The Docker socket group varies by host (e.g., GID 1001 on WSL2).
Hardcoding group_add: ["0"] doesn't work when the socket is owned
by a different group. The installer now detects the socket GID at
install time via stat. The main docker-compose.yml uses a
configurable DOCKER_GID env var (defaults to 0).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The cameleer-saas service needs Docker socket access for tenant
provisioning. Add the socket bind mount, group_add for permissions,
and explicit DOCKER_HOST=unix:///var/run/docker.sock to prevent
the Java Docker client from falling back to TCP (which happens on
WSL2 + Docker Desktop when DOCKER_HOST leaks from the host env).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Installer now asks deployment mode in simple mode:
- Multi-tenant vendor: creates saas-vendor role + assigns to admin
- Single tenant: asks for org name, creates Logto org + tenant record,
assigns admin as org owner
Reverts always-create-vendor-role — role is only created when vendor
mode is selected. TENANT_ORG_NAME env var passed to bootstrap for
single-tenant org creation.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ClickHouse: enable built-in Prometheus exporter at :9363/metrics via
config.d/prometheus.xml with metrics, events, and async_metrics.
Docker labels added for docker_sd_configs auto-discovery.
Tenant servers: add prometheus.scrape/path/port labels to provisioned
server containers pointing to /api/v1/prometheus:8081.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ClickHouse entrypoint needs write access to resolve from_env attribute
and apply CLICKHOUSE_PASSWORD to the default user config.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ClickHouse default user had no password, causing auth failures on recent
CH versions. Set password via from_env in clickhouse-users.xml, pass
credentials in JDBC URLs to SaaS services and tenant server containers.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Move all SaaS configuration properties under the cameleer.saas.*
namespace with all-lowercase dot-separated names and mechanical env var
mapping. Aligns with the server (cameleer.server.*) and agent
(cameleer.agent.*) conventions.
Changes:
- Move cameleer.identity.* → cameleer.saas.identity.*
- Move cameleer.provisioning.* → cameleer.saas.provisioning.*
- Move cameleer.certs.* → cameleer.saas.certs.*
- Rename kebab-case properties to concatenated lowercase
- Update all env vars to CAMELEER_SAAS_* mechanical mapping
- Update DockerTenantProvisioner to pass CAMELEER_SERVER_* env vars
to provisioned server containers (matching server's new convention)
- Spring JWT config now derives from SaaS properties via cross-reference
- Clean up orphaned properties in application-local.yml
- Update docker-compose.yml, docker-compose.dev.yml, .env.example
- Update CLAUDE.md, HOWTO.md, architecture.md, user-manual.md
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Replace inline FileField and native <input type="file"> with
FileInput from @cameleer/design-system (drag-and-drop, icons, clear)
- Update CertificatesPage and SsoPage to use FileInput + FormField
- Fix /certs volume permissions (chmod 775) so cameleer user can write
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Removed cameleer3-server and cameleer3-server-ui from docker-compose
(tenants provision their own server instances via the vendor console)
- Removed viewer/camel user from bootstrap (tenant users created during
provisioning)
- Removed Phase 7 server OIDC configuration (provisioned servers get
OIDC config from env vars, claim mappings via Logto Custom JWT)
- Removed server-related env vars from bootstrap (SERVER_ENDPOINT, etc.)
- Removed jardata volume from dev overlay
Clean slate: docker compose up gives you Traefik + PostgreSQL +
ClickHouse + Logto + SaaS platform + vendor seed. Everything else
(servers, tenants, users) created through the vendor console.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Deployed app containers are put on the cameleer-traefik network by the
orchestrator, but the server and Traefik were only on the compose-internal
network. This caused UnresolvedAddressException when apps tried to connect
to cameleer3-server:8081 for agent registration and SSE.
- Add cameleer-traefik network with fixed name (no compose project prefix)
- Attach server to cameleer-traefik with DNS alias "cameleer3-server"
- Attach Traefik to cameleer-traefik for routing to deployed apps
- Add dev overrides for Docker orchestration (socket, volumes, env vars)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Remove docker-java-core and docker-java-transport-zerodep from pom.xml
- Remove Docker socket mount, group_add, jardata volume from docker-compose.yml
- Remove CAMELEER_DOCKER_NETWORK and CLICKHOUSE_URL env vars from SaaS service
- Remove jardata volume definition
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
1. Docker socket security: remove root group from Dockerfile, use
group_add in docker-compose.yml for runtime-only socket access
2. M2M server communication: create ServerApiClient using Logto
client_credentials grant with API resource scope. Add M2M server
role in bootstrap. Replace hacky admin/admin login in
AgentStatusService.
3. Async deployment: extract DeploymentExecutor as separate @Service
so Spring's @Async proxy works (self-invocation bypasses proxy).
Deploy now returns immediately, health check runs in background.
4. Bootstrap: M2M server role (cameleer-m2m-server) with server:admin
scope, idempotent creation outside the M2M app creation block.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Docker Compose prefixes network names with the project name, so the
actual network is cameleer-saas_cameleer, not just cameleer. Pass
CAMELEER_DOCKER_NETWORK env var using COMPOSE_PROJECT_NAME.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add CI step to build cameleer-runtime-base image by downloading the
agent shaded JAR from Gitea Maven registry and pushing the image.
Wire CAMELEER_AUTH_TOKEN from docker-compose into RuntimeConfig so
deployed containers authenticate with cameleer3-server. Add agent.jar
to gitignore for local builds.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Traefik couldn't auto-link the logto router when two services
(logto, logto-console) exist on the same container. This broke
ALL default tenant routing (sign-in, OIDC, API).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The admin console (port 3002) calls the Management API on the
default tenant (port 443). Add Traefik CORS headers to allow
cross-origin requests from the admin console origin.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ADMIN_ENDPOINT set to HTTPS so OIDC issuer matches browser URL.
NODE_TLS_REJECT_UNAUTHORIZED=0 lets Logto's internal ky-based
OIDC self-discovery accept the self-signed cert through Traefik.
Remove in production with real certs.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Admin console HTTPS via Traefik conflicts with Logto's
ADMIN_ENDPOINT self-discovery. Parking this for now.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ADMIN_ENDPOINT=http://localhost:3002 for Logto self-calls.
TRUST_PROXY_HEADER makes Logto use X-Forwarded-Proto from Traefik
to generate HTTPS URLs for browser-facing OIDC flows.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add PUBLIC_HOST as network alias on the logto container so its
internal ADMIN_ENDPOINT calls (http://PUBLIC_HOST:3002) resolve
inside Docker directly, bypassing Traefik. Browser traffic goes
through Traefik on host port 3002 with TLS termination.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Remove tls=true from the logto-console router so the entrypoint
accepts plain HTTP. Logto's internal self-calls via ADMIN_ENDPOINT
use HTTP and pass through Traefik transparently. Browsers can
access via HTTP on port 3002.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Use separate port 3443 for TLS-terminated admin console access.
Port 3002 stays directly mapped from logto in dev for Logto's
internal OIDC self-discovery via ADMIN_ENDPOINT.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Traefik-only change: new entrypoint + router for TLS termination.
No changes to Logto ADMIN_ENDPOINT or bootstrap script.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Revert all Traefik port 3002 and ADMIN_ENDPOINT changes that broke
bootstrap. Admin console HTTPS access needs a different approach.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Logto calls ADMIN_ENDPOINT internally for OIDC discovery. Using
PUBLIC_HOST resolved to the host machine where Traefik now owns
port 3002, causing a routing loop. localhost resolves inside the
container directly.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add admin-console entrypoint to Traefik with TLS termination.
Route port 3002 through Traefik to logto:3002. Update Logto
ADMIN_ENDPOINT to use HTTPS.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add "Build and push Logto image" step to docker job
- Remove build: directive from logto service in docker-compose
- docker-compose now only pulls pre-built images, no local builds
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add docker/logto.Dockerfile: builds custom Logto image with sign-in
UI baked into /etc/logto/packages/experience/dist/
- Remove sign-in-ui init container, signinui volume, CUSTOM_UI_PATH
(CUSTOM_UI_PATH is Logto Cloud only, not available in OSS)
- Remove sign-in build stage from SaaS Dockerfile (now in logto.Dockerfile)
- Remove docker/saas-entrypoint.sh (no longer needed)
- LoginPage auto-redirects to Logto OIDC on mount instead of showing
"Sign in with Logto" button — seamless sign-in experience
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
CUSTOM_UI_PATH is a Logto Cloud feature, not available in OSS.
The correct approach for self-hosted Logto is to volume-mount
over /etc/logto/packages/experience/dist/.
- Use init container (sign-in-ui) to copy dist to shared volume
as root (fixes permission denied with cameleer user)
- Logto mounts signinui volume at experience/dist path
- Logto depends on sign-in-ui init container completion
- Remove saas-entrypoint.sh approach (no longer needed)
- Revert Dockerfile entrypoint to direct java -jar
- Permit /favicon.svg in SecurityConfig for sign-in page logo
Tested: full OIDC flow works end-to-end via Playwright.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replace Logto's default sign-in page with a custom React SPA that
matches the cameleer3-server login page using @cameleer/design-system.
- New Vite+React app at ui/sign-in/ with Experience API integration
- 4-step auth flow: init → verify password → identify → submit
- Design-system components: Card, Input, Button, FormField, Alert
- Same witty random subtitles as cameleer3-server LoginPage
- Dockerfile: add sign-in-frontend build stage, copy dist to image
- docker-compose: CUSTOM_UI_PATH on Logto, shared signinui volume
- SaaS entrypoint copies sign-in dist to shared volume on startup
- Add .gitattributes for LF line endings
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Browser sends Origin header on fetch calls even same-origin. Server
needs the public host in its CORS allowlist. Derived from PUBLIC_HOST.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Self-signed certs cause PKIX errors when the server fetches OIDC
discovery. CAMELEER_OIDC_TLS_SKIP_VERIFY=true disables cert
verification for OIDC calls only (server-team feature, pending build).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Nginx needs to see /assets/... not /server/assets/... to find the files.
Strip-prefix + BASE_PATH=/server now works correctly with the fixed image.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Server team fixed the BASE_PATH sed ordering bug. Remove our entrypoint
override and let the image's own entrypoint handle it.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The server-ui's entrypoint inserts <base href> THEN rewrites all
href="/" — including the just-inserted base tag, causing doubling.
Patched entrypoint rewrites asset paths first, then inserts <base>.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Server-ui needs BASE_PATH for React Router basename. Without strip-prefix,
no X-Forwarded-Prefix doubling. Server-ui handles full /server/ path itself.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Server-ui injects BASE_PATH=/server/ into <base href>. Without strip-prefix,
Traefik forwards /server/ path AND server-ui adds /server/ again = double prefix.
Strip /server before forwarding so server-ui sees / and produces correct <base href="/server/">.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Server-side path rewrite breaks React Router (browser URL stays at /
but basename is /platform). The SPA entry point is /platform/ — users
bookmark that. Root / goes to Logto catch-all which is correct for
direct OIDC flows.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Users hitting the root URL now get redirected to the SaaS app instead
of seeing Logto's unknown-session page.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
