ci(e2e): 新增 Ubuntu 24.04 全新安装+重跑 e2e 与 release→当前 commit 升级 e2e

2026-09-25 01:51:53 +08:00
1 parent 3ed6603201
commit cc85aac906
@@ -0,0 +1,154 @@
+# deploy/bootstrap.sh end to end on a fresh Ubuntu 24.04 x86_64 runner: the paths every
+# host goes through, run for real instead of by hand on a VM.
+#
+#   install  a full install of this commit, then the same commit again (a rerun must
+#            converge without restarting what did not change)
+#   upgrade  the newest published release, then this commit on top of it; skipped until
+#            a release exists
+#
+# Each job builds the control plane, the game images and the proxy from scratch, about
+# half an hour of runner time, so this runs on pushes that touch what gets installed, by
+# hand, and weekly (a moving upstream: apt mirrors, k3s's install script, Adoptium).
+# deploy/e2e_check.sh holds the assertions.
+name: e2e
+
+on:
+  push:
+    branches: [main]
+    paths:
+      - 'deploy/**'
+      - 'cmd/**'
+      - 'internal/**'
+      - 'panel/**'
+      - 'plugins/**'
+      - 'Dockerfile'
+      - 'go.mod'
+      - 'go.sum'
+      - '.github/workflows/e2e.yml'
+  workflow_dispatch:
+  schedule:
+    - cron: '23 4 * * 1'
+
+permissions:
+  contents: read
+
+concurrency:
+  group: e2e-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  install:
+    runs-on: ubuntu-24.04
+    timeout-minutes: 90
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4.4.0
+
+      # The images, k3s and the JRE need more room than a stock runner leaves free.
+      - name: Free disk space
+        run: sudo rm -rf /usr/share/dotnet /usr/local/lib/android /opt/ghc /opt/hostedtoolcache/CodeQL
+
+      # FELIS_SKIP_FETCH builds whatever sits in /opt/felis/src, the way the VM runs
+      # have always been done; the .git directory is what stamps the build. Root owns it
+      # so git, run as root by the installer, does not refuse it as dubious.
+      - name: Stage this commit as the installer's source
+        run: |
+          sudo mkdir -p /opt/felis
+          sudo cp -a "$GITHUB_WORKSPACE" /opt/felis/src
+          sudo chown -R root:root /opt/felis/src
+
+      - name: Install
+        run: sudo FELIS_SKIP_FETCH=1 FELIS_INSTALL_MODE=full bash /opt/felis/src/deploy/bootstrap.sh 2>&1 | tee install.log
+
+      - name: Check the install
+        run: sudo bash deploy/e2e_check.sh install
+
+      - name: Rerun the same commit
+        run: sudo FELIS_SKIP_FETCH=1 FELIS_INSTALL_MODE=full bash /opt/felis/src/deploy/bootstrap.sh 2>&1 | tee rerun.log
+
+      - name: Check the rerun
+        run: |
+          sudo bash deploy/e2e_check.sh rerun
+          grep -q 'felis-velocity unchanged; left running' rerun.log
+
+      - name: Diagnostics
+        if: failure()
+        run: |
+          export KUBECONFIG=/etc/rancher/k3s/k3s.yaml
+          k() { sudo -E /usr/local/bin/k3s kubectl "$@"; }
+          k get pods -A -o wide || true
+          k get events -A --sort-by=.lastTimestamp | tail -n 60 || true
+          for p in $(k get pods -A --no-headers 2>/dev/null | awk '$4 != "Running" && $4 != "Completed" {print $1 "/" $2}'); do
+            k -n "${p%%/*}" describe pod "${p#*/}" | tail -n 40 || true
+            k -n "${p%%/*}" logs "${p#*/}" --all-containers --tail=60 || true
+          done
+          sudo journalctl -u k3s -u felis-velocity -u postgresql --no-pager -n 120 || true
+
+      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+        if: always()
+        with:
+          name: e2e-install-logs
+          path: '*.log'
+          if-no-files-found: ignore
+
+  upgrade:
+    runs-on: ubuntu-24.04
+    timeout-minutes: 120
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4.4.0
+        with:
+          fetch-depth: 0
+
+      - name: Free disk space
+        run: sudo rm -rf /usr/share/dotnet /usr/local/lib/android /opt/ghc /opt/hostedtoolcache/CodeQL
+
+      - name: Find the newest release
+        id: release
+        env:
+          GH_TOKEN: ${{ github.token }}
+        run: |
+          tag="$(gh release view --repo "$GITHUB_REPOSITORY" --json tagName --jq .tagName 2>/dev/null || true)"
+          if [ -z "$tag" ]; then
+            echo "::notice::no published release yet; the upgrade path has nothing to start from"
+          fi
+          echo "tag=${tag}" >> "$GITHUB_OUTPUT"
+
+      # The release's own installer, fetching the release's own binary: what a host that
+      # installed that release is running today.
+      - name: Install the newest release
+        if: steps.release.outputs.tag != ''
+        env:
+          TAG: ${{ steps.release.outputs.tag }}
+          TOKEN: ${{ github.token }}
+        run: |
+          git show "${TAG}:deploy/bootstrap.sh" > release-bootstrap.sh
+          sudo FELIS_GITHUB_TOKEN="$TOKEN" FELIS_REF="$TAG" FELIS_INSTALL_MODE=full bash release-bootstrap.sh 2>&1 | tee release.log
+
+      - name: Check the release install
+        if: steps.release.outputs.tag != ''
+        run: sudo bash deploy/e2e_check.sh install
+
+      - name: Upgrade to this commit
+        if: steps.release.outputs.tag != ''
+        run: |
+          sudo rm -rf /opt/felis/src
+          sudo cp -a "$GITHUB_WORKSPACE" /opt/felis/src
+          sudo chown -R root:root /opt/felis/src
+          sudo FELIS_SKIP_FETCH=1 FELIS_INSTALL_MODE=full bash /opt/felis/src/deploy/bootstrap.sh 2>&1 | tee upgrade.log
+
+      - name: Check the upgrade
+        if: steps.release.outputs.tag != ''
+        run: sudo bash deploy/e2e_check.sh upgrade
+
+      - name: Diagnostics
+        if: failure()
+        run: |
+          export KUBECONFIG=/etc/rancher/k3s/k3s.yaml
+          sudo -E /usr/local/bin/k3s kubectl get pods -A -o wide || true
+          sudo journalctl -u k3s -u felis-velocity -u postgresql --no-pager -n 120 || true
+
+      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+        if: always()
+        with:
+          name: e2e-upgrade-logs
+          path: '*.log'
+          if-no-files-found: ignore
@@ -0,0 +1,118 @@
+#!/bin/bash
+# Checks a host that deploy/bootstrap.sh just installed (or re-ran on). The e2e workflow
+# runs it on a fresh GitHub runner after each of its two installer runs:
+#
+#   sudo bash deploy/e2e_check.sh install   # after the first run
+#   sudo bash deploy/e2e_check.sh rerun     # after the same commit ran again
+#   sudo bash deploy/e2e_check.sh upgrade   # after this commit ran over a release
+#
+# It asks what an operator's first minutes ask: the binary runs, the control plane is
+# rolled out and ready, the panel answers on its NodePort, the proxy answers a Minecraft
+# status ping, and the host timers are there. A rerun must also leave the proxy running
+# (it restarts only when what it runs changed) and keep every earlier answer.
+set -euo pipefail
+
+phase="${1:?usage: e2e_check.sh install|rerun|upgrade}"
+export KUBECONFIG=/etc/rancher/k3s/k3s.yaml
+KUBECTL=(/usr/local/bin/k3s kubectl)
+PID_FILE=/var/tmp/felis-e2e-velocity.pid
+fails=0
+
+pass() { printf 'PASS %s\n' "$*"; }
+fail() { printf 'FAIL %s\n' "$*"; fails=$((fails + 1)); }
+check() { # label command...
+  local label="$1"
+  shift
+  if "$@"; then pass "$label"; else fail "$label"; fi
+}
+
+check "felis version runs" sh -c '/usr/local/bin/felis version | grep -q "^felis "'
+
+for d in felis-api felis-operator registry; do
+  check "deployment ${d} is rolled out" "${KUBECTL[@]}" -n felis rollout status "deploy/${d}" --timeout=180s
+done
+
+node_ip="$(ip -4 route get 1.1.1.1 | awk '{for (i = 1; i <= NF; i++) if ($i == "src") { print $(i + 1); exit }}')"
+check "the panel serves its page on the NodePort" \
+  sh -c "curl -skf --retry 10 --retry-delay 3 --retry-all-errors https://${node_ip}:30443/ | grep -qi '<html'"
+# Readiness lives on the internal face only; a Service ClusterIP routes from the node.
+internal="$("${KUBECTL[@]}" -n felis get svc felis-api-internal -o jsonpath='{.spec.clusterIP}:{.spec.ports[0].port}')"
+check "felis-api is ready (database and cluster reachable)" \
+  curl -sf --retry 10 --retry-delay 3 --retry-all-errors -o /dev/null "http://${internal}/readyz"
+
+for unit in k3s postgresql felis-velocity; do
+  check "${unit} is active" systemctl is-active --quiet "$unit"
+done
+for timer in felis-db-backup.timer felis-watchdog.timer; do
+  check "${timer} is scheduled" systemctl is-enabled --quiet "$timer"
+done
+
+# A status ping is the proxy's own answer (ping passthrough is off), so it proves the JRE,
+# Velocity and its config without a login gate or a Mojang account.
+ping_proxy() {
+  python3 - <<'EOF'
+import json, socket, struct, sys
+
+def varint(n):
+    out = b""
+    n &= 0xFFFFFFFF
+    while True:
+        b, n = n & 0x7F, n >> 7
+        if n:
+            out += bytes([b | 0x80])
+        else:
+            return out + bytes([b])
+
+def read_varint(s):
+    n = 0
+    for i in range(5):
+        b = s.recv(1)
+        if not b:
+            raise EOFError("connection closed")
+        n |= (b[0] & 0x7F) << (7 * i)
+        if not b[0] & 0x80:
+            return n
+    raise ValueError("varint too long")
+
+host, port = "127.0.0.1", 25565
+s = socket.create_connection((host, port), timeout=10)
+hs = varint(0) + varint(767) + varint(len(host)) + host.encode() + struct.pack(">H", port) + varint(1)
+s.sendall(varint(len(hs)) + hs + varint(1) + varint(0))
+read_varint(s)
+read_varint(s)
+size = read_varint(s)
+data = b""
+while len(data) < size:
+    chunk = s.recv(size - len(data))
+    if not chunk:
+        raise EOFError("short status response")
+    data += chunk
+print(json.loads(data)["version"]["name"])
+EOF
+}
+if version="$(ping_proxy 2>&1)"; then
+  pass "the proxy answers a status ping (${version})"
+else
+  fail "the proxy answers a status ping: ${version}"
+fi
+
+pid="$(systemctl show -p MainPID --value felis-velocity)"
+case "$phase" in
+  install) printf '%s\n' "$pid" > "$PID_FILE" ;;
+  rerun)
+    if [ "$pid" = "$(cat "$PID_FILE" 2>/dev/null)" ]; then
+      pass "the rerun left the proxy running (pid ${pid})"
+    else
+      fail "the rerun restarted the proxy (pid $(cat "$PID_FILE" 2>/dev/null || echo '?') -> ${pid}) though nothing it runs changed"
+    fi
+    ;;
+  upgrade) ;; # a new release may well change what the proxy runs
+  *) fail "unknown phase ${phase}" ;;
+esac
+
+if [ "$fails" -eq 0 ]; then
+  echo "ALL PASS (${phase})"
+else
+  echo "${fails} FAILED (${phase})"
+fi
+exit "$fails"