How to query data for Primefaces dataTable using lazy loading and pagination

In case of very large resulting lists, the Java-side counting and the sublisting operations can be dangerous for the memory usage and consequently also on the performance side.

Instead, I usually go with the following approach: use 2 queries, one for counting the filtered resultSet (I let the db do the count), and another one for retrieving the paginated resultSet (I let the db extract the sublist). I have never experienced significant delays, even with tables containing millions of rows.

Follows a concrete example with sorting and filtering. All the code uses JPA standard (no Hibernate or Spring custom features) The CriteriaQuery approach is particularly indicated in such situations.

MyBean class

@ManagedBean
@ViewScoped
public class MyBean {
    @EJB
    private MyObjFacade myObjFacade;
    private LazyDataModel<MyObjType> model;        // getter and setter

    @PostConstruct
    public void init() {
        model = new LazyDataModel<MyObjType> () {

            @Override
            public List<MyObjType> load(int first, int pageSize, String sortField, SortOrder sortOrder, Map<String, String> filters) {
                model.setRowCount(myObjFacade.count(filters));
                return myObjFacade.getResultList(first, pageSize, sortField, sortOrder, filters);
            }
        };
        model.setRowCount(myObjFacade.count(new HashMap<String, String> ()));
    }
}

MyObjFacade class

@Stateless
public class MyObjFacade {
    @PersistenceContext
    private EntityManager em;
    @EJB
    private MyObjFacade myObjFacade;

    private Predicate getFilterCondition(CriteriaBuilder cb, Root<MyObjType> myObj, Map<String, String> filters) {
        Predicate filterCondition = cb.conjunction();
        String wildCard = "%";
        for (Map.Entry<String, String> filter : filters.entrySet()) {
            String value = wildCard + filter.getValue() + wildCard;
            if (!filter.getValue().equals("")) {
                javax.persistence.criteria.Path<String> path = myObj.get(filter.getKey());
                filterCondition = cb.and(filterCondition, cb.like(path, value));
            }
        }
        return filterCondition;
    }

    public int count(Map<String, String> filters) {
        CriteriaBuilder cb = getEntityManager().getCriteriaBuilder();
        CriteriaQuery<Long> cq = cb.createQuery(Long.class);
        Root<MyObjType> myObj = cq.from(MyObjType.class);
        cq.where(myObjFacade.getFilterCondition(cb, myObj, filters));
        cq.select(cb.count(myObj));
        return em.createQuery(cq).getSingleResult().intValue();
    }

    public List<MyObjType> getResultList(int first, int pageSize, String sortField, SortOrder sortOrder, Map<String, String> filters) {
        CriteriaBuilder cb = getEntityManager().getCriteriaBuilder();
        CriteriaQuery<MyObjType> cq = cb.createQuery(MyObjType.class);
        Root<MyObjType> myObj = cq.from(MyObjType.class);
        cq.where(myObjFacade.getFilterCondition(cb, myObj, filters));
        if (sortField != null) {
            if (sortOrder == SortOrder.ASCENDING) {
                cq.orderBy(cb.asc(myObj.get(sortField)));
            } else if (sortOrder == SortOrder.DESCENDING) {
                cq.orderBy(cb.desc(myObj.get(sortField)));
            }
        }
        return em.createQuery(cq).setFirstResult(first).setMaxResults(pageSize).getResultList();
    }
}

I'm not sure whether this is relevant in this instance, but adding to @perissf's observations, I would be concerned about the following:

if (entry.getValue() != null)
{
    criteria.add(Restrictions.ilike("requestNo",
                                    "%" + entry.getValue() + "%"));
}

For this would resolve into a query akin to

WHERE UPPER(request_no) LIKE '%VALUE%'

which would full table scan, as an index on request_no couldn't be used in this instance, which would be very slow for tables with large amount of rows for two reasons:

• UPPER(request_no) would need a functional index.
• like '%anything' would have to look through every value of request_no regardless of whether a functional index is present or not.