Insert or Update about requery

[npurushe]

Hi yes upsert is a good thing. Unfortunately the databases that support it all do it a different way so tricky to implement.

Is this for SQLite? SQLite only has insert or replace which is not the same as upsert unfortunately. I think for SQLite if you have some primary key it's best to check if that key already has a record first and retrieve it and base your insert/update decision on that, while of course being careful of concurrency.

[eugenkiss]

Yes, it's for SQLite in Android. Thanks for your tips. I've create two new versions of upsert (or rather a rough approximation thereof as you correctly mentioned that SQLite does not really have upsert support) extension functions.

fun <E: Persistable, K> SingleEntityStore<Persistable>.upsert(clazz: Class<E>, key: K, entity: E): Single<E> {
    return findByKey(clazz, key).toObservable().materialize().flatMap { notification ->
        if (notification.isOnError) {
            insert(entity).toObservable()
        } else {
            update(entity).toObservable()
        }
    }.toSingle()
}

fun <E: Persistable, K> BlockingEntityStore<Persistable>.upsert(clazz: Class<E>, key: K, entity: E): E {
    try {
        findByKey(clazz, key) == null
    } catch (e: NoSuchElementException) {
        return insert(entity)
    }
    return update(entity)
}

With the first one I had some concurrency problems (where I upserted a list of entities). I wasn't patient enough to fix them so I created the second extension function for a blocking version which now seems to work alright.

By the way, there seems to still be something fishy when I look at Requery's beforeExecuteQuery logs. Basically, I have five entities that I upsert and there are four distinct entities in the database. The first and the last of the five entities represent the same one in the database. The logs always show only four select statements but five insert or update statements. Can it be that a the last select is not executed because the entity is found in the cache? If yes, maybe that should be indicated in the logs?

[npurushe]

Hi yes if you use findByKey it's possible there are no queries made using the default caching strategy. I agree default logger should be extended to log that.