Spring Boot neo4j 性能

如何解决Spring Boot neo4j 性能

我们正在评估我们公司使用 neo4j 的可能性，我们使用 Java (>1.8)、Spring Boot (2.4.2) 和 SpringData-neo4j (2.4.2) 进行简单的查询和一些性能测试，neo4j 是使用官方 docker 镜像在 docker 容器中本地运行。

这是我们的设置：

json 结构

{
    "id": null,"addresses": [
      {
        "id": null,"city": "South Thresa","country": "Burundi","houseNumber": "861","street": "Auer Lake","zipCode": "03412"
      }
    ],"bankAccounts": [
      {
        "id": null,"bankAccountNumber": null,"bankName": null,"bic": "QVWPKO5H","blz": null,"countryCode": null,"divergentAccountHolder": null,"iban": "GB87DPZP84667918216002","sepaMandates": [
          {
            "id": null,"bankAccount": null,"signatureCity": "Lednerbury","signatureDate": "06.10.1984"
          }
        ]
      }
    ],"contacts": [
      {
        "id": null,"email": "tonisha.mitchell@gmail.com","faxNumber": "087-835-0885 x859","mobilePhoneNumber": "1-615-772-6922","phoneNumber": "(082) 785-8108 x6736"
      }
    ],"birthDate": "07.01.1966","firstName": "Daniel","lastName": "Hartmann","salutation": "Mrs.","title": "Lead Branding Orchestrator"
  }

实体：

@Node
public class Partner {
    @Id
    @GeneratedValue
    public Long id;
    @Relationship(type = "ADDRESS",direction = Relationship.Direction.INCOMING)
    public List<Address> addresses;
    @Relationship(type = "BANKACCOUNT",direction = Relationship.Direction.INCOMING)
    public List<BankAccount> bankAccounts;
    @Relationship(type = "CONTACTS",direction = Relationship.Direction.INCOMING)
    public List<Contact> contacts;
    @Property
    public String birthDate;
    @Property
    public String firstName;
    @Property
    public String lastName;
    @Property
    public String salutation;
    @Property
    public String title;
}

@Node
public class Contact{
    @Id
    @GeneratedValue
    public Long id;
    @Property
    public String email;
    @Property
    public String faxNumber;
    @Property
    public String mobilePhoneNumber;
    @Property
    public String phoneNumber;

@Node
public class BankAccount{
    @Id
    @GeneratedValue
    public Long id;
    @Property
    public String bankAccountNumber;
    @Property
    public String bankName;
    @Property
    public String bic;
    @Property
    public String blz;
    @Property
    public String countryCode;
    @Property
    public String divergentAccountHolder;
    @Property
    public String iban;
    @Relationship(type = "SEPA_MANDAT")
    public List<SepaMandate> sepaMandates;

@Node
public class Address{
    @Id
    @GeneratedValue
    public Long id;
    @Property
    public String city;
    @Property
    public String country;
    @Property
    public String houseNumber;
    @Property
    public String street;
    @Property
    public String zipCode;

@Node
public class SepaMandate{
    @Id
    @GeneratedValue
    public Long id;
    @Property
    public BankAccount bankAccount;
    @Property
    public String signatureCity;
    @Property
    public String signatureDate;

存储库：

public interface PartnerRepository extends Neo4jRepository<Partner,Long> {
  List<Partner> findByLastName(String name);
}

控制器有一个端点来触发大规模插入并停止时间

    @PutMapping("/partner/initialInsert")
    public void createPartner() {
        List<Partner> partners = mockData();
        StopWatch stopWatch = new StopWatch();
        stopWatch.start();
        partnerRepository.saveAll(partners);
        stopWatch.stop();
        System.err.println("Created: 100000 partnerEntries  ---->>> Duration ------------> " + stopWatch.toString());
    }

还有一些简单的查询/更新和插入端点

    @GetMapping("/partner")
    public List<Partner> getPartnerByLastName(@RequestParam(value = "lastname") String lastname) {
        StopWatch stopWatch = new StopWatch();
        stopWatch.start();
        List<Partner> partners = partnerRepository.findByLastName(lastname);
        stopWatch.stop();
        System.err.println("Selecting partners by lastname: " +lastname + " took: " + stopWatch.toString() + " and returns: " + partners.size() + " partners");
        return partners;
    }

    @PutMapping("/partner")
    public void createPartner(@RequestBody Partner partner) {
        StopWatch stopWatch = new StopWatch();
        stopWatch.start();
        partnerRepository.save(partner);
        stopWatch.stop();
        System.err.println("Updating/Inserting partner took: " + stopWatch.toString());
    }

    @PostMapping("/partner/bulk")
    public void bulkUpdatePartner(@RequestBody List<Partner> partners) {
        StopWatch stopWatch = new StopWatch();
        stopWatch.start();
        partnerRepository.saveAll(partners);
        stopWatch.stop();
        System.err.println("Updating/Inserting partner took: " + stopWatch.toString());
    }

以下是我们的测试结果：

通过批量 saveAll() 插入随机数据。

已创建：1000 个合作伙伴条目 ---->> 持续时间 ------> 00:00:29.775
已创建：10000 个合作伙伴条目 ---->>> 持续时间 ------> 00:04:40.259
已创建：100000 个合作伙伴条目 ---->> 持续时间 ------> 00:53:48.862

多次通过姓氏选择数据

按姓氏选择合作伙伴：Borer 取：00:00:00.404 返回：257 个合作伙伴
按姓氏选择合作伙伴：Borer 取：00:00:00.219 返回：257 个合作伙伴
按姓氏选择合作伙伴：Borer 取：00:00:00.173 返回：257 个合作伙伴
按姓氏选择合作伙伴：Borer 取：00:00:00.239 返回：257 个合作伙伴
按姓氏选择合作伙伴：Borer 取：00:00:00.194 返回：257 个合作伙伴
按姓氏选择合作伙伴：Borer 获取：00:00:00.266 并返回：257 个合作伙伴

更新/插入单个条目：

更新合作伙伴时间：00:00:00.297
插入合作伙伴花了：00:00:00.124

使用 jpa 更新批量 256 个合作伙伴：

更新合作伙伴时间：00:00:08.931

使用不带 jpa 的会话对象更新批量 256 合作伙伴：

更新合作伙伴时间：00:00:00.090

我们真的很好奇使用 jpa 函数 saveAll() 进行批量插入需要 50 分钟以上的时间来插入 100.000 个对象，以及通过 jpa 和会话对象更新 256 个条目之间的差异。现在的问题是，我们做错了什么，我们尝试使用没有任何特殊情况的简单设置，spring-data 集成有问题还是你们有类似的问题？我很期待你在 neo4j 方面的体验 :-)

问候丹:)

P.S：这就是我们更新会话对象的方式

try (Session session = driver.session()) {
            session.run("MATCH (a:Partner {lastName: 'Borer'}) SET a.firstName = '" + firstname + "'");
}

Spring Boot neo4j 性能

如何解决Spring Boot neo4j 性能

相关推荐