在分布式緩存l2上使用infinispan和hibernate進行連接鎖定

Question

我將infinispan用作分布式休眠緩存L2，在AWS上使用jgroups配置。 但是，在以下情況下，我在重負載下面臨問題：

• 最初有2個EC2實例可用
• 負載急劇增加
• 負載均衡啟動4個EC2實例
• 負載部分減少
• 負載平衡減少2個EC2實例
• 剩余的所有實例開始保持數據庫連接（Hikari池），直到沒有可用的連接為止，隨后的所有請求由於超時等待空閑實例而返回錯誤。

其余實例嘗試與舊實例通信，但沒有得到響應，在等待響應的同時保持連接。

所有實體都使用READ_WRITE策略。

Infinispan配置： org / infinispan / hibernate / cache / commons / builder / infinispan-configs.xml

region.factory_class： org.infinispan.hibernate.cache.commons.InfinispanRegionFactory

以下Jgroups配置已從 org / infinispan / infinispan-core / 9.2.0.Final / infinispan-core-9.2.0.Final.jar / default-configs / default-jgroups-tcp.xml中進行了編輯

    <config xmlns="urn:org:jgroups"
        xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
        xsi:schemaLocation="urn:org:jgroups http://www.jgroups.org/schema/jgroups-4.0.xsd">
   <TCP bind_port="7800"
        enable_diagnostics="false"
        thread_naming_pattern="pl"
        send_buf_size="640k"
        sock_conn_timeout="300"
        bundler_type="no-bundler"

        thread_pool.min_threads="${jgroups.thread_pool.min_threads:50}"
        thread_pool.max_threads="${jgroups.thread_pool.max_threads:500}"
        thread_pool.keep_alive_time="30000"
   />
   <AWS_ELB_PING
           region="sa-east-1"
           load_balancers_names="elb-name"
   />
   <MERGE3 min_interval="5000"
           max_interval="30000"
   />
   <FD_SOCK />
   <FD_ALL timeout="9000"
           interval="3000"
           timeout_check_interval="1000"
   />
   <VERIFY_SUSPECT timeout="5000" />
   <pbcast.NAKACK2 use_mcast_xmit="false"
                   xmit_interval="100"
                   xmit_table_num_rows="50"
                   xmit_table_msgs_per_row="1024"
                   xmit_table_max_compaction_time="30000"
                   resend_last_seqno="true"
   />
   <UNICAST3 xmit_interval="100"
             xmit_table_num_rows="50"
             xmit_table_msgs_per_row="1024"
             xmit_table_max_compaction_time="30000"
             conn_expiry_timeout="0"
   />
   <pbcast.STABLE stability_delay="500"
                  desired_avg_gossip="5000"
                  max_bytes="1M"
   />
   <pbcast.GMS print_local_addr="false"
               install_view_locally_first="true"
               join_timeout="${jgroups.join_timeout:5000}"
   />
   <MFC max_credits="2m"
        min_threshold="0.40"
   />
   <FRAG3/>
</config>

AWS_ELB_PING：此類是Discovery類的實現，在其中使用AWS ELB api發現所有可用的ip 。

我從下面的代碼中刪除了日志和一些樣板代碼：

public class AWS_ELB_PING extends Discovery {

    private static final String LIST_ELEMENT_SEPARATOR = ",";

    static {
        ClassConfigurator.addProtocol((short) 790, AWS_ELB_PING.class); // id must be unique
    }

    private String region;

    private String load_balancers_names;

    private int bind_port = 7800;

    private AmazonElasticLoadBalancing amazonELBClient;
    private AmazonEC2 amazonEC2Client;

    private List<String> getLoadBalancersNamesList() {
        return Arrays.asList(Optional.ofNullable(load_balancers_names).orElse("").split(LIST_ELEMENT_SEPARATOR));
    }

    @Override
    public void init() throws Exception {
        super.init();
        DefaultAWSCredentialsProviderChain awsCredentialsProviderChain = DefaultAWSCredentialsProviderChain.getInstance();
        amazonELBClient = AmazonElasticLoadBalancingClientBuilder.standard()
                .withRegion(region)
                .withCredentials(awsCredentialsProviderChain)
                .build();
        amazonEC2Client = AmazonEC2ClientBuilder.standard()
                .withRegion(region)
                .withCredentials(awsCredentialsProviderChain)
                .build();
    }

    @Override
    public void discoveryRequestReceived(final Address sender, final String logical_name,
                                         final PhysicalAddress physical_addr) {
        super.discoveryRequestReceived(sender, logical_name, physical_addr);
    }

    @Override
    public void findMembers(final List<Address> members, final boolean initialDiscovery, final Responses responses) {
        PhysicalAddress physicalAddress = null;
        PingData data = null;
        if (!use_ip_addrs || !initialDiscovery) {
            physicalAddress = (PhysicalAddress) super.down(new Event(Event.GET_PHYSICAL_ADDRESS, local_addr));
            data = new PingData(local_addr, false, NameCache.get(local_addr), physicalAddress);
            if (members != null && members.size() <= max_members_in_discovery_request) {
                data.mbrs(members);
            }
        }

        sendDiscoveryRequests(physicalAddress, data, initialDiscovery, getLoadBalancersInstances());
    }

    private Set<Instance> getLoadBalancersInstances() {
        final List<String> loadBalancerNames = getLoadBalancersNamesList();
        final List<LoadBalancerDescription> loadBalancerDescriptions = amazonELBClient
                .describeLoadBalancers(new DescribeLoadBalancersRequest().withLoadBalancerNames(loadBalancerNames))
                .getLoadBalancerDescriptions();

        checkLoadBalancersExists(loadBalancerNames, loadBalancerDescriptions);

        final List<String> instanceIds = loadBalancerDescriptions.stream()
                .flatMap(loadBalancer -> loadBalancer.getInstances().stream())
                .map(instance -> instance.getInstanceId())
                .collect(toList());

        return amazonEC2Client.describeInstances(new DescribeInstancesRequest().withInstanceIds(instanceIds))
                    .getReservations()
                    .stream()
                    .map(Reservation::getInstances)
                    .flatMap(List::stream)
                    .collect(Collectors.toSet());
    }

    private void checkLoadBalancersExists(final List<String> loadBalancerNames,
                                          final List<LoadBalancerDescription> loadBalancerDescriptions) {
        final Set<String> difference = Sets.difference(new HashSet<>(loadBalancerNames),
            loadBalancerDescriptions
                    .stream()
                    .map(LoadBalancerDescription::getLoadBalancerName)
                    .collect(Collectors.toSet()));
    }

    private PhysicalAddress toPhysicalAddress(final Instance instance) {
        try {
            return new IpAddress(instance.getPrivateIpAddress(), bind_port);
        } catch (final Exception e) {
            throw new RuntimeException(e);
        }
    }

    private void sendDiscoveryRequests(@Nullable final PhysicalAddress localAddress, @Nullable final PingData data,
                                       final boolean initialDiscovery, final Set<Instance> instances) {
        final PingHeader header = new PingHeader(PingHeader.GET_MBRS_REQ)
                    .clusterName(cluster_name)
                    .initialDiscovery(initialDiscovery);

        instances.stream()
                .map(this::toPhysicalAddress)
                .filter(physicalAddress -> !physicalAddress.equals(localAddress))
                .forEach(physicalAddress -> sendDiscoveryRequest(data, header, physicalAddress));
    }

    private void sendDiscoveryRequest(@Nullable final PingData data, final PingHeader header,
                      final PhysicalAddress destinationAddress) {
        final Message message = new Message(destinationAddress)
                .setFlag(Message.Flag.INTERNAL, Message.Flag.DONT_BUNDLE, Message.Flag.OOB)
                .putHeader(this.id, header);
        if (data != null) {
            message.setBuffer(marshal(data));
        }

        if (async_discovery_use_separate_thread_per_request) {
            timer.execute(() -> sendDiscoveryRequest(message), sends_can_block);
        } else {
            sendDiscoveryRequest(message);
        }
    }

    protected void sendDiscoveryRequest(final Message message) {
        try {
            super.down(message);
        } catch (final Throwable t) {
        }
    }

    @Override
    public boolean isDynamic() {
        return true;
    }

    @Override
    public void stop() {
        try {
            if (amazonEC2Client != null) {
                amazonEC2Client.shutdown();
            }
            if (amazonELBClient != null) {
                amazonELBClient.shutdown();
            }
        } catch (final Exception e) {
        } finally {
            super.stop();
        }
    }
}

有人已經遇到這種問題了嗎？

Answer 1

什么是你的問題; 您已經描述了您的設置，但沒有遇到問題...您是否對AWS_ELP_PING有參考？

Answer 2

盡管您可能想從現有的發現代碼（例如TCPPING或FILE_PING甚至是NATIVE_S3_PING）進行擴展，但代碼對我來說似乎還不錯。

“池耗盡”是什么意思？ 這是維護連接池的AWS客戶端嗎？ 還是說TCP中的線程池？ 后者具有bindler_type=no-bundler ; 嘗試刪除它（然后將使用transfer-queue-bundler ，它將創建消息批處理，而不是一個接一個地發送消息）。

如果您有一個疲憊的TCP線程池，那么獲取堆棧跟蹤信息，看看現有線程被阻塞了，將會很有意思。