From 63dd6adb5473e002889485f4afecf47a9d0d1094 Mon Sep 17 00:00:00 2001
From: Ye Cao <caoye.cao@alibaba-inc.com>
Date: Mon, 18 Dec 2023 11:11:34 +0800
Subject: [PATCH] Delete the retry logic while launching metadata service and
 reset the etcd client while probing (#1649)

Fixes #1648

Signed-off-by: Ye Cao <caoye.cao@alibaba-inc.com>
---
 src/server/services/etcd_meta_service.cc |  3 ++
 src/server/util/etcd_launcher.cc         | 36 ++++++++----------------
 src/server/util/redis_launcher.cc        | 36 ++++++++----------------
 3 files changed, 25 insertions(+), 50 deletions(-)

diff --git a/src/server/services/etcd_meta_service.cc b/src/server/services/etcd_meta_service.cc
index dd81a9dabd..161ed9fca6 100644
--- a/src/server/services/etcd_meta_service.cc
+++ b/src/server/services/etcd_meta_service.cc
@@ -380,6 +380,9 @@ void EtcdMetaService::retryDaeminWatch(
 }
 
 Status EtcdMetaService::probe() {
+  std::string const& etcd_endpoint =
+      etcd_spec_["etcd_endpoint"].get_ref<std::string const&>();
+  etcd_.reset(new etcd::Client(etcd_endpoint));
   if (EtcdLauncher::probeEtcdServer(etcd_, prefix_)) {
     return Status::OK();
   } else {
diff --git a/src/server/util/etcd_launcher.cc b/src/server/util/etcd_launcher.cc
index 71d966df68..f377c17f17 100644
--- a/src/server/util/etcd_launcher.cc
+++ b/src/server/util/etcd_launcher.cc
@@ -111,16 +111,6 @@ Status EtcdLauncher::LaunchEtcdServer(
     return Status::OK();
   }
 
-  LOG(INFO) << "Starting the etcd server";
-
-  // resolve etcd binary
-  std::string etcd_cmd = etcd_spec_["etcd_cmd"].get_ref<std::string const&>();
-  if (etcd_cmd.empty()) {
-    setenv("LC_ALL", "C", 1);  // makes boost's path works as expected.
-    etcd_cmd = boost::process::search_path("etcd").string();
-  }
-  LOG(INFO) << "Found etcd at: " << etcd_cmd;
-
   RETURN_ON_ERROR(initHostInfo());
   bool try_launch = false;
   if (local_hostnames_.find(endpoint_host_) != local_hostnames_.end() ||
@@ -129,22 +119,18 @@ Status EtcdLauncher::LaunchEtcdServer(
   }
 
   if (!try_launch) {
-    LOG(INFO) << "Will not launch an etcd instance.";
-    int retries = 0;
-    while (retries < max_probe_retries) {
-      if (probeEtcdServer(etcd_client, sync_lock)) {
-        break;
-      }
-      retries += 1;
-      sleep(1);
-    }
-    if (retries >= max_probe_retries) {
-      return Status::EtcdError(
-          "Etcd has been launched but failed to connect to it");
-    } else {
-      return Status::OK();
-    }
+    return Status::OK();
+  }
+
+  LOG(INFO) << "Starting the etcd server";
+
+  // resolve etcd binary
+  std::string etcd_cmd = etcd_spec_["etcd_cmd"].get_ref<std::string const&>();
+  if (etcd_cmd.empty()) {
+    setenv("LC_ALL", "C", 1);  // makes boost's path works as expected.
+    etcd_cmd = boost::process::search_path("etcd").string();
   }
+  LOG(INFO) << "Found etcd at: " << etcd_cmd;
 
   std::string host_to_advertise;
   if (host_to_advertise.empty()) {
diff --git a/src/server/util/redis_launcher.cc b/src/server/util/redis_launcher.cc
index a3f542cf8b..119a6b9e93 100644
--- a/src/server/util/redis_launcher.cc
+++ b/src/server/util/redis_launcher.cc
@@ -73,6 +73,17 @@ Status RedisLauncher::LaunchRedisServer(
     return Status::OK();
   }
 
+  RETURN_ON_ERROR(initHostInfo());
+  bool try_launch = false;
+  if (local_hostnames_.find(endpoint_host_) != local_hostnames_.end() ||
+      local_ip_addresses_.find(endpoint_host_) != local_ip_addresses_.end()) {
+    try_launch = true;
+  }
+
+  if (!try_launch) {
+    return Status::OK();
+  }
+
   LOG(INFO) << "Starting the redis server";
 
   // resolve redis binary
@@ -84,31 +95,6 @@ Status RedisLauncher::LaunchRedisServer(
   }
   LOG(INFO) << "Found redis at: " << redis_cmd;
 
-  RETURN_ON_ERROR(initHostInfo());
-  bool try_launch = false;
-  if (local_hostnames_.find(endpoint_host_) != local_hostnames_.end() ||
-      local_ip_addresses_.find(endpoint_host_) != local_ip_addresses_.end()) {
-    try_launch = true;
-  }
-
-  if (!try_launch) {
-    LOG(INFO) << "Will not launch a redis instance.";
-    int retries = 0;
-    while (retries < max_probe_retries) {
-      if (probeRedisServer(redis_client, syncredis_client, watch_client)) {
-        break;
-      }
-      retries += 1;
-      sleep(1);
-    }
-    if (retries >= max_probe_retries) {
-      return Status::RedisError(
-          "Redis has been launched but failed to connect to it");
-    } else {
-      return Status::OK();
-    }
-  }
-
   std::vector<std::string> args;
   args.emplace_back("--port");
   args.emplace_back(std::to_string(endpoint_port_));