Step 1: a static site
New line, in development. Installed on a four-node test cluster built from shc next
b43f0c168(ce5db649eb) on 2026-10-08: three nodes in locationhome, one infar. The output below is that run’s, trimmed.
The smallest app: one nginx container serving one page. Two files and the page.
nginx/ app.yaml compose.yaml site/index.htmlname: nginxversion: "1.27"title: nginxdescription: A static site served by nginxcategory: developmentlicense: id: BSD-2-Clauseservices: nginx: image: nginx:1.27-alpine expose: - "80" # the port nginx listens on; nothing is published yet volumes: - ./site:/usr/share/nginx/html:ro healthcheck: test: ["CMD", "wget", "-qO-", "http://127.0.0.1/"] interval: 10s retries: 3 deploy: resources: limits: cpus: "0.25" # this is the service's quota memory: 64M<!doctype html><title>hello</title><p>Hello from SHC.</p>Install it
Section titled “Install it”On a node, from the directory’s absolute path, as deployment web in tenant rig, publishing outside port 8080 to the container’s port 80:
$ shc install /root/tutorial/nginx -s web -t rig --expose 8080=nginx:80installing stack "web"web: startingweb: runningweb: nginx runninginstalled deployment "web"exposure: lan$ curl -s http://<node address>:8080/<!-- nginx/site/index.html --><!doctype html><title>hello</title><p>Hello from SHC.</p>$ shc ls -t rigSTACK TENANT ENV APP STATUS NODEweb rig prod /root/tutorial/nginx running far01A relative path works too: the CLI resolves ./nginx (and ~) in your shell’s directory before it sends anything, and X404064SRCNFD means only that the directory isn’t on this machine (SHC-1407, ce modules/app/commands/source_body.go). The directory exists on that node only, so that node packs it into the cluster’s app archive and hands the install to the render leader, which renders the archive like any other app; placement puts the replica on any node the rules allow, and that node reads the files it binds from the same archive (SHC-1722, ce modules/app/app_archive.go). Step 3 installs from a repository instead. --expose 8080=nginx:80 publishes the container’s port 80 on outside port 8080. exposure: lan is about an HTTP hostname through Traefik, which admits only LAN addresses (ce modules/app/commands/install.go). On a cluster with service VIPs, SHC listens on the nodes selected by publish.listen, or every node when unset. It prefers a ready replica on the accepting node; +spread allows any ready replica. This publish has no connection, so SHC renders no ports: entry and the node’s 127.0.0.1:8080 does not answer (ce modules/app/port_mappings.go, core/publish, modules/vip/compile). Use a listening node’s reachable address for the curl above; in this run each of the four nodes’ addresses answered with the page, though the replica ran on far01 alone.
What SHC wrote
Section titled “What SHC wrote”The deployment’s rendered compose.yaml (on the node that rendered it, the render leader, under /var/lib/shc/deployments/rig/web/prod/; the node that runs the replica holds no copy of that directory), trimmed for a cluster with service VIPs:
services: nginx: image: nginx:1.27-alpine deploy: resources: limits: {cpus: "0.375", memory: 96M} # quota x 1.5 reservations: {cpus: "0.25", memory: 64M} # quota x 1.0 dns: [10.42.0.1, 1.1.1.1, 8.8.8.8] # the node resolver first expose: ["80"] labels: shc.tenant: rig shc.stack: web shc.environment: prod shc.service: nginx shc.node: "" shc.restart.policy: unless-stopped shc.mount_digest: sha256:07522ee7... logging: driver: json-file options: {max-file: "3", max-size: 50m} networks: __shc_fabric: {ipv4_address: 10.128.1.130} # a pinned address on the fabric default: aliases: [prod.nginx.web] # <environment>.<service>.<stack> restart: "no" # the node agent restarts, not docker volumes: - /var/lib/shc/deployments/rig/web/prod/files/site:/usr/share/nginx/html:ro x-shc-quota: {cpu: "0.25", cpu_source: limit, memory: 64M, memory_source: limit} x-shc-restart: unless-stoppednetworks: default: ipam: config: [{subnet: 10.64.0.0/24, gateway: 10.64.0.1}] # SHC picks the subnetEvery line SHC added is on What SHC does to your compose file; the limit and reservation are the resource model. Beside it SHC wrote one replica project, replicas/nginx-r0/compose.yaml, with the same service as nginx-r0, dns_search: [web.prod.rig.shc.internal] and the replica’s labels (shc.ordinal: "0", shc.pod: nginx, shc.location: home, shc.spec_digest). A Linux node of a cluster on the shc fabric with service VIPs runs that replica project in place of compose.yaml; any other node runs compose.yaml. shc doctor (pods.replica-projects) says which this node does (ce modules/pods/doctor.go).
Remove it
Section titled “Remove it”$ shc uninstall web -t rig -yuninstalling stack "web"web: uninstalling stack "web" on node far01, which holds itweb: taking the safety backup of stack "web"web: taking stack "web" out of every node's service VIP tableweb: tearing down stack "web"web: taking down the replicas of stack "web" on node far01web: removing the copy of stack "web" on node home01uninstalled deployment "web"web: removed with the deployment: config lanEnforcementNext: Step 2: Postgres.