[NHibernate]大量存取時要注意的地方

2012-05-03

在替系統做效能調整時，發現NHibernate有幾個小地方若是沒有注意到的話，會讓效能越來越慢

情境一 : 有使用到OpensessionInView或是SessionScope去延長NHibernate Session存活的時間時，在這個時間下大量存取資料

   1:              using (new SessionScope())

   2:              {

   3:                  // 假設在data裡有一萬筆Person

   4:                  foreach (IPerson person in data)

   5:                  {

   6:                      //然後把這一萬筆Person存進DB裡

   7:                      PersonService.Create(person);

   8:                  }

   9:              }

以這個單純的情境來看，由於session一直都沒被釋放掉，所以每存一筆Person，Session就要去記住它，

當筆數越多session要記的Person就越多，所以效能也就會隨著筆數的增加而越來差。

若改成這樣:

   1:              foreach (IPerson person in data)

   2:              {

   3:                  using (new SessionScope())

   4:                  {

   5:                      //然後把這一萬筆Person存進DB裡

   6:                      PersonService.Create(person);

   7:                  }

   8:              }

每存完一個人就釋放掉的話，就可以穩定效能了。(範例過於簡單請不要太認真)

情境二 : 在複雜的關聯下，存多次不如存一次

   1:                  IPerson person = PersonService.MakeParty();

   2:                  IPersonAccount personAccount = this.PersonAccountService.MakePartyRole(person);

   3:                  // 建立person 跟 personAccount

   4:                  PersonService.CreateOrUpdate(person);

5:

   6:                  IMobilePhone phone = RheaFactory.DomainFactory.MakeDomain<IMobilePhone>();

   7:                  person.ContactMechs.Add(phone);

   8:                  phone.Party = person;

   9:                  // 建立person 跟 mobilePhone

  10:                  ContactMechService.CreateOrUpdate(phone);

11:

  12:                  IEmail email = RheaFactory.DomainFactory.MakeDomain<IEmail>();

  13:                  person.ContactMechs.Add(email);

  14:                  email.Party = person;

  15:                  // 建立person 跟 email

  16:                  ContactMechService.CreateOrUpdate(email);

17:

  18:                  IPostalAddress addr = RheaFactory.DomainFactory.MakeDomain<IPostalAddress>();

  19:                  person.ContactMechs.Add(addr);

  20:                  addr.Party = person;

  21:                  // 建立person 跟 postalAddress

  22:                  ContactMechService.CreateOrUpdate(addr);

23:

  24:                  IWorkingCareer work = RheaFactory.DomainFactory.MakeDomain<IWorkingCareer>();

  25:                  personAccount.WorkingCareers.Add(work);

  26:                  work.PersonAccount = personAccount;

27:

  28:                  ICustomerAnniversary anniversary = RheaFactory.DomainFactory.MakeDomain<ICustomerAnniversary>();

  29:                  personAccount.CustomerAnniversaries.Add(anniversary);

  30:                  // 建立personAccount 跟 WorkingCareer & anniversary

  31:                  PersonAccountService.CreateOrUpdate(personAccount);

這個範例程式每增加一組關聯就會存到DB一次，一般用NHibernate一定不會這樣用這是挺誇張的寫法，

不過或許會有中間存一次到兩次的情況，以前我一直覺得其實Persistence回DB的動作一次跟多次應該沒什麼差別，

但這中間其實是會有些微的差異，另外隨著Domain Model物件關聯的設計，也會對這種多次存的動作帶來效能上的差異

以這次調整的心得是，如在做複雜的關聯時，Persistence 的次數越少越好，最好能所有物件都設好後Persistence一次就行

如下:

   1:  IPerson person = PersonService.MakeParty();

   2:  IPersonAccount personAccount = this.PersonAccountService.MakePartyRole(person);

3:

   4:  IMobilePhone phone = RheaFactory.DomainFactory.MakeDomain<IMobilePhone>();

   5:  person.ContactMechs.Add(phone);

   6:  phone.Party = person;

7:

   8:  IEmail email = RheaFactory.DomainFactory.MakeDomain<IEmail>();

   9:  person.ContactMechs.Add(email);

  10:  email.Party = person;

11:

  12:  IPostalAddress addr = RheaFactory.DomainFactory.MakeDomain<IPostalAddress>();

  13:  person.ContactMechs.Add(addr);

  14:  addr.Party = person;

15:

  16:  IWorkingCareer work = RheaFactory.DomainFactory.MakeDomain<IWorkingCareer>();

  17:  personAccount.WorkingCareers.Add(work);

  18:  work.PersonAccount = personAccount;

19:

  20:  ICustomerAnniversary anniversary = RheaFactory.DomainFactory.MakeDomain<ICustomerAnniversary>();

  21:  personAccount.CustomerAnniversaries.Add(anniversary);

22:

  23:  // 上面都在做關聯，最後在用主要的那顆物件去Persistence就行

  24:  PersonService.CreateOrUpdate(person);

或許在單檔維護的情況下看不出效能怎樣，

但在大量存取這種分毫都要計較的情況下，這可以幫忙帶來些許的效能進步