如何在JPA（Spring Data JPA）中实现简单的全文搜索？

Question

i'm using JPA 2.1 (Hibernate 4 as impl) and Spring Data JPA 1.9.0. How do i implement full text search?

My scenario is as follows. I have a User entity and on the UI a have a table which display's most of users properties and i want the user to give text box enter there a search term and search in all properties.

I see 2 options to do this:

1. Load all users users from DB and filter them in Java
2. Write a JPQL query with many ORs and LIKE % :searchString %

Option 1 is not good for performance but quite nice to write.

Option 2 is performant beacuse executed on DB side but cumbersome to write.

Right now im suing option 1 because i need to translate boolean to "yes"/"no" and also have a profile enum where i want to search by it's field description and not by actual enum value.

In the User entity i have a method which returns all fields i want to be searched seperated by spaces:

   public String getSearchString(){
      return StringUtils.join(
              Arrays.asList(
                      login, 
                      firstName, 
                      lastName, 
                      email, 
                      active ? "yes" : "no", 
                      profile.getDescription())
              , " ");
   }

The in a service i load all users from DB and filter by this search string:

   @Override
   public List<User> getUsers(final String searchText) {
      final List<User> users = getUsers();
      if(StringUtils.isBlank(searchText)){
         return users;
      }
      CollectionUtils.filter(users, new Predicate<User>() {
         @Override
         public boolean evaluate(User object) {
            return StringUtils.containsIgnoreCase(object.getSearchString(), searchText);
         }
      });
      return users;
   }

On the other side in JPQL i end up with queries like this, which i dont think is the nice'est and easiest way to implement this functionality. Also there is a problem with translatin boolean to "yes" and "no".

@Query("SELECT r FROM User r WHERE "
        + "r.firstname LIKE '%' || :searchString || '%' "
        + "OR r.lastname LIKE '%' || :searchString || '%' "
        + "OR r.login LIKE '%' || :searchString || '%' "
        + "OR r.profile.description LIKE '%' || :searchString || '%' "
        + "OR r.active LIKE '%' || :searchString || '%' "
        + "OR r.email LIKE '%' || :searchString || '%'")
List<User> selectUsers(@Param("searchString")String searchString, Pageable page);

Is there a better solution to this problem?

Answer 1

Solved this by saving the search string on every persist and update to the DB. First created a column for the searchString:

   @Column(name = "SEARCH_STRING", length = 1000)
   private String searchString;

Storage is cheap, overhead on DB is not that big.

Then the saving on update and persist:

   @PreUpdate
   @PrePersist
   void updateSearchString() {
      final String fullSearchString = StringUtils.join(Arrays.asList(
              login,
              firstName,
              lastName,
              email,
              Boolean.TRUE.equals(active) ? "tak" : "nie",
              profile.getDescription()),
              " ");
      this.searchString = StringUtils.substring(fullSearchString, 0, 999);
   }

Then i can have a normal JPQL query with LIKE :

SELECT u FROM User u WHERE u.searchString LIKE '%' || :text || '%'

Or using Query By Example :

  ExampleMatcher matcher = ExampleMatcher.matching().
          withMatcher("searchString", ExampleMatcher.GenericPropertyMatcher.of(ExampleMatcher.StringMatcher.CONTAINING).ignoreCase());

Answer 2

Well, for the most cases, option 1 is not a really an alternative. If your application grow up of registers the memory and performance problems will hit you after some time. I guarantee you.

I cannot see a problem in the option 2 . Is not performant but it's simple to understand.

If performance on database is a problem, you can create a native query to call the native function of your database to do a full text search. Will not be a JPQL query, but for this kind of query you are trying to do this solution can be used.