דילוג לתוכן
מרכז התיעוד
לאתר LinkNodes Energy
LinkNodesENERGYתיעוד

טוענים את החיפוש…

    חיפוש גם בתוך המאמרים ↑ ↓ לניווט Esc לסגירה

    לאתר החברה

    שחרורים ואימות

    טיוטה היסטורית: מצב המעבר לאירופה

    רשומת תשתית מ־20 בנובמבר 2026. מעבר מסד הייצור נותר חסום; אינה הוראת מעבר או מצב שירות נוכחי.

    בעמוד הזה

    Status: source PRs #58–#62 merged; the tested Product map UI is live in Frankfurt. Production Supabase cutover remains blocked. Railway backup/restore is verified, but the services and volume remain in SFO after the Europe deployment approval was cancelled.

    This record supersedes status statements in the 2026-11-11 migration record. Historical evidence remains available there. Point-in-time parity must be checked again before a later cutover because Singapore remains writable.

    Scope and current authority#

    ComponentVerified stateRemaining release gate
    Merged application sourcePRs #58–#62 are merged; main is 92342e79a614c1d50889de5118e322a3a998b946The final merged-main workflows were blocked before execution by the actual Actions budget cap; the individual PR heads passed their full pipelines
    Vercel ProductUI-only deployment [operational-id-redacted], exact source da7f274ad14f42e44901fd8ad888f9e070187182, is READY in fra1 and resolves from the production domain; manual promotion guard verifiedNew billing and database cutover remain separate releases; the live cloud-browser map background limitation is recorded below
    Vercel Website and AdminPreviously verified production deployments READY in Frankfurt fra1No Website/Admin promotion or production environment switch is recorded in this update
    Supabase Singapore [project-ref-redacted]Live authoritative source in ap-southeast-1Retain until a verified target cutover and post-cutover reconciliation
    Supabase Frankfurt [project-ref-redacted]Healthy in eu-central-1; data synchronized; billing safety migration installedTarget server secret, real signed Clerk QA, Supabase backup/restore evidence and coordinated final cutover
    Railway Redis, worker and runtimeAll three healthy in SFO, one replica each; worker automatically redeployed after a source merge, with no Europe migrationExact six-field Amsterdam patch remains STAGED; deployment approval did not complete
    Website company formsPR #60 is merged; current live proxy still targets SingaporeTarget acceptance test, separate Website environment change and deployment
    HYPCredential-free fixture tests and database safety checks pass; external operations disabledHYP-issued test terminal and provider verification
    Native mobilePR #61 is merged; hosted native checks, source tests, Doctor and both platform exports passExpo authentication was declined; cloud configuration/signing verification and a signed device build remain incomplete

    The Product UI-only promotion is the customer-facing release recorded here. No Supabase source deletion or production database cutover, live payment/refund, or paid plan upgrade occurred. The agents did not increase the GitHub Actions budget.

    Frankfurt changes actually completed#

    The target-only operational migration geo_production_delta_sync_20261120 synchronized the authoritative Singapore delta:

    • Three audit-event inserts and two tenant-provider inserts.
    • One provider-preference revision/timestamp update and one verified-email timestamp update.
    • The operation used table locks and preimage assertions. Only the two specifically identified audit triggers were suspended and restored inside the transaction to avoid duplicate audit events; foreign keys and other triggers remained enforced.
    • All 43 application-table content fingerprints matched afterward. Before billing safety was applied, all 1,070 compared schema objects matched as well.

    The exact migration from billing PR #59, 20261120160000_billing_safety.sql, was then installed on the dormant Frankfurt target only. Its Git blob is e3d2759bc82eded8d13e3d2e98980f109bcc8fcb.

    Verification confirmed all ten reviewed function bodies, empty function search paths, denied execution for PUBLIC/anon/authenticated, the nine intended service-role grants, and denial of the legacy refund entry point even to service_role. Reconciliation returned zero items. All nine financial tables remained empty. All 43 data fingerprints still matched Singapore.

    The target schema now has 1,074 compared objects, including four new functions from billing safety; Singapore retains 1,070. This is an intentional reviewed schema difference, not unexplained drift. Source production was not changed. Both target cron jobs remain disabled; the two source cron jobs remain authoritative.

    The private Storage bucket configuration matches and contained no objects at inspection. The two Vault secrets match by hash; values were never added to repository content.

    Company form path#

    Both projects have ACTIVE lne-company-forms functions with the same bundle SHA-256:

    aa0b02799cc0505dff35adf783706921cc016f2cbd0040cd60beed3981fd81ac

    The existing verify_jwt=false setting is preserved. The function validates origin and Turnstile, uses Supabase-provided server environment values and retrieves the Turnstile value through lne_company_forms_secret. The Frankfurt RPC reads lne_company_forms_turnstile from Vault with an empty search path; the source/target secret hashes match.

    Merged PR #60 adds a server-only COMPANY_FORMS_SUPABASE_URL setting with strict origin validation and no fallback after target failure. Its earlier exact-commit Website preview was verified READY in fra1:

    No real form was submitted, no Website production variable changed and no live Turnstile acceptance is claimed.

    Supabase stop condition#

    The connected management API does not expose the target server-only key. The GitHub authentication method was submitted in the Supabase sign-in flow, which then required MFA. The MFA handoff was declined, so authenticated dashboard access was not completed. The Frankfurt server secret and Clerk trust settings were not verified through that flow. No further authentication attempt followed the decline.

    Provision the Frankfurt server secret directly into a protected, trusted QA deployment. Verify real Clerk-signed positive and negative tenant reads/writes and a Supabase restore before production. Then repeat final delta/parity checks under a controlled write window, deploy the compatible billing service, switch all consumers and hand over billing/retention cron exactly once. Keep Singapore available and reconcile any post-cutover writes before rollback. Source deletion remains a separate operation.

    Railway backup, restore and blocked deployment#

    Project [operational-uuid-redacted], production environment [operational-uuid-redacted].

    ServiceIDStill-live deployment
    Redis[operational-uuid-redacted][operational-uuid-redacted]
    BullMQ worker[operational-uuid-redacted][operational-uuid-redacted]
    BullMQ runtime[operational-uuid-redacted][operational-uuid-redacted]

    Redis volume [operational-uuid-redacted] remains the original 500 MB volume mounted at /data. No detach, deletion, resize or volume replacement was performed.

    The final readback at 16:58:11 UTC found all three services healthy in SFO and the physical volume still in SFO, with zero reported failures, warnings or issues. Redis and runtime retain the deployment IDs above. The worker was automatically redeployed from merge commit 93a9a088bfa93b3ee35ad8f12ab5bc679a5d8c6e after PR #61 merged: deployment created at 16:38:19 UTC, SUCCESS at 16:41:29 UTC. This automatic source deployment did not apply the staged Europe region patch. Readback inspection made no cloud mutation.

    The staged patch [operational-uuid-redacted] contains exactly six reviewed fields: remove SFO and add one europe-west4-drams3a replica for each service. The provider reports destructive=false; there are no variable, command, source, networking or replica-count changes. The intended Railway destination is EU West, Amsterdam, while Supabase/Vercel are Frankfurt.

    Backup and restore evidence#

    • Backup: LinkNodesEnergy-Redis-pre-Europe-20261120.rdb, 4,307 bytes, saved separately from source control.
    • SHA-256: 44e65eee4626d34fe812dbd5528bc57978d5950419d248694a2137ea191a731b.
    • Redis 8.2.11 validated the RDB format/checksum.
    • An isolated restore used a copied RDB in a private temporary directory within the existing provider container, a separate Unix socket, TCP disabled, persistence disabled, and no job consumers.
    • All 12 persistent keys matched the canonical live-data hash 2fce0166184ea25fd87e87c5d33e17e3c521eaf7. The thirteenth live key was an expiring BullMQ heartbeat.
    • The temporary process and files were removed; the live service still returned PONG.
    • Fresh read at 15:47:55 UTC / 17:47:55 Israel: completed=4, waiting=0, active=0, delayed=0, failed=0, prioritized=0, waiting-children=0, paused jobs=0, queue_paused=0.

    Attempt outcome and next validation#

    An earlier connector call was interrupted without a provider result. Official accept_deploy approval attempts, including the latest attempt, were cancelled; the provider returned: “Cancelled — the user did not approve this action. No changes were made.” No successful Europe deployment workflow was returned. The final readback confirms the same six-field patch remains STAGED and the actual services and volume remain in SFO. The automatic worker source deployment described above is separate from the cancelled region change. No dashboard workaround was used.

    After the required connector deployment approval is completed, verify:

    1. All three deployments settle SUCCESS/ONLINE with their actual replicas in Europe; the physical volume region also shows Europe.
    2. Source/image/command equivalence, the same volume ID/mount/size, private DNS and Redis PONG.
    3. New deployment-scoped boot smoke jobs finish, all pending/failed/stalled counts are zero, and both worker health endpoints pass.
    4. Redis persistence and normal retention behavior. Old completed jobs may be pruned by the existing 24-hour retention policy after new boot jobs; do not mistake expected pruning for data loss.

    Stop on a failed migration, unexpected plan/cost change, volume replacement or health regression. Preserve the prior deployments and verified RDB. Application deployment rollback does not itself restore Redis data.

    Application and release evidence#

    Merged source and exact-head hosted CI#

    All five task PRs are merged. The final combined source is 92342e79a614c1d50889de5118e322a3a998b946. Each PR passed its complete Energy workflow on the exact head below before merge, including check, database, browser and Website jobs. Native PR #61 also passed its multiplatform workflow.

    Merged PRReviewed changeExact tested headSuccessful hosted pipeline
    #58Current-location map default and the user's existing vehicle image; approved surrounding UI/assets preservedda7f274ad14f42e44901fd8ad888f9e070187182Energy checks 44773927140
    #59Strict billing outcomes, atomic refund reservations, idempotency and conservative lease recovery; supersedes closed drafts #53, #55 and #562328db174499f74df3fcb4474af232571b2a526fEnergy checks 44771595761
    #60Validated server-only company-form destination configuration8844ea266d72d0dee0e6d015c932f752cbbcfbbfEnergy checks 44771617895
    #61Native location-first map, account-safe vehicle marker, 24 behavioral tests and the 13 compatible Expo SDK 58 package alignmentsbed59e0db916481eac3e8873443396b556574e7bEnergy checks 44771604725; multiplatform checks 44771604812
    #62Deterministic system exports and inventory assertions corrected to the existing migration baseline857f8470d91badfacd02ae4caf964b756079d701Energy checks 44771514370

    Provider bounds PR #54 remains a separate open draft for transaction date-window validation; it is not one of the five merged PRs. No live provider was activated.

    Combined local validation: 633 tests across 59 files, root lint and production build passed. Billing additionally passed 59 adapter/service fixtures and 11 database safety checks including true concurrent races; all 18 source migrations and the existing database regression suites passed on isolated PostgreSQL 17. Native TypeScript, 24/24 tests, Expo Doctor 20/20, iOS export and Android export passed.

    GPS browser acceptance and retained screenshots#

    Local Chromium socket creation was denied and sandbox escalation was rejected, so local browser execution is not claimed. Hosted GitHub Actions subsequently ran the production-demo browser suite successfully on exact head da7f274ad14f42e44901fd8ad888f9e070187182.

    The completed browser job 136418569161 reports 79 passed, 1 skipped, 80 total. The sole skip is the existing providers.spec.ts:106 live directory/station endpoint case. All 14 geolocation.spec.ts cases passed, with no GPS skips: automatic start without a manual fallback; map-data failure; explicit city selection versus a late fix; denied, timed-out and old fixes; coarse/narrow dark-mode layout; page visibility and permission revocation; freshness expiry; Smart marker/theme behavior; silent-service timeout and watch release; station permalink discovery; temporary signal recovery; and native Chromium geolocation updates.

    The map acceptance artifact 14650350394 contains coarse-mobile.png and map-vehicle-dark.png. The 25,534-byte ZIP was downloaded and verified against SHA-256 d6070846d5480f86c7addbeb4b4b3cd26ba43055a9e82feb1612be442bde3e6c. Visual review found the approved Smart image loaded and centered, and the narrow location card contained its approximate-location text and controls without horizontal overflow. The map background is deliberately synthetic in the fixture; these images validate marker/layout/theme behavior, not live external tile availability or physical-device GPS.

    Product UI-only production release#

    The official Vercel promotion request for [operational-id-redacted] returned without error. Subsequent deployment lookup using product.linknodesenergy.com resolves that deployment, exact source da7f274ad14f42e44901fd8ad888f9e070187182, with READY / production / fra1 state.

    • Live health: HTTP 200, status=ok, surface=product, mode=demo.
    • Live map catalog: HTTP 200; 1,773 stations, 12 operators and 1,316 localities.
    • The previous production deployment is retained as the rollback target.
    • The deployed source is the tested UI-only PR #58 head. The combined main source, including new billing behavior, was not promoted by this release. No billing activation flag, environment variable, database destination or cron ownership was changed.

    At 17:01:22 UTC, a fresh Vercel dashboard reload verified that the Auto-assign Custom Production Domains control remained unchecked, the manual-promotion note remained visible and Save was disabled. No setting was edited during this readback. The manual release guard remains in place; no deployment protection was weakened.

    Live public-browser review loaded the Product sign-in surface and existing Personal demo dashboard. The demo map route /fleet-demo/[operational-uuid-redacted]/map rendered the updated location panel and preserved layout. Location permission was denied in this cloud browser, so no live GPS position is claimed. Its map background displayed the existing unavailable-background fallback, including after one safe retry; the cause remains unverified. The browser captured no failed requests that explained it. This demo had no selected providers, so its zero-result list was expected. Live external map-tile success is not claimed from this review.

    Source comparison found the map initializer, worker URL, style selection and error catch match baseline c0982ded08192dc9ea6b379efe7ab9bb89d073cd. A read-only comparison against the retained baseline deployment was blocked by its protected deployment URL at catalog fetch, before map initialization, so it could not establish whether the same background issue occurs there. No source change or protection change followed. A post-promotion Vercel query for error/fatal log groups over a ten-minute window returned no groups; that observation is limited to the queried window.

    Final merged-main CI and external gates#

    GitHub Actions ran the exact-head PR pipelines above successfully. The later final-main Energy run 44776138027 and Postman run 44776137916, both on 92342e79a614c1d50889de5118e322a3a998b946, were blocked before steps executed. Live billing inspection showed included usage $12/$12 and the repository Actions budget $3/$3, Usage blocked. This is the observed current cap, superseding the earlier $0-budget snapshot. The agents did not increase a budget, rerun these blocked runs, use alternate runners or weaken any gate to bypass it. Successful PR results do not constitute a completed final merged-main run.

    Expo authentication was declined. GitHub repository and environment Actions-secret lists were empty at inspection; no Expo signing or Maps credentials were provisioned through this work. Expo cloud configuration/signing and the Android Maps key remain unverified. No signed IPA/APK/AAB or store submission exists.

    Keep HYP payment/refund activation flags false until HYP-issued test credentials and provider-side checkout, token, callback, ambiguity, refund and reconciliation checks are available. The current test environment is synthetic and credential-free, not a connected HYP sandbox.