Создан бакет variant-cache в shared MinIO (minio.kzntsv.site/books-vds) для sharp resize+watermark вариантов
(S3-эквивалент легаси App_Data/imageCache; без него 500 NoSuchBucket на всех sharp-картинках).
Грант: snolla S3-креды == MinIO root (сверено) → mode-server-fs даёт полный доступ by construction, IAM не нужен.
Write-verify под деплойными кредами: put→stat→get(md5 match 78fab481)→delete→gone. Другие бакеты не тронуты.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Оператор сменил watermark-подход на in-process sharp → снял глиф со SHARED imgproxy стека 29 (books-vds).
Вырезал IMGPROXY_WATERMARK_DATA (byte-identical pre-watermark бэкапу, cross-check true) → PUT+redeploy+purge.
Verify: pilorama98 (главный консюмер) 3 рендера == baseline (43950/172042/160010B); stostayer НЕ imgproxy-консюмер (не задет); tandem featured→чистый (ожидаемо). Глиф-бинарь 7127e92a оставлен для sharp.
tandemmebel-deploy теперь ждёт пересбора образа на sharp-пин.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Ops-подхват по тиражу tandemmebel. Воркшоп попросил выставить брендовый
глиф-вотермарк на imgproxy стек 29 (books-vds, отдаёт картинки всем тенантам).
Первый base64 (в inbox И в закоммиченном таске) был БИТЫЙ на источнике —
tasks_create порезал 10КБ-поле: sha f8f0…≠e995…, IEND нет, зацикленный хвост.
Поймал sha256-сверкой ДО прода → не запушил битьё (imgproxy бы упал на парсе
WATERMARK_DATA = отвал картинок у всех тенантов).
Воркшоп до-доставил logo.png бинарём в git (7127e92a, sha256 e995…971fc, 8015B).
PUT стека 29 (Portainer portainer.kzntsv.site ep1, X-API-Key, IMGPROXY_WATERMARK_DATA
inline env, PullImage:false, БЕЗ глобального _OPACITY).
Smoke cache-busted: A tandem вотермарк виден глазами (27294→27820B);
B pilorama контроль byte-identical 160010B — глобальная opacity не просочилась.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
DNS zone is on Yandex DNS (dns1/dns2.yandex.net), not reg.ru. Flip propagated
unevenly (dns2 first, dns1 lagged); cutover gate = BOTH authoritative Yandex NS
agree apex+www -> 89.253.255.94 (monitor waited for it) so LE HTTP-01 can't hit
RUVDS via a stale auth server and burn the rate-limit.
Stack 17 traefik-rule -> Host(labtools.ru)||Host(www.labtools.ru) via Portainer
PUT (env 8/8 preserved, pullImage=false). LE cert issued (~t+20s; brief self-signed
window during issuance). Live-smoke GREEN: both hosts 200 (nav+catalog+products),
redirects parity (trailing/lowercase/.php), sitemap index host=labtools.ru (this
tenant's siteUrl is non-www), theme assets byte-identical (RUVDS 301 apex->www made
the direct md5 an artifact; -L => MD5-OK). Staging host removed (->404). External:
apex->VDS 200; www still split-brain on RUVDS cache (200, identical content, TTL 6h).
RUVDS left intact as rollback.
Task closed. Whole batch done: labtools.ru + labtools.pro LIVE on VDS.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
DNS flip confirmed at authoritative ns1.reg.ru (Resolve-DnsName) + 8.8.8.8 ->
89.253.255.94 (labtools.pro/www; NOT labtools.ru). Earlier checks hit stale TTL
cache; authoritative NS was the truth.
Stack 19 traefik-rule -> Host(labtools.pro)||Host(www.labtools.pro) via Portainer
PUT (env 8/8 preserved, pullImage=false). LE cert issued instantly (ssl_verify=0).
Live-smoke GREEN: both hosts 200 (nav+catalog+products), redirects parity, sitemap
index host=www, real content (lang=en). Staging host removed (->404). External
check from RUVDS box: labtools.pro/www -> 89.253.255.94 HTTP 200 trusted cert.
RUVDS labtools.pro left intact as rollback.
Task closed.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
ADR-0010: forms moved to @snollajs/forms-api, @snollajs/snolla dropped
entirely, core@0.1.1 (Buffer over btoa). Stack 16 → pilonuxt:cf2bba2.
Pre-deploy tree-check of .output/server/node_modules 4/4: 0x snolla,
0x btoa, exactly 1x data@0.9.1, core 0.1.1 + content-api 0.14 + forms-api
0.1.0. Smoke green: SSR /+/catalog 200, robots/yandex from DB (ADR-0009
no regress), form ?path=/checkout empty -> 422 JSON, /nope -> 404 (no
valid submitted). Smoke gotcha: 422 only with Accept: application/json.
Closes [deploy-pilonuxt-forms-api-drop-snolla].
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Stack 16 redeployed to registry.kzntsv.site/pilonuxt:40bb383 (ADR-0009:
DB-backed staticPages + robots.txt). Build from monorepo with forced client
regen (rm src/generated + schema-gen 0.6 → getStaticPage/getRobotsTxt);
home push hit traefik-499 on .output layer (425MB) → save|ssh load + push
from VDS. Smoke acceptance 6/6 verified by body (robots full from DB w/
Yandex Clean-param + Sitemap, yandex/google verifications, /nope→404).
Closes [deploy-pilonuxt-static-pages-robots].
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
4th layer fixed by peer (data 0.9.0->0.9.1). Rebuilt from 6c6d52d
(workspace Dockerfile + sharp @img copy), pushed from VDS, deployed stack 16.
Container smoke all 200: product title<=60 + meta 'доставка по СПб и ЛО';
/catalog/planken 'Планкен прямой и скошенный — купить в СПб'; products 200x3;
home/shipping/blog meta ok. stderr clean. Compose -> 6c6d52d.
Pending (cross-repo, needs push-ok): commit sharp Dockerfile + docker-deploy.md to monorepo.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Dockerfile runner now copies @img (incl sharp-libvips-linux-x64) into .output;
verified SHARP_OK in container. Redeploy: sharp gone, but products 500 from
getSiteById findByPk undefined on snolla 0.17 ecommerce route (pages route OK).
Rolled back to 19a4a84 (all 200). Remaining blocker = snolla/peer zone.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Redeploy from eaee56f: snolla-500 gone, home/categories 200 with SEO,
but sharp native libvips missing in Nitro .output → product listing 500.
Rolled back to 19a4a84 again (catalog works). Platform-specific, peer's
local prod-bundle couldn't catch it.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Build+deploy mechanics solved (workspace Dockerfile, VDS-push for 499).
Prod 500: @snollajs/snolla@0.11.0 getSiteById (Sites model undefined),
nested under content-api@0.11.0 alongside direct snolla@0.16.0.
Rolled back to 19a4a84 (verified 200). Dep fix = web/snolla owner.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
После deploy нового кода (образ master-bf2a8c5, registryGc v2-DELETE rewrite)
прогнан registryGc {dryRun:true} через scheduler-dispatch (POST /tasks/registryGc
+ x-agenda-job-id): success, 401 нет нигде, repos books-* видны (7 processed),
отчёт сгенерён. Acceptance auth/dryRun met.
Finding отдан books (НЕ блокер кред-таски): GC бежит но drop=0 всегда
(keep=tags-1), keepLastN не применяется — getCreated→null для всех манифестов
(configDigest не извлекается, вероятно OCI image-index от buildx),
planDeletions защищает null-dated группы. Bug в books lib/registryV2.js.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Ingest .wiki/concepts/stostayer-web-deploy-runbook.md — канал деплоя легаси
web (build offline → push docker.stostayer.ru → Portainer-стек 16 через
container-IP API с хоста, мимо Angie BA) + rollback + гоча VPN-IP бан
хостером при retry-push-шторме.
БЛОКЕР задокументирован: легаси packages/web (node16/CJS/Nuxt2) не
пересобрать ни с какого свежего дерева — 5 ESM-ставших депов (@snollajs/snolla,
@snollajs/content-api, @stostayer/api, @stostayer/data ×2). 0.3.18 заморожен.
Фикс формы (120bc07) едет с web4-cutover.
Таска stostayer-web-complaint-form-deploy → 🔵 park (blocker=web4-cutover);
попытка деплоя 0.3.19 откатана на 0.3.18, сайт восстановлен. Дохлый 0.3.19
оставлен в registry по решению vitya.
+ index.md, log.md.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
