feat: etcd cluster single node failure APISIX startup failure

apisix/cli/etcd.lua

@@ -33,6 +33,7 @@ local str_format = string.format
 local str_sub = string.sub
 local table_concat = table.concat
 local table_insert = table.insert
+local io_stderr = io.stderr
 
 local _M = {}
 
@@ -218,24 +219,22 @@ function _M.init(env, args)
 
             local cluster_version = body["etcdcluster"]
             if compare_semantic_version(cluster_version, env.min_etcd_version) then
-                util.die("etcd cluster version ", cluster_version,
-                        " is less than the required version ",
-                        env.min_etcd_version,
-                        ", please upgrade your etcd cluster\n")
+                util.die("etcd cluster version ", cluster_version, " is less than the required version ",
+                        env.min_etcd_version, ", please upgrade your etcd cluster\n")
             end
 
             table_insert(etcd_healthy_hosts, host)
         else
-            print(str_format("request etcd endpoint \'%s\' error, %s\n", version_url, err))
+            io_stderr:write(str_format("request etcd endpoint \'%s\' error, %s\n", version_url, err))
         end
     end
 
     if #etcd_healthy_hosts <= 0 then
         util.die("all etcd nodes are unavailable\n")
     end
 
-    if host_count >= 2 and #etcd_healthy_hosts < 2 then
-        util.die("etcd cluster must have two or more healthy nodes\n")
+    if host_count >= 2 and (#etcd_healthy_hosts / host_count * 100) < 50 then
+        util.die("the etcd cluster needs at least 50% and above healthy nodes\n")
     end
 
     local etcd_ok = false

rockspec/apisix-master-0.rockspec

@@ -34,7 +34,7 @@ dependencies = {
     "lua-resty-ctxdump = 0.1-0",
     "lua-resty-dns-client = 5.2.0",
     "lua-resty-template = 2.0",
-    "lua-resty-etcd = 1.5.5",
+    "lua-resty-etcd = 1.6.0",
     "api7-lua-resty-http = 0.2.0",
     "lua-resty-balancer = 0.02rc5",
     "lua-resty-ngxvar = 0.5.2",

t/cli/test_etcd_ha.sh

@@ -31,7 +31,7 @@ etcd:
 
 docker-compose -f ./t/cli/docker-compose-etcd-cluster.yaml up -d
 
-# case 1: stop one etcd nodes (result: start successful)
+# case 1: stop one etcd node (result: start successful)
 docker stop ${ETCD_NAME_0}
 
 out=$(make init 2>&1)
@@ -49,7 +49,7 @@ out=$(make init 2>&1)
 if echo "$out" | grep "23791" | grep "connection refused"; then
     echo "passed: APISIX failed to start, etcd cluster must have two or more healthy nodes"
 else
-    echo "failed: etcd has stopped two nodes, APISIX should fail to start"
+    echo "failed: two etcd nodes have been stopped, APISIX should fail to start"
     exit 1
 fi