我正在使用AWS Elasticsearch集群，目前有9个实例，其中包括1个分片和8个副本。当我尝试按比例缩小群集以使其具有2个实例（1个分片和1个副本）时，出现此错误：// g++ -Wall -Wextra -pedantic -std=c++17 -O3 a.cpp
#include <random>
#include <memory>
#include <variant>
#include <chrono>
#include <iostream>
using chronores = std::nano;
static constexpr char const resstr[] = "ns";
namespace helper {
template <template <unsigned> typename T, unsigned X, unsigned UB, typename... Args>
struct mkvar {
using type = typename mkvar<T,X+1,UB,Args...,T<X>>::type;
};
template <template <unsigned> typename T, unsigned UB, typename... Args>
struct mkvar<T,UB,UB,Args...> {
using type = std::variant<Args...,T<UB>>;
};
template <template <unsigned> typename T, unsigned LB, unsigned UB>
using mkvar_t = typename mkvar<T,LB,UB>::type;
template <unsigned X>
struct Num {
static constexpr unsigned value = X;
using inc = Num<X+1>;
};
template <typename NumX, typename NumUB, template <unsigned> typename T, bool use_variant>
struct ctor_Num {
static constexpr auto X = NumX::value;
static constexpr auto UB = NumUB::value;
template <typename Container>
static void run(unsigned x, Container& container) {
if (x == X) {
if constexpr (use_variant) {
container.emplace_back(T<X>());
} else {
container.emplace_back(std::make_unique<T<X>>());
}
} else {
ctor_Num<typename NumX::inc,NumUB,T,use_variant>::run(x,container);
}
}
};
template <typename NumX, template <unsigned> typename T, bool use_variant>
struct ctor_Num<typename NumX::inc,NumX,T,use_variant> {
template <typename Container>
static void run(unsigned, Container&) { }
};
template <unsigned X, unsigned UB, template <unsigned> typename T, bool use_variant, typename Container>
inline void ctor(unsigned x, Container& container) {
return ctor_Num<Num<X>,Num<UB>,T,use_variant>::run(x,container);
}
struct Time {
double& time;
std::chrono::time_point<std::chrono::steady_clock> start;
Time(double& time) : time(time) {
start = std::chrono::steady_clock::now();
}
~Time() {
auto const finish = std::chrono::steady_clock::now();
time += std::chrono::duration<double,chronores>(finish-start).count();
}
};
}
template <unsigned LB, unsigned UB>
struct measure {
struct A {
virtual unsigned f() const noexcept = 0;
virtual ~A() noexcept {}
};
template <unsigned X>
struct B : A {
virtual unsigned f() const noexcept override { return X; }
};
template <unsigned X>
struct C {
unsigned f() const noexcept { return X; }
};
static void main(std::size_t const N, std::size_t const R = 10, bool warmup = false) {
if (!warmup) main(N,1,true);
using namespace helper;
std::vector<std::unique_ptr<A>> bs;
bs.reserve(N);
std::vector<mkvar_t<C,LB,UB>> cs;
cs.reserve(N);
std::uniform_int_distribution<unsigned> distr(LB,UB);
double time_ctor_virtual = 0;
double time_ctor_variant = 0;
double time_call_virtual = 0;
double time_call_variant = 0;
unsigned volatile sum = 0;
std::mt19937 mt(42); mt.discard(100);
for (std::size_t r = 0; r < R; ++r) {
bs.clear();
cs.clear();
{
Time scope(time_ctor_virtual);
for (std::size_t i = 0; i < N; ++i) {
bs.emplace_back(std::make_unique<B<UB>>());
}
}
{
Time scope(time_ctor_variant);
for (std::size_t i = 0; i < N; ++i) {
cs.emplace_back(C<UB>());
}
}
bs.clear();
cs.clear();
for (std::size_t i = 0; i < N; ++i) {
auto const rn = distr(mt);
// effectively calls bs.emplace_back(std::make_unique<B<rn>>())
ctor<LB,UB,B,false>(rn,bs);
// effectively calls cs.emplace_back(C<rn>())
ctor<LB,UB,C,true >(rn,cs);
}
{
Time scope(time_call_variant);
for (std::size_t i = 0; i < N; ++i) {
sum += std::visit([](auto const& c) { return c.f(); },cs[i]);
}
}
{
Time scope(time_call_virtual);
for (std::size_t i = 0; i < N; ++i) {
sum += bs[i]->f();
}
}
}
(void)sum;
if (!warmup) {
std::cout << "[" << LB << "," << UB << "] time ctor virtual: " << (time_ctor_virtual/N/R) << " " << resstr << "\n";
std::cout << "[" << LB << "," << UB << "] time ctor variant: " << (time_ctor_variant/N/R) << " " << resstr << "\n";
std::cout << "[" << LB << "," << UB << "] time call virtual: " << (time_call_virtual/N/R) << " " << resstr << "\n";
std::cout << "[" << LB << "," << UB << "] time call variant: " << (time_call_variant/N/R) << " " << resstr << "\n";
}
}
};
int main() {
static constexpr std::size_t N = 400000;
measure<0,0>::main(N);
std::cout << "\n";
measure<0,1>::main(N);
std::cout << "\n";
measure<0,7>::main(N);
std::cout << "\n";
measure<0,31>::main(N);
std::cout << "\n";
measure<0,127>::main(N);
std::cout << "\n";
}
。
我希望它从群集中删除多余的副本。为什么它不允许我这样做？是否尝试将多余的副本合并为一个？解决此问题的正确方法是什么？
答案 0 :(得分：1)
Elasticsearch必须能够抵抗单个节点的故障。通过在quorum个节点接受集群状态更新后认为成功，集群可以实现这种弹性。仲裁是群集中符合主机资格的节点的精心选择的子集。
必须仔细选择仲裁人数，以使集群无法选择两个独立的主机，这些主机的决策不一致，最终导致数据丢失。知道more..
缩小之前的准备工作
将索引复制因子降低到1，以节省空间并在缩放过程中加快分片重定位，因为需要创建和移动的分片较少。而且，这可以节省重复数据中的大量空间。
curl -X PUT "localhost:9200/twitter/_settings?pretty" -H 'Content-Type: application/json' -d'
{
"index" : {
"number_of_replicas" : 1
}
}
'
在开始缩减之前，请先优雅地重新平衡集群。
集群必须保持绿色并保持健康，请检查shards和status
健康
curl -X GET "localhost:9200/_cluster/health?pretty"
预期输出
{
"cluster_name" : "\"es-data-cluster\"",
"status" : "green",
"timed_out" : false,
"number_of_nodes" : 1,
"number_of_data_nodes" : 1,
"active_primary_shards" : 0,
"active_shards" : 0,
"relocating_shards" : 0,
"initializing_shards" : 0,
"unassigned_shards" : 0,
"delayed_unassigned_shards" : 0,
"number_of_pending_tasks" : 0,
"number_of_in_flight_fetch" : 0,
"task_max_waiting_in_queue_millis" : 0,
"active_shards_percent_as_number" : 100.0
}
碎片
curl -X GET "localhost:9200/_cat/shards"
预期输出
twitter 2 p STARTED 0 0b 172.18.0.2 es-node
twitter 1 p STARTED 0 0b 172.18.0.2 es-node
twitter 0 p STARTED 0 230b 172.18.0.2 es-node
当群集状态为green且所有分片均为STARTED时，最好按比例缩小。
缩小比例的步骤
删除一个数据节点-群集将进入黄色状态。 现在观察
STARTED和UNASSIGNED分片如果日志中显示Marking shards as stale，则表示该分片不再可用，将被删除。然后，elasticsearch的内置功能开始重新平衡碎片。
curl -X GET "localhost:9200/_cluster/allocation/explain?pretty"
此命令将详细说明集群中的分片分配。
等待绿色-则集群已复制丢失的碎片。
集群运行状况为
red，因此至少有一个未分配的主分片。您需要关注未分配的群集。
答案 1 :(得分：0)
您的设置不会更改，因为您按比例缩小了集群。由于您已指定1个主碎片和8个副本，因此elasticsearch会尝试将这些副本移动到其他节点。您需要先更新settings（1个主分片和1个副本），然后按比例缩小集群。
PUT /INDEX_NAME_HERE/_settings
{
"index" : {
"number_of_replicas" : 1
}
}