Skip to content

CIS API reporting performance audit and architecture proposal

Date: 4 November 2026
Status: Source audit and proposed design
Audience: Backend engineers, database engineers and technical leads

Executive recommendation

Prioritize bounded CSV extraction and safe filter binding, followed by incremental productivity generation. Introduce durable background export jobs for large reports and consolidate compatible dashboard aggregates. Tune Oracle indexes and reporting views after collecting execution statistics. These changes target memory growth, repeated database work and competition with transactional API traffic.

The findings below describe source behavior, not measured production latency. No live database execution plans, deployed view definitions, row counts or load-test results were collected. Severity represents likely scaling impact. No runtime code or database changes are included in this proposal.

Scope and current flow

The audit covers ReportingResource, ReportingService, GenericReportFactory, representative archive and examination production exports, user summary exports, ReportUtils, productivity generation and general dashboard counters. It is not an exhaustive correctness certification of every report or the separate archive and examination dashboards.

flowchart TD
    A[Client requests PDF or CSV] --> B[ReportingResource]
    B --> C[ReportingService and report factory]
    C --> D[Generator builds SQL and queries Oracle]
    D --> E[All result rows loaded into a List]
    E --> F{Format}
    F -->|CSV| G[Write complete list to temporary file]
    F -->|PDF| H[Build HTML then XHTML and PDF layout]
    G --> I[Return file after generation completes]
    H --> I

F1 Complete result sets are retained in memory

Priority: High. ExaminationProductionReport.generateReportData and ArchiveProductionReport.generateReportData return a complete JDBC result List. CSV writers consume that list; PDF generators place the full collection in a Thymeleaf context. PDF additionally creates HTML, XHTML and renderer state. Memory use grows with matching rows and concurrent exports. Archive and examination production queries also sort by date, which can add database sort work.

Remedy: Process JDBC rows with a callback and write CSV incrementally to a buffered artifact file. Keep only a bounded fetch batch and output buffer in memory. Use proper CSV escaping, explicit UTF-8 and try-with-resources. Cap PDF rows or pages and offer detailed CSV for large datasets. Do not silently truncate an export. Keep deterministic ordering with a unique tie-breaker when chunking.

Acceptance: Compare 10 thousand and 100 thousand row exports under identical concurrency. CSV peak heap should remain within an agreed bounded budget rather than grow with the full dataset. Validate row counts, ordering, commas, quotes, newlines and Unicode. Fetch size alone does not remove the retained List.

F2 Productivity generation rebuilds historical actions

Priority: High. WorkflowRepository.getProductivityData selects all workflow actions. ProductivityReportGenerator.generateWorkflowProductivity runs in one transaction, deletes prior results per action and inserts daily results. Work grows with total action-days and retains a large persistence context. Existing Hibernate batching settings do not eliminate that algorithmic cost.

Remedy: Select changed actions plus open actions, using a stable checkpoint and keyset chunks. Bulk replace or merge derived rows within short transactions. Flush and clear between chunks when using JPA. Record successful checkpoints only after the corresponding writes commit. Add uniqueness at the agreed action and day grain after checking existing duplicates. Recalculate affected historical periods when calendars, leave or office hours change.

Acceptance: Reprocessing an identical chunk produces the same results without duplicates. Crash recovery neither skips actions nor exposes partially replaced results for an action. Track processed actions, generated days, elapsed time and failed chunks. Preserve source workflow data.

F3 Productivity loops repeat calendar database lookups

Priority: High. Each action-day can query national holidays, office holidays, user leave and office timings before saving a daily record. The loop is O(total action-days), with up to four reference reads per normal working day in the visible call path.

Remedy: Bulk-load relevant calendar dates, province timing intervals and leave intervals for each processing chunk. Resolve date membership in memory. Bound cache size and use run-scoped snapshots for reproducibility. Define business timezone, weekend behavior, partial-day treatment and retroactive calendar corrections before replacing the calculation.

Acceptance: Reference-query counts scale with chunks and required reference ranges rather than individual action-days. Cover same-day work, first and last partial days, after-hours periods, holidays, leave, open actions and missing office timings.

F4 Dynamic filter SQL contains request supplied structure and literals

Priority: High. ReportUtils.getQueryFromFilters concatenates dbCaption, dbOperation, prefixes and data supplied through ReportFiltersData. This creates SQL variations and an SQL injection risk. ReportingVm does not enforce a maximum date span, format enumeration or nonempty province list. The inspected dynamic production generators do not separately bind the top-level province and date fields.

Remedy: Accept stable filter identifiers and typed values. Resolve SQL columns, allowed operators and mandatory scope predicates from server-owned definitions. Bind values through NamedParameterJdbcTemplate. Validate report ID, format, date order, report-specific required filters, list sizes and authorized province scope. Never trust a supplied column or function name. Treat absent filters explicitly rather than emitting an empty WHERE clause.

Acceptance: Reject unknown filters and operators; preserve valid filter semantics. Verify that province restrictions cannot be omitted or widened through dynamic filters. Compare SQL text stability for requests differing only in values. Test malicious strings as bound data.

F5 Dashboard counters execute separate sequential queries

Priority: Medium to high. DashBoardService builds several counter responses from four repository calls, including registration, user summary, process, reservation and lodgement counters. Repeated requests can repeat scans over overlapping source populations. Actual scan cost depends on deployed views and Oracle plans.

Remedy: Consolidate compatible counters with conditional aggregation over a common filtered dataset. Aggregate each metric at its correct entity grain before joining to avoid inflated counts or money totals. Retain separate queries when populations differ. Add short-lived caching only where freshness requirements permit; include authorization scope, normalized filters, timezone and data version in the key.

Acceptance: Compare every old and new metric on fixtures with multiple roles, actions, invoices and payments. Record database calls and logical reads per dashboard response. Group months by a chronological month key; format labels at the presentation boundary.

F6 Date expressions and JDBC settings need targeted tuning

Priority: Medium. UserSummaryReport filters with trunc(USERCREATIONDATE) between date parameters. This can prevent an ordinary timestamp index from serving a range efficiently unless Oracle can transform the predicate or an appropriate function-based index exists. JDBC report queries do not inherit Hibernate fetch-size settings.

Remedy: Prefer timestamp >= :fromInclusive AND timestamp < :toExclusive with the next day as the exclusive end boundary, using the agreed timezone. Set report-specific JDBC fetch sizes and statement timeouts. Evaluate base-table composite indexes against real selectivity and execution plans. Do not add an index to every filter column or assume a view name implies precomputed data.

Acceptance: Inspect executed plans for narrow and broad date ranges and one or many provinces. Compare actual rows, logical reads, elapsed time and sort spills. Verify inclusion of the full final day. Choose indexes only after inspecting deployed view definitions, statistics and existing indexes.

F7 Synchronous generation and unmanaged artifacts reduce isolation

Priority: Medium. ReportingResource waits for generation before returning a file. ReportUtils and individual generators create temporary files; no successful-download cleanup is visible in the inspected report path. Report execution shares the application's database resources. The checked application configuration defaults the Hikari maximum to 15, but runtime overrides were not verified.

Remedy: Add durable asynchronous jobs for large exports with bounded database and rendering concurrency. Store completed files in managed artifact storage with retention expiry. Publish an artifact only after successful close and validation; remove partial files on failure. Use an explicit reporting connection budget, optionally a separate small pool or worker deployment after measurement. Increasing pool size alone can increase Oracle contention.

Acceptance: A burst of exports respects configured limits and does not exhaust transactional connection capacity. Retries, cancellations, process restarts and retention sweeps leave consistent job and artifact states. Test client disconnects and failed downloads.

Proposed export architecture

Retain the strategy registry around GenericReport, while separating query extraction, rendering and job orchestration. A ReportDefinitionRegistry owns allowed filters and report capabilities. A ReportQueryAdapter produces typed, authorized, bound queries. CsvArtifactWriter streams rows; PdfArtifactRenderer renders bounded datasets. ReportJobService owns durable lifecycle and ArtifactStore owns publication and expiry.

flowchart TD
    A[Client] --> B[API validates report and authorizes scope]
    B --> C{Small bounded report}
    C -->|Yes| D[Synchronous compatibility route]
    C -->|No| E[Persist durable job and return 202 with job ID]
    E --> F[Worker claims job with lease]
    F --> G[Query adapter with bound filters]
    D --> G
    G --> H[Oracle within reporting connection budget]
    H --> I{Output format}
    I -->|CSV| J[Row callback and buffered artifact writer]
    I -->|PDF| K[Bounded dataset and limited PDF renderer]
    J --> L[Validate and publish artifact]
    K --> L
    L --> M[Mark job succeeded or return synchronous file]
    A --> N[Authorized status and download endpoint]
    N --> M
    L --> O[Retention service expires artifact]

Proposed API additions are POST /report/jobs, GET /report/jobs/{id}, GET /report/jobs/{id}/download and DELETE /report/jobs/{id} for cancellation. Keep existing POST /report behavior for supported bounded requests. Add the asynchronous route with a corresponding UI status and download flow; do not silently change a binary-response contract to JSON.

Persist job ID, requester, authorized scope, canonical filters, format, definition version, source cutoff, status, progress, attempts, lease expiry, timestamps, artifact reference and expiry. Use QUEUED, RUNNING, SUCCEEDED, FAILED, CANCELLED and EXPIRED states. Workers claim jobs atomically, renew leases and use bounded retries. Write attempt-specific staging artifacts so a stale worker cannot overwrite a newer result. A durable database job table is sufficient initially; a message broker is optional, with a transactional outbox if introduced.

Reauthorize download access. Never use an unprotected artifact URL as authorization. Recheck permissions before extraction; define how revoked access affects completed jobs. Deduplication and artifact reuse must include requester or equivalent authorization scope, report definition version, canonical filters, format and source version. Check cancellation between extraction chunks and rendering stages.

For a consistent export, prefer one cursor or an agreed Oracle snapshot mechanism. Multi-query keyset extraction needs an explicit source snapshot or cutoff policy; a timestamp cutoff alone does not freeze updates to existing rows. Long snapshots can consume undo and must be tested with the database team. Streaming to a file decouples Oracle connection time from a slow client download.

Initial implementation

The existing POST /report route resolves typed report strategies through ReportDefinitionRegistry and uses GenericReportFactory for legacy strategies. To enroll a report, register a Spring bean implementing ScopedReport with its own immutable ReportExportFilters type. The strategy adapts and validates the request, then receives the same typed filters after authorization. Reports that support jobs implement BackgroundCsvReport and own the versioned filter decoder. Each query adapter binds its report's typed values to server-owned SQL.

The pilot uses the shared immutable ReportFilters value object, accepting either the original top-level fields or a filters object with fromDate, toDate, provinces and userTypes. A specialized report can compose this standard scope with additional immutable criteria without changing the framework contract. Persisted version 1 filter JSON retains this shape.

User Summary requires a report grant from an assigned role in each requested province. An assignment with no province grants no scope through this path. The pilot accepts up to 366 inclusive dates and up to 100 provinces and user types per request. CSV uses a single cursor and a buffered UTF-8 file. PDF rejects results above 2,000 rows and directs users to CSV. The report endpoint removes temporary file artifacts when the request completes. These limits are initial implementation choices pending representative load tests.

Shared export contracts and the strategy registry live in reporting.core, with shared filters in core.filters and artifact writers in core.csv and core.pdf. Each migrated report owns its strategy and query adapter under reporting.reports.<domain>. Additive criteria such as Archive Production's document types live under that domain's filters subpackage. User Summary reuses the shared filter.

User Summary PDF rendering

The pilot PDF adapter renders the trusted Thymeleaf template, normalizes it to XHTML and uses Flying Saucer to lay out the document before allocating a temporary output file. It accepts the bundled report logo, flag and renderer stylesheet. Failed writes remove the partial artifact and return a sanitized HTTP 500. Completed downloads use the MVC cleanup interceptor.

The query reads at most 2,001 rows to detect the 2,000-row PDF limit. Province labels come from a scalar query. The cis.reporting.pdf.stage timer records extraction, template, layout and write separately. Use representative staging timings and heap measurements to decide whether PDF needs separate permits from synchronous CSV.

Background CSV pilot

Set CIS_REPORTING_JOBS_ENABLED=true in the CIS API module environment file on a deployment with migrations V5 and V6 applied, scheduling enabled and the existing Redis scheduler lock available. The feature defaults to off. Polling defaults to 10 seconds and is configurable with CIS_REPORTING_JOBS_POLL_DELAY. The other reporting job settings are listed in the CIS API environment examples for dev, UAT and production. Extraction and retention run on a dedicated scheduler. The worker processes one job per Redis scheduler lock and uses the shared Oracle pool. During extraction it can use a second connection briefly to check the claim every 5,000 rows.

  • POST /api/v1/report/jobs accepts the existing User Summary CSV request and responds with 202 and a job location.
  • GET /api/v1/report/jobs/{id} returns the owner's state and attempt count.
  • GET /api/v1/report/jobs/{id}/download returns CSV after rechecking the requester's current grants.
  • DELETE /api/v1/report/jobs/{id} cancels an owned queued or running job.

The feature lives in the CIS API reporting package, with separate export, job, persistence and HTTP components. Spring Data JPA stores the requester, versioned filters and lifecycle metadata. JDBC streams the Oracle extraction cursor and transfers artifact BLOBs without loading them into entity state. A token-guarded update publishes the BLOB and success state together. The worker uses a ten-minute lease and retries up to three attempts. Each attempt reads one Oracle cursor using the data visible at execution time. A cancelled or superseded attempt cannot publish. Artifacts are capped at 64 MiB during generation, expire after one day and are cleared by the worker. Retention runs independently of extraction. Set app.reporting.jobs.maintenance-enabled=true to keep retention running while submission and extraction are paused. Downloads stage the BLOB locally before releasing the database connection and streaming the file.

Legacy dynamic filter SQL, additional report enrollment and frontend polling remain outside the pilot. Validate the Oracle migration, runtime role grants, restart recovery and representative concurrency before enabling the feature.

Full Platform Enrollment & Architectural Uniformity

All 35 reports across all CIS modules are now fully enrolled into the modern reporting framework: - Reservation (6 reports): Business Rule, Notification, Production, Status, Transfer, User. - Examination (6 reports): Business Rule, Divisional, Document Capture Per User Per Day, Notification, Production, User Production. - Lodgement (5 reports): Production, Monthly Statistics, User, Notification, Business Rule. - Workflow (5 reports): Production, User Production, Notification, Business Rule, Request Statistical. - User Management (6 reports): User Summary (relocated to user management), User Login Summary, User Hourly Login Distribution, User Registration Quarterly, User Update Quarterly, User Update History. - Archive (8 reports): Archive Production, Archive Business Rule, Archive Monthly Statistics, Archive Notification, Archive User, PSDT Details, Scanning Monthly Statistics, Scanning Production. - Maintenance (5 reports): Monthly Statistics, Notification, Overridden Business Rule, Production, User.

Architectural Packaging Standards

  1. Report Strategies: com.dw.ngms.cis.reporting.reports.<domain>.<ReportName>Report
  2. Query Adapters: com.dw.ngms.cis.reporting.reports.<domain>.adapter.<ReportName>QueryAdapter (cleanly separated from the strategy package to eliminate parent package clutter).
  3. Domain Filters: com.dw.ngms.cis.reporting.reports.<domain>.filters.* (or the shared ReportFilters contract).
  4. Code Hygiene & Author Headers: Every strategy, query adapter, filter, and test class contains the required author header:
    /**
     * Description of class.
     * <p>Pattern: ...</p>
     *
     * @author Prateek goel
     * @since 2026-11-06 23:45:00
     */
    
  5. Security & Performance: Strict bounded PDF generation (2,000 row ceiling), CSV artifact ceiling (64 MiB), semaphore concurrency caps (2 concurrent synchronous renders per report), formula injection escaping, parameter-bound sargable date queries (>= :fromDate AND < :toExclusive), and zero legacy dynamic string interpolation.
  6. Legacy Code Cleanup: All 35 legacy report implementations across com.dw.ngms.cis.service.report have been removed.

Pilot operations

  • Default admission permits 32 outstanding jobs across replicas, including retained successful artifacts, three active jobs per requester and six submissions per requester per rolling minute. Change these using max-outstanding, max-active-per-requester and max-submissions-per-minute under app.reporting.jobs. Admission uses a short Oracle lock and returns HTTP 429 with Retry-After when saturated.
  • The default artifact size ceiling is 64 MiB. The outstanding limit bounds committed artifacts to at most 2 GiB at default settings, excluding Oracle LOB allocation overhead. Each API instance allows four staged background downloads by default (max-downloads) and two synchronous User Summary renders. Use an instance-local temporary volume with bounded capacity and platform cleanup after process termination. Background download responses are marked Cache-Control: no-store.
  • Maintenance clears expired artifacts and deletes up to 500 old terminal rows each pass. It runs once per minute by default and keeps 30 days of terminal metadata (maintenance-delay, history-days).
  • Report grant changes require an administrator authority. CSV rows are quoted, and formula-like text fields receive a leading apostrophe for spreadsheet viewing. Resaving in a spreadsheet can remove escaping, and downstream CSV consumers should account for the text prefix. OWASP CSV injection guidance
  • Monitor cis.reporting.jobs.queued, cis.reporting.jobs.oldest.queued.seconds, cis.reporting.jobs.events and cis.reporting.jobs.attempt through the existing Micrometer registry. Configure alerts using the target environment's latency and capacity objectives. Include Oracle tablespace and API temporary-volume usage in rollout checks.
  • The API returns a Location on submission. Poll the job resource until it reaches a terminal state, then download a successful artifact. Stop polling when the job is cancelled, failed or expired. On 429, honor Retry-After. Frontend integration is a separate deployment.

Generate a combined JaCoCo report for the framework with ./gradlew.bat :cis-api:jacocoReportingReport on a Docker-enabled host. The report combines unit and Oracle integration execution data for com.dw.ngms.cis.reporting.

For debugging, search application logs by jobId for queued, claimed, published and cancelled jobs, or by exportId for synchronous exports and PDF stage failures. Enable DEBUG for com.dw.ngms.cis.reporting and com.dw.ngms.cis.service.report.ReportingService to see PDF stage timings, rejected requests and download decisions. The reporting diagnostics use identifiers, state, elapsed time and exception types without emitting filter values or artifact contents.

Production readiness gates

  1. Migrate each exposed archive and examination dynamic-filter report to a server-owned definition and parameter binding. Verify report grants and mandatory province predicates with real report configuration and representative authorized fixtures. Treat the existing SQL concatenation paths as a security release blocker for a reporting-wide rollout.
  2. Run the Oracle job integration tests with migrations V5 and V6 and exercise submission through download with deployed authentication and Redis. Cover concurrent workers, stale leases, restart recovery, cancellation, revoked grants, expiry, failed uploads and disconnected downloads.
  3. Measure narrow and broad exports alongside transactional traffic. Establish limits for queue depth, submission rate, database connections, artifact storage and retention based on those results. Add metrics and alerts for oldest queued job, failures, retries, expiry and active connections.
  4. Connect the frontend polling, cancellation and download flow. Document operator recovery and retention behavior, including cleanup while job processing is paused. Enable the feature for a controlled User Summary pilot before expanding report enrollment.

The export framework currently enrolls User Summary, Archive Production, the full suite of Examination reports (Production, Notification, User Production, Divisional, Business Rule, and Document Capture Per User Per Day), and the full suite of Reservation reports (Production, Status, Transfer, Business Rule, Manager Notification, and User) with typed parameter-bound SQL queries and background CSV support. The productivity and dashboard phases below are separate implementation tracks.

Incremental productivity architecture

flowchart TD
    A[Scheduled run with lease] --> B[Read checkpoint and capture run cutoff]
    B --> C[Select changed and open actions in keyset chunks]
    C --> D[Bulk load calendars office hours and leave]
    D --> E[Calculate daily productivity in memory]
    E --> F[Replace derived rows and commit checkpoint atomically]
    F --> G{More chunks}
    G -->|Yes| C
    G -->|No| H[Mark run complete and publish metrics]
    I[Calendar or leave correction] --> J[Queue affected actions and dates]
    J --> C

Use a stable change sequence or a timestamp plus unique ID, with overlap and deduplication where timestamp precision requires it. Revisit open actions even without a new source change because elapsed work changes over time. Capture one business-time cutoff per run. Commit derived rows and checkpoint together; make replacement idempotent. Keep an explicit backfill command for historical rebuilds.

The existing scheduler lock has a 14 minute maximum. Confirm actual runtime and scheduler topology before choosing lease limits. Support renewal or a bounded execution contract and prevent concurrent writers for the same action. A scheduler lock alone does not coordinate separate workflow-triggered recalculation paths.

Dashboard read architecture

flowchart TD
    A[Authorized dashboard request] --> B[Normalize filters and scope]
    B --> C{Scoped cache hit}
    C -->|Yes| D[Return values with data freshness time]
    C -->|No| E[Consolidated aggregate query]
    E --> F[Oracle reporting views or daily summary tables]
    F --> G[Store bounded cache entry]
    G --> D
    H[Incremental summary refresh] --> F
    H --> I[Advance data version or invalidate affected keys]

Start with query consolidation and measurement. Add daily summaries or materialized views only for expensive repeated aggregates with an agreed freshness contract. Choose dimensions and measures from actual dashboard requirements. Counts of distinct users cannot generally be obtained by summing daily distinct counts; averages need numerators and denominators. Detail exports should continue to use an appropriate detail source.

Correctness prerequisites

ReportingResource assigns PDF content type and then unconditionally overwrites it with text/csv. Correct response headers before validating downloaded files. MaintenanceProductionReport currently selects blank or null columns from VW_WH_BUSINESS_RULES with ROWNUM < 10; its parameters are not used. Treat this as incomplete reporting behavior, not a successful fast report.

ReportingService.getReport dispatches directly to a generator without a visible report-role permission check in that method. Verify the full security path and enforce report and province scope on execution, separately from metadata listing. This is a gap in the inspected execution path, not proof that every deployment exposes reports anonymously.

ProductivityReportGeneratorTest prints a duration, has no assertions and comments out the generator call. Add representative expected-result fixtures before changing calculations. Review the current partial-day loop for omitted final-day work, overcounted first days and negative after-hours durations.

Delivery phases and validation

Phase 1 establishes baseline telemetry and correct export headers, validation, safe filters and authorization checks. Add CSV extraction that writes incrementally, managed artifact cleanup and explicit limits. Preserve existing response envelopes and valid filter behavior.

Phase 2 adds durable jobs and bounded workers. Pilot a single high-volume CSV report before extending all generators. Add job status and download UI behavior. Keep the synchronous route behind a configured threshold or capability rule.

Phase 3 replaces full productivity rebuilds with incremental chunks, preloaded references and restart-safe writes. Compare derived totals and per-day records against reviewed fixtures and representative existing results before switching consumers. Preserve raw workflow records.

Phase 4 consolidates dashboard queries and introduces scoped caching. Add reporting summaries only if measurements justify them. New schema changes require new migrations; preserve applied scripts and follow repository ownership and deployment conventions.

Validate narrow and broad date ranges, one and many provinces, empty results, high row counts, repeated requests and concurrent exports alongside normal API traffic. Record p50 and p95 latency, SQL elapsed time, logical reads, returned rows, peak heap, GC pause time, active and pending connections, queue depth, oldest job age, renderer concurrency, temporary storage and failure rate. Set numeric service targets from measured workload and business freshness needs rather than promising an unverified speedup.

Use feature flags per report for safe rollout. Compare CSV content and aggregate totals before promotion. Roll back routing to the previous generator if needed while retaining job history and published artifacts under the retention policy. Database changes should be additive so rollback does not require deleting source data.

Evidence and technical references

Repository locations below are relative to cis-api/src/main/java/com/dw/ngms/cis unless stated otherwise. Line references identify the reviewed source snapshot and may move as development continues.

  • F1: service/report/Examination/ExaminationProductionReport.java lines 47, 78 and 106; service/report/Archive/ArchiveProductionReport.java lines 78 and 159.
  • F2 and F3: service/report/ProductivityReportGenerator.java lines 55 through 149; persistence/repository/workflow/WorkflowRepository.java lines 561 through 585; service/HolidayCalendarService.java line 62; service/OfficeTimingsService.java; service/user/UserLeaveCalendarService.java.
  • F4: utilities/ReportUtils.java lines 78 through 130; persistence/domains/report/ReportFiltersData.java; web/vm/report/ReportingVm.java.
  • F5: service/DashBoardService.java lines 53, 77, 178, 401 and 531; persistence/repository/dashboard repositories.
  • F6: service/report/UserSummaryReport.java lines 92 through 118; cis-api/src/main/resources/application.yaml lines 65 through 107.
  • F7: web/rest/ReportingResource.java lines 61 through 74; utilities/ReportUtils.java lines 38 through 47; service/report/ReportingService.java lines 87 through 90.
  • Correctness: service/report/maintenance/MaintenanceProductionReport.java lines 95 through 119; service/report/ReportingService.java lines 81 through 85; cis-api/src/test/java/com/dw/ngms/cis/service/report/ProductivityReportGeneratorTest.java.
  • Spring JDBC statement settings and row callbacks: https://docs.spring.io/spring-framework/docs/6.0.x/javadoc-api/org/springframework/jdbc/core/JdbcTemplate.html
  • Oracle executed cursor plans and required privileges: https://docs.oracle.com/en/database/oracle/oracle-database/19/arpls/DBMS_XPLAN.html

Capture representative executed cursor plans with DBMS_XPLAN.DISPLAY_CURSOR and ALLSTATS LAST after collecting execution statistics. Inspect deployed view definitions, indexes and cardinality estimates before choosing physical database changes. The framework reference documents established JDBC capabilities; implementation should use the repository's current dependency versions.