fix(cluster): 以宁波专用节点替换深圳 Worker

移除深圳节点及中继拓扑,新增第二台宁波 K3s agent 的全互联 WireGuard 接入和严格 UFW 门禁。\n\nWorker Deployment 与容量控制器仅选择 easyai.io/worker=true 节点,使原宁波混部节点退出 Worker 资源预算,生产基线恢复为宁波专用节点与香港节点各一实例。\n\n已通过 Go 全量测试、go vet、gofmt、迁移安全检查、bash -n、ShellCheck、发布脚本测试和 Kubernetes 清单渲染。
This commit is contained in:
2026-08-01 10:47:01 +08:00
parent d3b36cf63d
commit 53589fe3ef
12 changed files with 119 additions and 155 deletions
@@ -172,7 +172,9 @@ func (client *KubernetesClient) SiteState(ctx context.Context, site string) (Kub
} `json:"status"`
} `json:"items"`
}
nodePath := "/api/v1/nodes?labelSelector=" + url.QueryEscape("easyai.io/site="+site)
nodePath := "/api/v1/nodes?labelSelector=" + url.QueryEscape(
"easyai.io/site="+site+",easyai.io/worker=true",
)
if err := client.getJSON(ctx, nodePath, &nodes); err != nil {
return KubernetesSiteState{}, err
}
@@ -27,6 +27,10 @@ func TestKubernetesClientReadsSiteAndMutatesOnlyWorkerScale(t *testing.T) {
}]}}}
}`))
case request.Method == http.MethodGet && request.URL.Path == "/api/v1/nodes":
if request.URL.Query().Get("labelSelector") != "easyai.io/site=hongkong,easyai.io/worker=true" {
http.Error(w, "unexpected Worker node selector", http.StatusBadRequest)
return
}
_, _ = w.Write([]byte(`{"items":[{"metadata":{"name":"easyai-hongkong"},"status":{
"allocatable":{"memory":"8Gi","cpu":"4"},
"conditions":[{"type":"MemoryPressure","status":"False"}]
@@ -458,6 +458,7 @@ spec:
nodeSelector:
easyai.io/site: ningbo
easyai.io/workload: "true"
easyai.io/worker: "true"
tolerations:
- key: easyai.io/worker-only
operator: Equal
@@ -622,6 +623,7 @@ spec:
nodeSelector:
easyai.io/site: hongkong
easyai.io/workload: "true"
easyai.io/worker: "true"
tolerations:
- key: easyai.io/worker-only
operator: Equal
+11 -10
View File
@@ -6,22 +6,23 @@
CloudNativePG;洛杉矶只作为 K3s server/etcd 仲裁节点,并带
`easyai.io/witness=true:NoSchedule` 污点。
深圳扩展节点是纯 K3s agent,不加入 etcd 或 CloudNativePG。它使用节点名
`easyai-shenzhen`、WireGuard 地址 `10.77.0.4`,并归入逻辑 `hongkong` Worker 池。
第二台宁波服务器是纯 K3s agent,不加入 etcd 或 CloudNativePG。它使用节点名
`easyai-ningbo-worker-2`、WireGuard 地址 `10.77.0.4`,并归入逻辑 `ningbo` Worker 池。
`easyai.io/worker-only=true:NoSchedule` 污点确保除 Worker 外的业务 Pod 不会调度到该机;
香港池副本通过 hostname topology spread 分散到香港、深圳。接入使用:
`easyai.io/worker=true` 标签确保容量控制器只把专用 Worker 节点纳入资源预算。接入使用:
```bash
./scripts/cluster/add-shenzhen-worker-node.sh all
./scripts/cluster/add-ningbo-worker-node.sh all
```
香港—深圳公网直连实测无法满足生产链路门禁,因此两地 `10.77.0.2/10.77.0.4` 数据前缀
固定经宁波 WireGuard 中继;香港—深圳直接 peer 只保留加密心跳。巡检仍以两端实际数据路径
10 包丢包率低于 1%、平均 RTT 低于 80 ms 为硬门禁,禁止自动回退到劣化的公网直连。
新节点与宁波、香港、洛杉矶使用 WireGuard 全互联直连,不经过中继。公网 K3s 直连虽然可以
单独保护 API TLS,但当前 Flannel、Pod 和数据库网络依赖 `10.77.0.0/24`,禁止改成未加密的
公网 VXLAN。巡检以 10 包丢包率低于 1%为硬门禁;宁波和香港路径 RTT 低于 80 ms,洛杉矶
路径低于 300 ms。
本地 `.env.local` 只保存 `AI_GATEWAY_SHENZHEN_HOST` 和 SSH 接入密码,必须保持 `0600`
不得进入 Git。深圳已有宿主机服务时,接入脚本保留其公开端口,封锁 K3s API、kubelet、
VXLAN 和除既有 `31058` 外的 NodePort 公网入口。深圳防火墙继续由原有 UFW 管理,接入脚本
本地 `.env.local` 只保存 `AI_GATEWAY_NINGBO_WORKER_HOST` 和 SSH 接入密码,必须保持 `0600`
不得进入 Git。新节点只保留原有 SSH 入口,封锁 K3s API、kubelet、VXLAN 和全部 NodePort
公网入口。节点防火墙继续由原有 UFW 管理,接入脚本
禁止安装与 UFW 冲突的 `iptables-persistent`;UFW 未安装或未启用时必须在预检阶段停止。
生产上线前的双节点高媒体负载、Worker 强杀和 P24/P28/P32 容量搜索见
+10 -9
View File
@@ -89,21 +89,22 @@ AI_GATEWAY_WORKER_REPLICAS_HONGKONG
各一个 2 GiB Worker,再按实时内存、CPU 和 PostgreSQL 预算验证 `1+2`、条件允许时的
`2+2`,以及安全 drain 回到 `1+1`
宁波因正式混部不具备 Worker 余量时,验收可改用 `0+2` 基线:
生产基线使用两个独立物理 Worker 节点:第二台宁波服务器承载 `easyai-worker-ningbo`
香港服务器承载 `easyai-worker-hongkong`。两个 Deployment 都要求节点带
`easyai.io/worker=true`,原宁波混部节点不再承载 Worker。默认验收配置为:
```bash
AI_GATEWAY_ACCEPTANCE_BASE_REPLICAS_NINGBO=0
AI_GATEWAY_ACCEPTANCE_BASE_REPLICAS_HONGKONG=2
AI_GATEWAY_ACCEPTANCE_AUTOSCALING_MIN_REPLICAS_NINGBO=0
AI_GATEWAY_ACCEPTANCE_BASE_REPLICAS_NINGBO=1
AI_GATEWAY_ACCEPTANCE_BASE_REPLICAS_HONGKONG=1
AI_GATEWAY_ACCEPTANCE_AUTOSCALING_MIN_REPLICAS_NINGBO=1
AI_GATEWAY_ACCEPTANCE_AUTOSCALING_MIN_REPLICAS_HONGKONG=1
AI_GATEWAY_ACCEPTANCE_AUTOSCALING_MAX_REPLICAS_NINGBO=0
AI_GATEWAY_ACCEPTANCE_AUTOSCALING_MAX_REPLICAS_NINGBO=2
AI_GATEWAY_ACCEPTANCE_AUTOSCALING_MAX_REPLICAS_HONGKONG=2
```
此时香港逻辑池的两个 Worker 必须通过 hostname topology spread 分布在香港和深圳物理节点
固定容量仍是两个实例,自动伸缩阶段验证 `0+1 → 0+2 → 0+1`,最终 certified profile
恢复 `0+2` 最小副本以保留双物理节点容灾。资源前置门禁采样实际启用的 Worker 站点节点,
宁波仍受运行时 80% 目标和 85% 硬门禁约束。
资源前置门禁和容量控制器只采样 `easyai.io/worker=true` 节点,避免把原宁波混部节点
可分配资源错误计入弹性上限。每站点能否升到两个副本仍由实测 RSS、CPU 和 PostgreSQL
连接预算决定,不因新增节点而预先承诺。
| 档位 | 单实例执行槽 | 数据库池 | 媒体并发 | 全局执行槽 |
|---|---:|---:|---:|---:|
@@ -8,15 +8,15 @@ source "$script_dir/common.sh"
action=${1:-all}
[[ $action == preflight || $action == network || $action == install ||
$action == verify || $action == all ]] || {
echo 'usage: add-shenzhen-worker-node.sh [preflight|network|install|verify|all]' >&2
echo 'usage: add-ningbo-worker-node.sh [preflight|network|install|verify|all]' >&2
exit 64
}
load_cluster_env
require_commands ssh scp shasum
: "${CLUSTER_SHENZHEN_HOST:?AI_GATEWAY_SHENZHEN_HOST is required}"
: "${CLUSTER_NINGBO_WORKER_HOST:?AI_GATEWAY_NINGBO_WORKER_HOST is required}"
node_name=easyai-shenzhen
node_name=easyai-ningbo-worker-2
node_ip=10.77.0.4
k3s_version=v1.36.2+k3s1
k3s_sha256=65a55ec56c24eab44383086166ec620a491952b7e23941a49ddca6e8a4c4b4de
@@ -38,11 +38,12 @@ assert_join_material() {
}
preflight_node() {
cluster_ssh "$CLUSTER_SHENZHEN_HOST" bash -s <<'REMOTE'
cluster_ssh "$CLUSTER_NINGBO_WORKER_HOST" bash -s <<'REMOTE'
set -euo pipefail
[[ $(id -u) -eq 0 && $(uname -m) == x86_64 ]]
. /etc/os-release
[[ $ID == debian && ${VERSION_ID%%.*} -ge 12 ]]
[[ ($ID == debian && ${VERSION_ID%%.*} -ge 12) ||
($ID == ubuntu && ${VERSION_ID%%.*} -ge 24) ]]
(( $(nproc) >= 4 ))
available_kib=$(awk '/MemAvailable:/ {print $2}' /proc/meminfo)
(( available_kib >= 6 * 1024 * 1024 ))
@@ -52,7 +53,7 @@ timedatectl show -p NTPSynchronized --value | grep -Fxq yes
[[ ! -e /etc/rancher/k3s/config.yaml || -f /etc/rancher/k3s/config.yaml ]]
command -v ufw >/dev/null
ufw status | grep -Fqx 'Status: active'
printf 'shenzhen_worker_preflight=PASS cpu=%s memory_available_mib=%s disk_available_gib=%s\n' \
printf 'ningbo_worker_preflight=PASS cpu=%s memory_available_mib=%s disk_available_gib=%s\n' \
"$(nproc)" "$((available_kib / 1024))" "$((available_bytes / 1024 / 1024 / 1024))"
REMOTE
}
@@ -62,9 +63,9 @@ configure_wireguard() {
local -a server_ips=(10.77.0.1 10.77.0.2 10.77.0.3)
local -a server_endpoints=("${CLUSTER_NINGBO_HOST#*@}" "${CLUSTER_HONGKONG_HOST#*@}" "${CLUSTER_LOS_ANGELES_HOST#*@}")
local -a server_keys=()
local shenzhen_key host key
local worker_key host key
shenzhen_key=$(cluster_ssh "$CLUSTER_SHENZHEN_HOST" 'bash -s' <<'REMOTE' | tail -n 1
worker_key=$(cluster_ssh "$CLUSTER_NINGBO_WORKER_HOST" 'bash -s' <<'REMOTE' | tail -n 1
set -euo pipefail
export DEBIAN_FRONTEND=noninteractive
if ! command -v wg >/dev/null 2>&1; then
@@ -81,7 +82,7 @@ wg pubkey </etc/wireguard/privatekey >/etc/wireguard/publickey
cat /etc/wireguard/publickey
REMOTE
)
[[ $shenzhen_key =~ ^[A-Za-z0-9+/]{43}=$ ]]
[[ $worker_key =~ ^[A-Za-z0-9+/]{43}=$ ]]
for host in "${server_hosts[@]}"; do
key=$(cluster_ssh "$host" 'wg show wg0 public-key')
@@ -89,7 +90,7 @@ REMOTE
server_keys+=("$key")
done
cluster_ssh "$CLUSTER_SHENZHEN_HOST" bash -s -- \
cluster_ssh "$CLUSTER_NINGBO_WORKER_HOST" bash -s -- \
"${server_keys[0]}" "${server_endpoints[0]}" "${server_ips[0]}" \
"${server_keys[1]}" "${server_endpoints[1]}" "${server_ips[1]}" \
"${server_keys[2]}" "${server_endpoints[2]}" "${server_ips[2]}" <<'REMOTE'
@@ -131,43 +132,13 @@ systemctl restart wg-quick@wg0
REMOTE
for host in "${server_hosts[@]}"; do
cluster_ssh "$host" bash -s -- "$shenzhen_key" "${CLUSTER_SHENZHEN_HOST#*@}" "$node_ip" <<'REMOTE'
cluster_ssh "$host" bash -s -- "$worker_key" "${CLUSTER_NINGBO_WORKER_HOST#*@}" "$node_ip" <<'REMOTE'
set -euo pipefail
wg set wg0 peer "$1" allowed-ips "$3/32" endpoint "$2:51820" persistent-keepalive 25
wg-quick save wg0
REMOTE
done
# The Hong Kong-Shenzhen public route is unstable. Keep the direct peers for
# encrypted liveness, but route their data prefixes through Ningbo's healthy
# WireGuard paths. AllowedIPs remain unique, so WireGuard selects the relay
# deterministically instead of oscillating between direct and relayed paths.
cluster_ssh "$CLUSTER_HONGKONG_HOST" bash -s -- \
"${server_keys[0]}" "$shenzhen_key" <<'REMOTE'
set -euo pipefail
wg set wg0 peer "$1" allowed-ips 10.77.0.1/32,10.77.0.4/32
wg set wg0 peer "$2" allowed-ips 0.0.0.0/32
wg-quick save wg0
REMOTE
cluster_ssh "$CLUSTER_SHENZHEN_HOST" bash -s -- \
"${server_keys[0]}" "${server_keys[1]}" <<'REMOTE'
set -euo pipefail
wg set wg0 peer "$1" allowed-ips 10.77.0.1/32,10.77.0.2/32
wg set wg0 peer "$2" allowed-ips 0.0.0.0/32
wg-quick save wg0
REMOTE
cluster_ssh "$CLUSTER_NINGBO_HOST" 'bash -s' <<'REMOTE'
set -euo pipefail
sysctl -qw net.ipv4.ip_forward=1
ensure_forward_rule() {
if ! iptables -C FORWARD "$@" 2>/dev/null; then iptables -I FORWARD 1 "$@"; fi
}
ensure_forward_rule -i wg0 -o wg0 -s 10.77.0.2/32 -d 10.77.0.4/32 \
-m comment --comment easyai-wireguard-hongkong-shenzhen -j ACCEPT
ensure_forward_rule -i wg0 -o wg0 -s 10.77.0.4/32 -d 10.77.0.2/32 \
-m comment --comment easyai-wireguard-shenzhen-hongkong -j ACCEPT
netfilter-persistent save >/dev/null
REMOTE
verify_wireguard
}
@@ -183,19 +154,19 @@ verify_path() {
}
verify_wireguard() {
verify_path "$CLUSTER_NINGBO_HOST" "$node_ip" 80 ningbo_to_shenzhen
verify_path "$CLUSTER_SHENZHEN_HOST" 10.77.0.1 80 shenzhen_to_ningbo
verify_path "$CLUSTER_HONGKONG_HOST" "$node_ip" 80 hongkong_to_shenzhen
verify_path "$CLUSTER_SHENZHEN_HOST" 10.77.0.2 80 shenzhen_to_hongkong
verify_path "$CLUSTER_LOS_ANGELES_HOST" "$node_ip" 300 losangeles_to_shenzhen
verify_path "$CLUSTER_SHENZHEN_HOST" 10.77.0.3 300 shenzhen_to_losangeles
verify_path "$CLUSTER_NINGBO_HOST" "$node_ip" 80 ningbo_to_worker_2
verify_path "$CLUSTER_NINGBO_WORKER_HOST" 10.77.0.1 80 worker_2_to_ningbo
verify_path "$CLUSTER_HONGKONG_HOST" "$node_ip" 80 hongkong_to_worker_2
verify_path "$CLUSTER_NINGBO_WORKER_HOST" 10.77.0.2 80 worker_2_to_hongkong
verify_path "$CLUSTER_LOS_ANGELES_HOST" "$node_ip" 300 losangeles_to_worker_2
verify_path "$CLUSTER_NINGBO_WORKER_HOST" 10.77.0.3 300 worker_2_to_losangeles
local host recent
for host in "$CLUSTER_NINGBO_HOST" "$CLUSTER_HONGKONG_HOST" "$CLUSTER_LOS_ANGELES_HOST" "$CLUSTER_SHENZHEN_HOST"; do
for host in "$CLUSTER_NINGBO_HOST" "$CLUSTER_HONGKONG_HOST" "$CLUSTER_LOS_ANGELES_HOST" "$CLUSTER_NINGBO_WORKER_HOST"; do
recent=$(cluster_ssh "$host" \
"wg show wg0 latest-handshakes | awk '\$2 > 0 && systime()-\$2 < 180 {count++} END {print count+0}'")
[[ $recent == 3 ]]
done
echo 'shenzhen_wireguard=PASS peers=4 recent_handshakes=12 hongkong_shenzhen_via=ningbo'
echo 'ningbo_worker_wireguard=PASS peers=4 recent_handshakes=12 routing=direct'
}
install_agent() {
@@ -205,12 +176,12 @@ install_agent() {
: "${K3S_TOKEN:?}"
[[ $K3S_TOKEN =~ ^[0-9a-f]{64}$ ]]
cluster_ssh "$CLUSTER_SHENZHEN_HOST" 'install -d -m 0700 /etc/rancher/k3s'
printf '%s\n' "$K3S_TOKEN" | cluster_ssh "$CLUSTER_SHENZHEN_HOST" \
cluster_ssh "$CLUSTER_NINGBO_WORKER_HOST" 'install -d -m 0700 /etc/rancher/k3s'
printf '%s\n' "$K3S_TOKEN" | cluster_ssh "$CLUSTER_NINGBO_WORKER_HOST" \
'umask 077; cat >/etc/rancher/k3s/cluster-token; chmod 0600 /etc/rancher/k3s/cluster-token'
cluster_scp "$k3s_binary" "$CLUSTER_SHENZHEN_HOST:/usr/local/bin/k3s"
cluster_scp "$registries_file" "$CLUSTER_SHENZHEN_HOST:/etc/rancher/k3s/registries.yaml"
cluster_ssh "$CLUSTER_SHENZHEN_HOST" bash -s -- "$k3s_version" "$k3s_sha256" <<'REMOTE'
cluster_scp "$k3s_binary" "$CLUSTER_NINGBO_WORKER_HOST:/usr/local/bin/k3s"
cluster_scp "$registries_file" "$CLUSTER_NINGBO_WORKER_HOST:/etc/rancher/k3s/registries.yaml"
cluster_ssh "$CLUSTER_NINGBO_WORKER_HOST" bash -s -- "$k3s_version" "$k3s_sha256" <<'REMOTE'
set -euo pipefail
version=$1
expected_sha256=$2
@@ -220,11 +191,11 @@ umask 077
cat >/etc/rancher/k3s/config.yaml <<'EOF'
server: https://10.77.0.1:6443
token-file: /etc/rancher/k3s/cluster-token
node-name: easyai-shenzhen
node-name: easyai-ningbo-worker-2
node-ip: 10.77.0.4
flannel-iface: wg0
node-label:
- easyai.io/site=hongkong
- easyai.io/site=ningbo
- easyai.io/workload=true
- easyai.io/database=false
- easyai.io/worker=true
@@ -241,10 +212,8 @@ ufw deny in on "$public_interface" to any port 6443 proto tcp comment easyai-k3s
ufw deny in on "$public_interface" to any port 10250 proto tcp comment easyai-kubelet-public-deny >/dev/null
ufw deny in on "$public_interface" to any port 8472 proto udp comment easyai-flannel-public-deny >/dev/null
for protocol in tcp udp; do
ufw deny in on "$public_interface" to any port 30000:31057 proto "$protocol" \
comment easyai-nodeport-public-deny-low >/dev/null
ufw deny in on "$public_interface" to any port 31059:32767 proto "$protocol" \
comment easyai-nodeport-public-deny-high >/dev/null
ufw deny in on "$public_interface" to any port 30000:32767 proto "$protocol" \
comment easyai-nodeport-public-deny >/dev/null
done
install_script=$(mktemp)
trap 'rm -f -- "$install_script"' EXIT
@@ -267,6 +236,13 @@ REMOTE
sleep 2
done
[[ $ready == true ]]
cluster_ssh "$CLUSTER_NINGBO_HOST" 'bash -s' <<'REMOTE'
set -euo pipefail
kubectl='k3s kubectl'
$kubectl label node easyai-ningbo easyai.io/worker=false --overwrite >/dev/null
$kubectl label node easyai-hongkong easyai.io/worker=true --overwrite >/dev/null
$kubectl label node easyai-los-angeles easyai.io/worker=false --overwrite >/dev/null
REMOTE
verify_agent
}
@@ -277,13 +253,15 @@ set -euo pipefail
node=$1
kubectl='k3s kubectl'
[[ $($kubectl get node "$node" -o jsonpath='{.status.conditions[?(@.type=="Ready")].status}') == True ]]
[[ $($kubectl get node "$node" -o jsonpath='{.metadata.labels.easyai\.io/site}') == hongkong ]]
[[ $($kubectl get node "$node" -o jsonpath='{.metadata.labels.easyai\.io/site}') == ningbo ]]
[[ $($kubectl get node "$node" -o jsonpath='{.metadata.labels.easyai\.io/workload}') == true ]]
[[ $($kubectl get node "$node" -o jsonpath='{.metadata.labels.easyai\.io/database}') == false ]]
[[ $($kubectl get node easyai-hongkong -o jsonpath='{.metadata.labels.easyai\.io/worker}') == true ]]
[[ $($kubectl get node easyai-ningbo -o jsonpath='{.metadata.labels.easyai\.io/worker}') == false ]]
[[ $($kubectl get node "$node" -o json | jq '[.spec.taints[]? | select(.key=="easyai.io/worker-only" and .effect=="NoSchedule")] | length') -eq 1 ]]
[[ $($kubectl get pods -A --field-selector "spec.nodeName=$node" -o json |
jq '[.items[] | select(.metadata.namespace!="kube-system")] | length') -eq 0 ]]
printf 'shenzhen_k3s_agent=PASS node=%s role=worker-only etcd_member=false\n' "$node"
printf 'ningbo_worker_k3s_agent=PASS node=%s role=worker-only etcd_member=false\n' "$node"
REMOTE
}
+3 -3
View File
@@ -256,9 +256,9 @@ REMOTE
cluster_ssh "$CLUSTER_NINGBO_HOST" 'bash -s' <<'REMOTE'
set -euo pipefail
kubectl='k3s kubectl'
$kubectl label node easyai-ningbo easyai.io/site=ningbo easyai.io/workload=true easyai.io/database=true --overwrite
$kubectl label node easyai-hongkong easyai.io/site=hongkong easyai.io/workload=true easyai.io/database=true --overwrite
$kubectl label node easyai-los-angeles easyai.io/site=los-angeles easyai.io/workload=false easyai.io/database=false --overwrite
$kubectl label node easyai-ningbo easyai.io/site=ningbo easyai.io/workload=true easyai.io/database=true easyai.io/worker=false --overwrite
$kubectl label node easyai-hongkong easyai.io/site=hongkong easyai.io/workload=true easyai.io/database=true easyai.io/worker=true --overwrite
$kubectl label node easyai-los-angeles easyai.io/site=los-angeles easyai.io/workload=false easyai.io/database=false easyai.io/worker=false --overwrite
$kubectl taint node easyai-los-angeles easyai.io/witness=true:NoSchedule --overwrite
k3s secrets-encrypt status
k3s etcd-snapshot save --name post-bootstrap
+4 -37
View File
@@ -11,9 +11,9 @@ require_commands ssh scp
hosts=("$CLUSTER_NINGBO_HOST" "$CLUSTER_HONGKONG_HOST" "$CLUSTER_LOS_ANGELES_HOST")
public_endpoints=("${CLUSTER_NINGBO_HOST#*@}" "${CLUSTER_HONGKONG_HOST#*@}" "${CLUSTER_LOS_ANGELES_HOST#*@}")
wireguard_ips=(10.77.0.1 10.77.0.2 10.77.0.3)
if [[ -n $CLUSTER_SHENZHEN_HOST ]]; then
hosts+=("$CLUSTER_SHENZHEN_HOST")
public_endpoints+=("${CLUSTER_SHENZHEN_HOST#*@}")
if [[ -n $CLUSTER_NINGBO_WORKER_HOST ]]; then
hosts+=("$CLUSTER_NINGBO_WORKER_HOST")
public_endpoints+=("${CLUSTER_NINGBO_WORKER_HOST#*@}")
wireguard_ips+=(10.77.0.4)
fi
public_keys=()
@@ -90,35 +90,6 @@ systemctl restart wg-quick@wg0
REMOTE
done
if (( ${#hosts[@]} == 4 )); then
cluster_ssh "$CLUSTER_HONGKONG_HOST" bash -s -- \
"${public_keys[0]}" "${public_keys[3]}" <<'REMOTE'
set -euo pipefail
wg set wg0 peer "$1" allowed-ips 10.77.0.1/32,10.77.0.4/32
wg set wg0 peer "$2" allowed-ips 0.0.0.0/32
wg-quick save wg0
REMOTE
cluster_ssh "$CLUSTER_SHENZHEN_HOST" bash -s -- \
"${public_keys[0]}" "${public_keys[1]}" <<'REMOTE'
set -euo pipefail
wg set wg0 peer "$1" allowed-ips 10.77.0.1/32,10.77.0.2/32
wg set wg0 peer "$2" allowed-ips 0.0.0.0/32
wg-quick save wg0
REMOTE
cluster_ssh "$CLUSTER_NINGBO_HOST" 'bash -s' <<'REMOTE'
set -euo pipefail
sysctl -qw net.ipv4.ip_forward=1
ensure_forward_rule() {
if ! iptables -C FORWARD "$@" 2>/dev/null; then iptables -I FORWARD 1 "$@"; fi
}
ensure_forward_rule -i wg0 -o wg0 -s 10.77.0.2/32 -d 10.77.0.4/32 \
-m comment --comment easyai-wireguard-hongkong-shenzhen -j ACCEPT
ensure_forward_rule -i wg0 -o wg0 -s 10.77.0.4/32 -d 10.77.0.2/32 \
-m comment --comment easyai-wireguard-shenzhen-hongkong -j ACCEPT
netfilter-persistent save >/dev/null
REMOTE
fi
for source_index in "${!hosts[@]}"; do
for target_index in "${!hosts[@]}"; do
[[ $source_index -eq $target_index ]] && continue
@@ -136,8 +107,4 @@ for host in "${hosts[@]}"; do
}
done
if (( ${#hosts[@]} == 4 )); then
echo 'wireguard_bootstrap=PASS peers=4 hongkong_shenzhen_via=ningbo'
else
echo "wireguard_bootstrap=PASS peers=${#hosts[@]}"
fi
echo "wireguard_bootstrap=PASS peers=${#hosts[@]} routing=direct"
+4 -4
View File
@@ -28,8 +28,8 @@ load_cluster_env() {
CLUSTER_NINGBO_HOST=${AI_GATEWAY_NINGBO_HOST:-$AI_GATEWAY_DEPLOY_HOST}
CLUSTER_LOS_ANGELES_HOST=${AI_GATEWAY_LOS_ANGELES_HOST:-$AI_GATEWAY_DEPLOY_HOST_2}
CLUSTER_HONGKONG_HOST=${AI_GATEWAY_HONGKONG_HOST:-$AI_GATEWAY_DEPLOY_HOST_3}
CLUSTER_SHENZHEN_HOST=${AI_GATEWAY_SHENZHEN_HOST:-}
export CLUSTER_NINGBO_HOST CLUSTER_LOS_ANGELES_HOST CLUSTER_HONGKONG_HOST CLUSTER_SHENZHEN_HOST
CLUSTER_NINGBO_WORKER_HOST=${AI_GATEWAY_NINGBO_WORKER_HOST:-}
export CLUSTER_NINGBO_HOST CLUSTER_LOS_ANGELES_HOST CLUSTER_HONGKONG_HOST CLUSTER_NINGBO_WORKER_HOST
for host in "$CLUSTER_NINGBO_HOST" "$CLUSTER_HONGKONG_HOST" "$CLUSTER_LOS_ANGELES_HOST"; do
[[ $host =~ ^root@[A-Za-z0-9.-]+$ ]] || {
@@ -37,8 +37,8 @@ load_cluster_env() {
return 1
}
done
if [[ -n $CLUSTER_SHENZHEN_HOST && ! $CLUSTER_SHENZHEN_HOST =~ ^root@[A-Za-z0-9.-]+$ ]]; then
echo "Shenzhen Worker host must use root@host syntax" >&2
if [[ -n $CLUSTER_NINGBO_WORKER_HOST && ! $CLUSTER_NINGBO_WORKER_HOST =~ ^root@[A-Za-z0-9.-]+$ ]]; then
echo "dedicated Ningbo Worker host must use root@host syntax" >&2
return 1
fi
[[ $ALI_REGION =~ ^[a-z0-9-]+$ && $ALI_BUCKET =~ ^[A-Za-z0-9.-]+$ ]] || {
@@ -140,8 +140,8 @@ pool_saturated_pods() {
check_wireguard() {
local -a hosts=("$CLUSTER_NINGBO_HOST" "$CLUSTER_HONGKONG_HOST" "$CLUSTER_LOS_ANGELES_HOST")
local -a ips=(10.77.0.1 10.77.0.2 10.77.0.3)
[[ -z $CLUSTER_SHENZHEN_HOST ]] || {
hosts+=("$CLUSTER_SHENZHEN_HOST")
[[ -z $CLUSTER_NINGBO_WORKER_HOST ]] || {
hosts+=("$CLUSTER_NINGBO_WORKER_HOST")
ips+=(10.77.0.4)
}
local source target output loss average handshakes peer_count
+28 -20
View File
@@ -652,17 +652,25 @@ verify_release_cas() {
-o 'jsonpath={.spec.replicas}') == "$AI_GATEWAY_ACCEPTANCE_BASE_REPLICAS_NINGBO" ]]
[[ $(remote_kubectl get deployment easyai-worker-hongkong -n "$namespace" \
-o 'jsonpath={.spec.replicas}') == "$AI_GATEWAY_ACCEPTANCE_BASE_REPLICAS_HONGKONG" ]]
if (( AI_GATEWAY_ACCEPTANCE_BASE_REPLICAS_NINGBO == 0 &&
AI_GATEWAY_ACCEPTANCE_BASE_REPLICAS_HONGKONG == 2 )) &&
[[ -n $CLUSTER_SHENZHEN_HOST ]]; then
local worker_node pod_count
for worker_node in easyai-hongkong easyai-shenzhen; do
if [[ -n $CLUSTER_NINGBO_WORKER_HOST ]]; then
local expected_node expected_replicas site pod_count
for site in ningbo hongkong; do
case $site in
ningbo)
expected_node=easyai-ningbo-worker-2
expected_replicas=$AI_GATEWAY_ACCEPTANCE_BASE_REPLICAS_NINGBO
;;
hongkong)
expected_node=easyai-hongkong
expected_replicas=$AI_GATEWAY_ACCEPTANCE_BASE_REPLICAS_HONGKONG
;;
esac
pod_count=$(remote_kubectl get pods -n "$namespace" \
-l 'app.kubernetes.io/name=easyai-worker,easyai.io/site=hongkong' \
--field-selector "spec.nodeName=$worker_node" -o json | jq \
-l "app.kubernetes.io/name=easyai-worker,easyai.io/site=$site" \
--field-selector "spec.nodeName=$expected_node" -o json | jq \
'[.items[] | select(.status.phase=="Running")] | length')
[[ $pod_count == 1 ]] || {
echo "Hong Kong logical Worker pool is not spread across both physical nodes: node=$worker_node running_pods=$pod_count" >&2
[[ $pod_count == "$expected_replicas" ]] || {
echo "Worker placement does not match the dedicated node: site=$site node=$expected_node running_pods=$pod_count expected=$expected_replicas" >&2
return 1
}
done
@@ -674,12 +682,12 @@ verify_resource_preconditions() {
local worker_nodes los_angeles_nodes invalid_mixed_workloads witness_workloads
worker_nodes=
if (( AI_GATEWAY_ACCEPTANCE_BASE_REPLICAS_NINGBO > 0 )); then
worker_nodes+=$(remote_kubectl get nodes -l 'easyai.io/site=ningbo' \
worker_nodes+=$(remote_kubectl get nodes -l 'easyai.io/site=ningbo,easyai.io/worker=true' \
-o jsonpath='{range .items[*]}{.metadata.name}{"\n"}{end}')
worker_nodes+=$'\n'
fi
if (( AI_GATEWAY_ACCEPTANCE_BASE_REPLICAS_HONGKONG > 0 )); then
worker_nodes+=$(remote_kubectl get nodes -l 'easyai.io/site=hongkong' \
worker_nodes+=$(remote_kubectl get nodes -l 'easyai.io/site=hongkong,easyai.io/worker=true' \
-o jsonpath='{range .items[*]}{.metadata.name}{"\n"}{end}')
fi
worker_nodes=$(printf '%s\n' "$worker_nodes" | sed '/^$/d' | LC_ALL=C sort -u)
@@ -2409,13 +2417,13 @@ verify_cluster_links() {
verify_wireguard "$CLUSTER_LOS_ANGELES_HOST" 10.77.0.1 300 'los_angeles_to_ningbo'
verify_wireguard "$CLUSTER_HONGKONG_HOST" 10.77.0.3 300 'hongkong_to_los_angeles'
verify_wireguard "$CLUSTER_LOS_ANGELES_HOST" 10.77.0.2 300 'los_angeles_to_hongkong'
if [[ -n $CLUSTER_SHENZHEN_HOST ]]; then
verify_wireguard "$CLUSTER_NINGBO_HOST" 10.77.0.4 80 'ningbo_to_shenzhen'
verify_wireguard "$CLUSTER_SHENZHEN_HOST" 10.77.0.1 80 'shenzhen_to_ningbo'
verify_wireguard "$CLUSTER_HONGKONG_HOST" 10.77.0.4 80 'hongkong_to_shenzhen'
verify_wireguard "$CLUSTER_SHENZHEN_HOST" 10.77.0.2 80 'shenzhen_to_hongkong'
verify_wireguard "$CLUSTER_LOS_ANGELES_HOST" 10.77.0.4 300 'los_angeles_to_shenzhen'
verify_wireguard "$CLUSTER_SHENZHEN_HOST" 10.77.0.3 300 'shenzhen_to_los_angeles'
if [[ -n $CLUSTER_NINGBO_WORKER_HOST ]]; then
verify_wireguard "$CLUSTER_NINGBO_HOST" 10.77.0.4 80 'ningbo_to_worker_2'
verify_wireguard "$CLUSTER_NINGBO_WORKER_HOST" 10.77.0.1 80 'worker_2_to_ningbo'
verify_wireguard "$CLUSTER_HONGKONG_HOST" 10.77.0.4 80 'hongkong_to_worker_2'
verify_wireguard "$CLUSTER_NINGBO_WORKER_HOST" 10.77.0.2 80 'worker_2_to_hongkong'
verify_wireguard "$CLUSTER_LOS_ANGELES_HOST" 10.77.0.4 300 'los_angeles_to_worker_2'
verify_wireguard "$CLUSTER_NINGBO_WORKER_HOST" 10.77.0.3 300 'worker_2_to_los_angeles'
fi
}
@@ -2434,8 +2442,8 @@ verify_control_plane_and_recent_logs() {
[[ $error_count == 0 ]] || return 1
fi
done
if [[ -n $CLUSTER_SHENZHEN_HOST && -n $runtime_observation_started_at ]]; then
error_count=$(cluster_ssh "$CLUSTER_SHENZHEN_HOST" \
if [[ -n $CLUSTER_NINGBO_WORKER_HOST && -n $runtime_observation_started_at ]]; then
error_count=$(cluster_ssh "$CLUSTER_NINGBO_WORKER_HOST" \
"journalctl -u k3s-agent --since '$runtime_observation_started_at' --no-pager 2>/dev/null | awk 'BEGIN {IGNORECASE=1} /connection refused/ || /i\/o timeout/ || /node.*not ready/ || /failed to sync/ {count++} END {print count+0}'")
[[ $error_count == 0 ]] || return 1
fi
+8 -7
View File
@@ -15,10 +15,10 @@ load_cluster_env
hosts=("$CLUSTER_NINGBO_HOST" "$CLUSTER_HONGKONG_HOST" "$CLUSTER_LOS_ANGELES_HOST")
wireguard_ips=(10.77.0.1 10.77.0.2 10.77.0.3)
site_labels=(ningbo hongkong losangeles)
if [[ -n $CLUSTER_SHENZHEN_HOST ]]; then
hosts+=("$CLUSTER_SHENZHEN_HOST")
if [[ -n $CLUSTER_NINGBO_WORKER_HOST ]]; then
hosts+=("$CLUSTER_NINGBO_WORKER_HOST")
wireguard_ips+=(10.77.0.4)
site_labels+=(shenzhen)
site_labels+=(ningbo-worker-2)
fi
peer_count=$((${#hosts[@]} - 1))
for source_index in "${!hosts[@]}"; do
@@ -55,7 +55,7 @@ done
echo "wireguard_acceptance=PASS peers=${#hosts[@]} handshakes=$((${#hosts[@]} * peer_count)) packet_loss_lt_1=true"
expected_nodes=3
[[ -z $CLUSTER_SHENZHEN_HOST ]] || expected_nodes=4
[[ -z $CLUSTER_NINGBO_WORKER_HOST ]] || expected_nodes=4
cluster_state=$(cluster_ssh "$CLUSTER_NINGBO_HOST" bash -s -- "$expected_nodes" <<'REMOTE'
set -euo pipefail
expected_nodes=$1
@@ -72,13 +72,14 @@ printf 'nodes=%s\n' "$($kubectl get nodes --no-headers | wc -l)"
REMOTE
)
[[ $cluster_state == nodes="$expected_nodes" ]]
if [[ -n $CLUSTER_SHENZHEN_HOST ]]; then
if [[ -n $CLUSTER_NINGBO_WORKER_HOST ]]; then
worker_state=$(cluster_ssh "$CLUSTER_NINGBO_HOST" 'bash -s' <<'REMOTE'
set -euo pipefail
kubectl='k3s kubectl'
node=easyai-shenzhen
node=easyai-ningbo-worker-2
[[ $($kubectl get node "$node" -o jsonpath='{.status.conditions[?(@.type=="Ready")].status}') == True ]]
[[ $($kubectl get node "$node" -o jsonpath='{.metadata.labels.easyai\.io/site}') == hongkong ]]
[[ $($kubectl get node "$node" -o jsonpath='{.metadata.labels.easyai\.io/site}') == ningbo ]]
[[ $($kubectl get node "$node" -o jsonpath='{.metadata.labels.easyai\.io/worker}') == true ]]
[[ $($kubectl get node "$node" -o jsonpath='{.metadata.labels.easyai\.io/database}') == false ]]
[[ $($kubectl get node "$node" -o json | jq '[.spec.taints[]? | select(.key=="easyai.io/worker-only" and .effect=="NoSchedule")] | length') -eq 1 ]]
[[ $($kubectl get pods -A --field-selector "spec.nodeName=$node" -o json |