus = ERROR` on CUR and `Execution.ExecutionStatus.StatusCode = DELIVERY_FAILURE` on BCM. ## Tests 42 new tests across 5 classes: - `CurIntegrationTest` (11) — full CRUD on report definitions, duplicate / not-found / validation paths - `BcmDataExportsIntegrationTest` (16) — full CRUD on exports + executions, all validation paths - `FocusRowProjectorTest` (9) — pure-unit projection: usage rows, USD-denominated record types (Credit / Refund / Tax), missing resource id, tags JSON, billing-period anchoring, region+account flow-through, rate cache reuse - `ParquetEmitterIntegrationTest` (3) — DuckDB readback, staging cleanup, concurrent run isolation - `CurEmissionIntegrationTest` (3) — end-to-end: PutReportDefinition -> emit -> Parquet readable through DuckDB; ReportStatus updates; CreateExport produces an execution record Full Floci suite stays green (4194/4194, excluding the pre-existing local-env flakes documented during PR #821 / #832 and the two duck-required tests added here that need the `floci-duck` sidecar and a fixed Floci HTTP port to verify cross-container S3 access). ## Documentation - `docs/services/cur.md` — operations, validation rules, emission pipeline, configuration, examples. - `docs/services/bcm-data-exports.md` — operations, validation, execution lifecycle, configuration, examples. - README service table updated with rows for both, plus a note that CUR / BCM Data Exports share one Parquet engine. Refs: https://github.com/orgs/floci-io/discussions/791 Refs: #823 Builds on: #821, #832

Implements the `AWSPriceListService.*` JSON 1.1 surface backed by a bundled static snapshot on the classpath. Covers the five public operations: `DescribeServices`, `GetAttributeValues`, `GetProducts`, `ListPriceLists`, and `GetPriceListFileUrl`, with pagination (Base64 offset tokens) and the array-of-JSON-strings `PriceList` shape the real AWS SDKs expect. Rationale: Floci today cannot back FinOps tooling (cost detection, budget guards, CUR parsers, etc.) because the billing APIs are absent. Pricing is the stateless prerequisite for the Cost Explorer and CUR services that follow — cost lines are synthesized as `resource state x pricing snapshot`, so the snapshot data path is the foundation. Snapshot layout on the classpath (and mirrored on the filesystem when `FLOCI_SERVICES_PRICING_SNAPSHOT_PATH` is set): pricing-snapshots/services.json pricing-snapshots/attribute-values/<ServiceCode>/<Attr>.json pricing-snapshots/products/<ServiceCode>/<Region>.json pricing-snapshots/price-lists/<ServiceCode>.json The bundled snapshot is intentionally minimal — `AmazonEC2`, `AmazonS3`, and `AWSLambda` in `us-east-1` — to exercise SDK parsing and filter logic without bloating the image. Users needing broader coverage point the override env var at a full snapshot generated from the AWS Price List bulk API. Wiring follows the existing JSON 1.1 stateless pattern (Textract, Secrets Manager): descriptor in `ResolvedServiceCatalog`, handler injected into `AwsJson11Controller`, service config on `EmulatorConfig.ServicesConfig`. No new protocol, no changes to existing services, no new runtime dependencies. Tests: `PricingIntegrationTest` covers all five operations via RestAssured JSON 1.1 wire-format assertions, including validation errors, unknown-service errors, filter matching, and pagination tokens (17 tests, all green). Refs: https://github.com/orgs/floci-io/discussions/791

Implements the `AWSInsightsIndexService.*` JSON 1.1 surface backed by synthesizing cost and usage from Floci's existing resource state multiplied by the bundled AWS Pricing snapshot (floci-io#821). Covers the nine Cost Explorer operations the FinOps tooling ecosystem most commonly exercises: - `GetCostAndUsage` / `GetCostAndUsageWithResources` — full `TimePeriod`, `Granularity` (DAILY / MONTHLY / HOURLY), `Metrics` (UnblendedCost / BlendedCost / AmortizedCost / NetUnblendedCost / NetAmortizedCost / UsageQuantity / NormalizedUsageAmount), `GroupBy` (DIMENSION / TAG / COST_CATEGORY), and the recursive `Filter` expression tree (`And` / `Or` / `Not` / `Dimensions` / `Tags` / `CostCategories`). - `GetDimensionValues` — returns the dimension values present in the synthesized data; honors filters and search strings. - `GetTags` — returns tag keys / values across enumerated resources. - `GetReservationCoverage` / `GetReservationUtilization` — zero-totalled stubs so callers that hit these endpoints during smoke tests don't fail on `UnknownOperationException`. - `GetSavingsPlansCoverage` / `GetSavingsPlansUtilization` — same stubbed shape. - `GetCostCategories` — empty list (cost-category management is out-of-scope for this PR). `GROUP_BY=RECORD_TYPE` distinguishes `Usage`, `Credit`, `Tax`, `Refund`, `DiscountedUsage`, and the four `SavingsPlan*` variants. `Tax` / `Refund` / `DiscountedUsage` / `SavingsPlan*` are reserved for follow-up PRs; `Credit` is emitted when `FLOCI_SERVICES_CE_CREDIT_USD_MONTHLY > 0`, capped at the synthesized monthly usage so net cost never goes below zero. ## Architecture A new SPI lives in `core/common/`: public interface ResourceUsageEnumerator { Stream<UsageLine> enumerate(Instant start, Instant end, String region); } CDI auto-discovers all `@ApplicationScoped` beans implementing it. Cost Explorer iterates `Instance<ResourceUsageEnumerator>` per request, so adding a new Floci service with cost data is a matter of dropping a new enumerator bean next to the service — zero changes to `CostExplorerService`. The same enumerators will feed the upcoming CUR and BCM Data Exports Parquet writer (floci-io#823) without duplication. The bundled enumerators are: - `Ec2UsageEnumerator` — `BoxUsage:<instanceType>` * hours per running instance, priced from the Pricing snapshot. - `S3UsageEnumerator` — `TimedStorage-ByteHrs` * GB-month per bucket, priced from the snapshot. - `LambdaUsageEnumerator` — catalog-only line per function (zero quantity until invocation tracking lands in a follow-up). - `UnpricedServicesEnumerator` — emits zero-quantity catalog rows for the remaining ~30 Floci services so they appear in `GetDimensionValues SERVICE` without contributing billed cost. Each priced enumerator additionally emits a zero-quantity catalog line so the service shows up in `GetDimensionValues SERVICE` even when no resources exist. ## Tests `CostExplorerIntegrationTest` covers all nine operations via RestAssured wire-format assertions: granularity bucketing, filter-expression evaluation, group-by combinations, dimension and tag listings, RI / SP stubs, validation errors. 20 tests, all green. The full Floci test suite stays green with the additions (4030 tests run, 0 failures, 0 errors, excluding the pre-existing local-env flakes documented during PR floci-io#821). Refs: https://github.com/orgs/floci-io/discussions/791 Refs: floci-io#822

@cfranzen

* fix(pricing): bundle snapshot resources into native image The bundled pricing-snapshots tree was missing from the GraalVM native-image build, so the standard Floci container (which ships the native binary) had no service codes available — the Pricing snapshot loader silently returned null because the JSON files weren't present at runtime. Two changes: 1. Register `pricing-snapshots/**` under `quarkus.native.resources.includes` so GraalVM bundles every snapshot file into the image. Quarkus's static analysis cannot detect resource paths constructed dynamically (the loader joins `pricing-snapshots/` with a per-call relative path), so an explicit include is required. 2. Read snapshot resources via the class's own `ClassLoader` rather than the thread context loader. In a native image the thread context loader can be the bootstrap loader, which sees no application resources, while the class's loader sees everything bundled into the image. Reported by @cfranzen against floci-io#821 — Pricing service returned no service codes when called via Testcontainers and `load()` produced no log output, indicating the snapshot files were absent rather than malformed. Refs: floci-io#821 (comment) * ci: retrigger after sdk-test-node flake on apigatewayv2 websocket disconnect

This was referenced May 12, 2026

feat: add Cost Explorer (ce:*) service #822

Closed

feat: add CUR + BCM Data Exports services #823

Closed

ShubhamDX marked this pull request as ready for review May 12, 2026 18:18

hectorvent requested changes May 13, 2026

View reviewed changes

Comment thread src/main/java/io/github/hectorvent/floci/config/EmulatorConfig.java Outdated

Comment thread src/main/java/io/github/hectorvent/floci/services/pricing/PricingService.java

hectorvent added enhancement New feature or request pricing waiting-contributor labels May 13, 2026

ShubhamDX force-pushed the feat/pricing-api branch from c4cbd9b to 9ed87c0 Compare May 13, 2026 02:54

hectorvent approved these changes May 13, 2026

View reviewed changes

hectorvent merged commit e0224e9 into floci-io:main May 13, 2026
10 checks passed

ShubhamDX deleted the feat/pricing-api branch May 13, 2026 07:12

chenrui333 mentioned this pull request May 13, 2026

floci 1.5.15 chenrui333/homebrew-tap#6963

Closed

ShubhamDX mentioned this pull request May 13, 2026

feat(ce): add Cost Explorer service #832

Merged

7 tasks

ShubhamDX mentioned this pull request May 14, 2026

fix(pricing): bundle snapshot resources into native image #836

Merged

2 tasks

ShubhamDX mentioned this pull request May 15, 2026

feat(cur, bcm-data-exports): add Cost and Usage Reports + BCM Data Exports #850

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat(pricing): add AWS Price List Service support#821

feat(pricing): add AWS Price List Service support#821
hectorvent merged 1 commit into
floci-io:mainfrom
ShubhamDX:feat/pricing-api

ShubhamDX commented May 12, 2026

Uh oh!

ShubhamDX commented May 12, 2026

Uh oh!

ShubhamDX commented May 12, 2026 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

hectorvent left a comment

Uh oh!

Uh oh!

cfranzen commented May 13, 2026 •

edited

Loading

Uh oh!

ShubhamDX commented May 14, 2026

Uh oh!

ShubhamDX commented May 14, 2026

Uh oh!

ShubhamDX commented May 14, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

ShubhamDX commented May 12, 2026

Summary

Why

Design choices

Test plan

Scope boundaries (out of scope for this PR)

Uh oh!

ShubhamDX commented May 12, 2026

Uh oh!

ShubhamDX commented May 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

hectorvent left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

cfranzen commented May 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ShubhamDX commented May 14, 2026

Uh oh!

ShubhamDX commented May 14, 2026

Uh oh!

ShubhamDX commented May 14, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

ShubhamDX commented May 12, 2026 •

edited

Loading

cfranzen commented May 13, 2026 •

edited

Loading