Promotes the ASCII-safe address transliteration fix (go-fulfillr#24) to prod
after dev verification.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Deploys the ASCII-safe address transliteration fix (go-fulfillr#24) to the
staging back-office. Prod image tag unchanged.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Deploys the go-store banner-columns build to the prod back-office so the Coupons
admin round-trips the global-sale banner fields on prod. Applied via
`make deploy TAGS=fulfillr`.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Deploys the go-store banner-columns bump (coupon banner_label/banner_text) to
the staging back-office so the Coupons admin round-trips the global-sale fields.
Applied via `make deploy TAGS=fulfillr-dev`. Prod tag unchanged.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
v3.0.0 dropped pgvecto.rs support, which crash-looped immich-server
("No vector extension found") and left photos.debyl.io on a white page.
Swap db_image to Immich's bundled postgres image (VectorChord 0.4.3 +
pgvecto.rs 0.2.0) so v3 reindexes the existing embeddings into
VectorChord in place; bump server/ml from v2.7.5 to v3.0.0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Stripe relink endpoints (stripe-candidates / link-stripe) + ch_ refund support
now live on prod back-office.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
After a power cycle a transient HMAC Secrets Manager blip leaves
go-fulfillr's gated routes unregistered (404) with the process still up,
so nothing restarts it. Add a podman healthcheck probing the new
dependency-free /api/v1/health/startup (503 until those routes register)
with healthcheck_failure_action: restart, so podman restarts the
container in place and the next boot self-heals.
- fulfillr.yml + fulfillr-dev.yml: healthcheck via busybox wget (ships in
the alpine image), interval 30s / timeout 5s / retries 3 /
start_period 30s (covers the ~14s HMAC retry backoff), failure_action
restart. Existing restart_policy on-failure:3 kept (process-exit case).
- main.yml: bump fulfillr + fulfillr-dev image to 20260628.1930 (the
build carrying the /health/startup probe).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The esp32-stm32-vcu firmware build generates common-yaml headers with
`python3 generate.py`, which needs pyyaml + jinja2. The runner's base Python is
PEP 668 externally-managed (pip install fails) and the IDF venv isn't on PATH in
the docker-exec step shell, so install both as distro packages. Lets firmware
jobs run a plain `python3 generate.py` with no pip and no IDF sourcing.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds email_sent/email_failed timeline events for ticket label emails.
Deployed to fulfillr-dev and fulfillr (prod) on home.debyl.io.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Ticket refund + replacement-shipment endpoints and guarded transitions.
Deployed to fulfillr-dev and fulfillr (prod) on home.debyl.io.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- remove snipcart_api_key from dev/production config (Snipcart decommissioned
post-migration)
- add review-outreach and cart-recovery schedule_name/schedule_group blocks
(dev + prod) for the EventBridge-driven outreach and cart-recovery jobs
- bump fulfillr image 20260607.0217 -> 20260613.0117
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- Containerfile.ci: add python3-yaml + python3-jinja2 and the
gcc-arm-none-eabi / binutils / libnewlib toolchain for embedded builds
- bind-mount the runner's SSH key + known_hosts read-only into each job
container at /root/.ssh so submodule clones over
ssh://git@git.skudak.com:2222 succeed; staged as a dedicated
container_file_t-labelled ci-ssh copy (tasks/user.yml) and allowlisted
via valid_volumes (config.yaml.j2)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The Ollama role and SearXNG container backed FISTO AI responses in the
greg-time Discord bot. greg-time 3.9.6 drops both (plus the Gemini path)
in favor of a single xAI Grok backend, so:
- remove the ollama role and its wiring in deploy_home.yml
- remove the searxng container task, template, and searxng_path default
- gregtime: swap OLLAMA_*/SEARXNG_URL/GEMINI_API_KEY env for XAI_API_KEY,
bump image 3.6.5 -> 3.9.6
- vault: add xai_api_key, drop gemini_api_key
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The EasyPost tracker webhook moved to debyltech-api (publicly reachable Lambda);
the fulfillr host is LAN-restricted and no longer hosts it, so the carve-out is
no longer needed. Removes the handle blocks for prod and dev.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The Fulfillr host is IP-restricted, so EasyPost's servers can't reach it. Add a
narrow `handle /webhooks/easypost` before the IP restriction (handle blocks are
mutually exclusive, first match wins) for prod (:9054) and dev (:9055) so the
HMAC-verified tracker webhook is reachable while the rest of the host stays locked.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Lambda packaging steps in some workflows shell out to `zip`; the image
only had `unzip`. Add `zip` alongside it.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add a second go-fulfillr container (fulfillr-dev) wired to the staging
Turso store + EasyPost/Stripe test keys via dev.json, served at
fulfillr-dev.debyltech.com (Caddy -> :9055), LAN-restricted like prod.
- fulfillr-dev.yml + dev.json.j2: the staging container, volumes, config
- defaults: fulfillr_dev_* vars; prod store URL stubbed off until cutover
- Caddyfile + caddy.yml: fulfillr-dev site block and static mount
- awsddns.yml: Route53 DDNS for the fulfillr-dev hostname
- production.json.j2: add store_database_url/store_auth, rename stripe key
var to fulfillr_stripe_api_key
- vault.yml: dev + store/stripe secrets
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Switch the act_runners from :host execution to docker:// images backed by
a rootless podman socket under the gitea-runner user, so each job runs in
its own ephemeral container with per-job Go caches. This eliminates the
cross-repo GOMODCACHE/go-build poisoning that forced the debyl runner to
capacity:1.
- deps.yml: enable the rootless --user podman.socket, ensure subuid/subgid,
register gitea_runner_uid; drop the rootful system socket override,
podman-docker and host golang
- images.yml + Containerfile.ci/.espidf: build localhost/gitea-ci and
localhost/gitea-ci-espidf into the runner's rootless image store
- config.yaml.j2: docker:// labels (per-runner overridable), docker_host
-> rootless socket, force_pull false
- act_runner.service.j2: XDG_RUNTIME_DIR + DOCKER_HOST -> user socket
- defaults: uniform capacity:4 (drop the debyl capacity:1 workaround);
esp_idf_version now tags the espressif/idf-based image
- main.yml: import images.yml, drop the host esp-idf install (firmware jobs
use the espressif/idf job container instead)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Bump fulfillr image to the build with the tickets feature, and add the
tickets_table to the fulfillr production.json config (new debyltech-tickets-prod
DynamoDB table) so the /api/v1/tickets routes register.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- add stripe_api_key to fulfillr production.json template
- add restricted Stripe key to ansible vault (encrypted)
- bump fulfillr image to the CI build containing the Stripe endpoints
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Picks up the case-status simplification from go-fulfillr 309550d
(only "open" and "closed" are accepted on PATCH; "new" is rejected).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Bump fulfillr container image from 20260124.0411 to 20260509.1940
(built from go-fulfillr commit 48b9f60 which adds /api/v1/cases
endpoints for the contact-form CRM dashboard).
- Add fulfillr_cases_table default ("debyltech-cases-prod") so the
HasCasesConfig() guard flips on at startup and the cases routes
register.
- Add cases_table to production.json.j2 so it lands in /config inside
the container.
Verified after deploy: GET /api/v1/cases returns the existing test
cases, PATCH succeeds, GSI1PK rewrite works.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Bastian de Byl
bastian