Description
I have the need for nested transactions and have already implemented a simple but working solution. I would like to bring this to your attention and double check if this is the desired behavior or not.
Imagine an API handler needs to ensure that an operation across two different BMCs is transactional.
pub async fn some_api_handler(State(mm): State<ModelManager>) -> Result<()> {
let mm = mm.new_with_txn()?;
mm.dbx().begin_txn().await?;
FooBmc::create(&mm, &foo_payload).await?;
BarBmc::create(&mm, &bar_payload).await?;
mm.dbx().commit_txn().await?;
Ok(())
}
However, BarBmc
needs to make sure that it's create
function is always in a transaction, regardless of whether the calling site already has a running transaction or not.
impl BarBmc {
pub async fn create(mm: &ModelManager, payload: &BarPayload) -> Result<()> {
let mm = mm.new_with_txn()?;
mm.dbx().begin_txn().await?;
let new_id = base::create::<Self, _>(&mm, payload).await?;
Self::something_else(&mm, new_id).await?;
mm.dbx().commit_txn().await?;
Ok(())
}
}
If you call BarBmc::create
without a transaction, everything will work.
If you call BarBmc::create
inside a running transaction, you get a (code: 5) database is locked error
(probably only in SQLite).
A simple solution would be that mm.new_with_transaction()
should not reset the txn_holder
.
pub fn new_with_txn(&self) -> ModelManager {
+ if self.dbx.with_txn() {
+ return ModelManager {
+ dbx: self.dbx.clone(),
+ };
+ };
let dbx = Dbx::new(self.dbx.db().clone(), true);
ModelManager { dbx }
}
impl Dbx {
pub fn new(db_pool: Db, with_txn: bool) -> Self {
Self {
db_pool,
txn_holder: Arc::default(),
with_txn,
}
}
+
+ pub fn with_txn(&self) -> bool {
+ self.with_txn
+ }
}
Let me know what you think.