Avoid querying with secondary columns which might fetch and lock additional rows (#20474)

* Accessing UserSession by primary key

This resolves problematic locking queries databases running on SERIALIZABLE isolation level like CockroachDB

Closes #16977

* Avoid querying with expiring column

This resolves problematic locking queries databases running on SERIALIZABLE isolation level like CockroachDB

Closes #16977
This commit is contained in:
Alexander Schwartz 2023-05-23 09:19:58 +02:00 committed by GitHub
parent 2672c47bc8
commit 7f64ca0048
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
3 changed files with 41 additions and 18 deletions

View file

@ -23,10 +23,10 @@ import java.util.List;
import java.util.Map;
import java.util.Objects;
import java.util.UUID;
import java.util.stream.Collectors;
import java.util.stream.Stream;
import jakarta.persistence.EntityManager;
import jakarta.persistence.LockModeType;
import jakarta.persistence.Parameter;
import jakarta.persistence.PersistenceException;
import jakarta.persistence.TypedQuery;
import jakarta.persistence.criteria.CriteriaBuilder;
@ -181,7 +181,10 @@ public abstract class JpaMapStorage<RE extends JpaRootEntity, E extends Abstract
}
JpaPredicateFunction<RE> predicateFunc = mcb.getPredicateFunc();
if (this.isExpirableEntity) {
if (this.isExpirableEntity && (queryParameters.getLimit() != null || queryParameters.getOffset() != null)) {
// only when using pagination exclude expired entities in the query directly
// for all other queries, remove the expired results later as those additional predicates might confuse the database
// to use a bad index (see: CockroachDB), and we assume that expired entities are cleaned from the DB regularly
predicateFunc = predicateFunc != null ? predicateFunc.andThen(predicate -> cb.and(predicate, notExpired(cb, query::subquery, root)))
: this::notExpired;
}
@ -197,6 +200,10 @@ public abstract class JpaMapStorage<RE extends JpaRootEntity, E extends Abstract
// In order to cache the result, the full result needs to be retrieved.
// There is also no difference to that in Hibernate, as Hibernate will first retrieve all elements from the ResultSet.
List<RE> resultList = emQuery.getResultList();
if (isExpirableEntity) {
// remove expired entities when those haven't been excluded by a predicate
resultList = resultList.stream().filter(e -> !isExpired((ExpirableEntity) e, true)).collect(Collectors.toList());
}
cache.put(queryCacheKey, resultList);
return closing(resultList.stream()).map(this::mapToEntityDelegateUnique);

View file

@ -144,14 +144,15 @@ public class MapSingleUseObjectProvider implements SingleUseObjectProvider {
DefaultModelCriteria<SingleUseObjectValueModel> mcb = criteria();
mcb = mcb.compare(SingleUseObjectValueModel.SearchableFields.OBJECT_KEY, ModelCriteriaBuilder.Operator.EQ, key);
MapSingleUseObjectEntity singleUseEntity = singleUseObjectTx.read(withCriteria(mcb)).findFirst().orElse(null);
if (singleUseEntity != null) {
if (isExpired(singleUseEntity, false)) {
singleUseObjectTx.delete(singleUseEntity.getId());
return singleUseObjectTx.read(withCriteria(mcb))
.filter(entity -> {
if (isExpired(entity, false)) {
singleUseObjectTx.delete(entity.getId());
return false;
} else {
return singleUseEntity;
}
}
return null;
return true;
}
})
.findFirst().orElse(null);
}
}

View file

@ -187,13 +187,19 @@ public class MapUserSessionProvider implements UserSessionProvider {
return userEntityToAdapterFunc(realm).apply(userSessionEntity);
}
DefaultModelCriteria<UserSessionModel> mcb = realmAndOfflineCriteriaBuilder(realm, false)
.compare(UserSessionModel.SearchableFields.ID, Operator.EQ, id);
// This is an exceptional case where not to use the criteria query:
// As the ID is already known, and we expect in almost all cases to have exactly one row being returned,
// the provider fetches the instance by ID and does the filtering in the Java code afterward instead
// of using the criteria query. When using a criteria query in earlier versions, the store (CockroachDB) would pick
// a wrong optimization path and lock too many DB rows which would result in transaction-not-serializable exceptions
// on concurrent transactions. This change has been done in the assumption that all stores would be faster
// to evaluate the fetch-by-id than a criteria query.
userSessionEntity = storeWithRealm(realm).read(id);
if (userSessionEntity != null && Objects.equals(userSessionEntity.getRealmId(), realm.getId()) && !userSessionEntity.isOffline()) {
return userEntityToAdapterFunc(realm).apply(userSessionEntity);
}
return storeWithRealm(realm).read(withCriteria(mcb))
.findFirst()
.map(userEntityToAdapterFunc(realm))
.orElse(null);
return null;
}
@Override
@ -314,7 +320,16 @@ public class MapUserSessionProvider implements UserSessionProvider {
LOG.tracef("removeUserSession(%s, %s)%s", realm, session, getShortStackTrace());
storeWithRealm(realm).delete(withCriteria(mcb));
// This is an exceptional case where not to use the criteria query:
// As the ID is already known, the provider does the filtering in the Java code and uses delete-by-id
// instead of using the criteria query to delete rows.
// When using a criteria query in earlier versions, the store (CockroachDB) would pick
// a wrong optimization path and lock too many DB rows which would result in transaction-not-serializable exceptions
// on concurrent transactions. This change has been done in the assumption that delete-by-id would be faster than
// delete-by-criteria for all stores.
if (Objects.equals(session.getRealm(), realm) && !session.isOffline()) {
storeWithRealm(realm).delete(session.getId());
}
}
@Override