Tuesday, April 28, 2009

object references an unsaved transient instance

    Match m = new Match();

    for (int playerId : playerIds) {
        Player p = playerDao.read(playerId);
        // update player's stat
        PlayerStats ps = playerStatsDao.findByPlayerAndYear(playerId, year);

org.hibernate.TransientObjectException: object references an unsaved transient instance - save the transient instance before flushing: poker.db.model.Match.

On 2nd iteration of the loop, in line 8, while querying playerstats, it threw the above exception. It was doing a *flush* before the query since playerstats was dirty atm.

I really made an effort to dig into it. It turns out that it was caused by the *inverse* setting of a many-to-many relationship. I used to think it doesn't matter as long as you set either side of it.

In Class Match:
public void addParticipant(Player p) {

PlayStats * <---> 1 Player * <---> * Match
All relationships are bi-directional. PlayStats is owner side by default for the many-to-one. And I set (inverse=true) on Player side by rolling a dice. I thought when I called match.addPaticipant(), both collections are updated and the relation is built. It shouldn't make any difference by setting inverse.

I imagine that the reference link is uni-directional for underlying presistence with the inverse setting. When Player is the owner of this many-to-many, the link is PlayerStats ->Player->Match, since Match object is a transient object (not saved yet) when we tried to save/update PlayerStats, it threw the above exception. However, if we change the owner side to Match, the reference link is broken, PlayerStats->Player, Match->Player, save/update PlayerStats has nothing to do with the state of Match object anymore. Thus it runs perfectly!

* Inverse=true means relationship owner to avoid unnecessary update. Basically on many side, insert a new record already with FK value, you don't need another update statement to set it.

Alternatively, the second solution is to set FlushMode of query in playerStatsDao.findByPlayerAndYear().

public PlayerStats findByPlayerAndYear(int playerId, int year) {
    String queryString = "from PlayerStats ps where ps.player.id = :playerId "
                          + "and ps.year = :year";
    Query query = getSession().createQuery(queryString)
                .setInteger("playerId", playerId)
                .setInteger("year", year)
    return (PlayerStats) query.uniqueResult();

By default, FlushMode is AUTO, flush synchronize the in-memory persistent object with underlying database store by running the SQLs, the changes not visible to others until tx.commit() though.

Note that flush occurs in three cases: transcation commit, session.flush() invoked or before the query. By changing flush to tx commit time, we avoid the "object references an unsaved transient instance" issue. It is also good for performance.

No comments: