DEFERRED was a bit too deferred - it allowed the dirty pages to be
written out to WAL before checking for violations, resulting in the
violations effectively being committed even though the transaction ended
up aborting
Closes#3784Closes#3785
DEFERRED was a bit too deferred - it allowed the dirty pages to be
written out to WAL before checking for violations, resulting in the
violations effectively being committed even though the transaction
ended up aborting
Without this change and running:
```
cd stress
while cargo run -- --nr-threads=4 -i 1000 --verbose --busy-timeout=0; do; done
```
I can produce a deadlock quite reliably.
With this change, I can't.
Even with 5 second busy timeout (the default), the run makes progress
although it is slow as hell because of the busy timeout.
Full disclosure: i couldn't figure out based on parking lot RwLock
semantics why this would fix it, so maybe it just lessens the
probability
Reviewed-by: Preston Thorpe <preston@turso.tech>
Closes#3759
Without this change and running:
```
cd stress
cargo run -- --nr-threads=4 -i 1000 --verbose --busy-timeout=0
```
I can produce a deadlock quite reliably.
With this change, I can't.
Even with 5 second busy timeout (the default), the run makes progress although it is slow as hell because of the busy timeout.
Rolling back a transaction on Error should result in
`connection.auto_commit` being set back to true.
Added a regression test for this where a UNIQUE constraint violation
rolls back the transaction and trying to COMMIT will fail.
Currently, our default conflict resolution strategy is ROLLBACK, which
ends the transaction. In SQLite, the default is ABORT, which rolls back
the current statement but allows the transaction to continue.
We should migrate to default ABORT once we support subtransactions.
Closes#3746
Reviewed-by: Preston Thorpe <preston@turso.tech>
Closes#3747
This patch pushes unsafe Send and Sync to individual components instead
of doing it at Database level. This makes it easier for us to
incrementally fix thread-safety, but avoid developers adding more thread
unsafe code.
Reviewed-by: Jussi Saurio <jussi.saurio@gmail.com>
Closes#3745
`INSERT OR IGNORE INTO t VALUES (...)` can trivially be rewritten to
`INSERT INTO t VALUES (..) ON CONFLICT DO NOTHING`
This PR does this rewriting, as well as finishes a large refactor on
INSERT translation in general.. I just need a break from the rest of
this feature tbh.. just was getting under my skin and I have been in
`translate` land for too long.
Closes#3742
Rolling back a transaction should result in `connection.auto_commit` being set
back to true.
Added a regression test for this where a UNIQUE constraint violation rolls back
the transaction and trying to COMMIT will fail.
Currently, our default conflict resolution strategy is ROLLBACK,
which ends the transaction. In SQLite, the default is ABORT, which rolls back
the current statement but allows the transaction to continue.
We should migrate to default ABORT once we support subtransactions.
This patch pushes unsafe Send and Sync to individual components instead
of doing it at Database level. This makes it easier for us to
incrementally fix thread-safety, but avoid developers adding more thread
unsafe code.
This PR introduces a `Context` object that is stored in the `Completion`
that currently only stores a `Waker`. In the future, I want to add some
sort of abort signal so that we can abort tasks that share the same
Context. To pass the Waker, I introduced a `step_with_waker` function in
`Statement` that delegates to an internal `_step` function. `_step` is
the previous `step` but just with the `Option<&Waker>` argument.
I was going to try and have the BusyHandler by truly async as well, but
I decided to not do it here, because it will be slightly complicated to
achieve.
Closes#3535
The previous implementation of CompletionGroup would call the group's
callback function directly when the last completion finished:
if prev == 1 {
let group_result = group.result.get().and_then(|e| *e);
(group.complete)(group_result.map_or(Ok(0), Err));
}
This broke nested completion groups because parent groups track their
children via the Completion::callback() method. By calling the function
pointer directly, we bypassed the completion chain and parent groups
never received notification that their child had completed.
The fix stores a reference to the group's own Completion object in
self_completion during build(). When the last child finishes, we call
group_completion.callback() instead of invoking the function directly.
This properly propagates through the completion hierarchy, ensuring
parent groups decrement their outstanding count and eventually complete.
This matches the behavior of individual completions and maintains the
invariant that all completions notify their parents through the unified
callback() mechanism.
The previous implementation of CompletionGroup::add() would filter out
successfully-finished completions:
if !completion.finished() || completion.failed() {
self.completions.push(completion.clone());
}
This caused a problem when combined with drain() in the calling code.
Completions that were already finished would be removed from the source
vector by drain() but not added to the group, effectively losing track
of them.
This breaks the invariant that all completions passed to a group must
be tracked, regardless of their state. The build() method already
handles finished completions correctly by not including them in the
outstanding count.
The fix is to always add all completions and let build() handle their
state appropriately, matching the behavior of the old io_yield_many!()
macro.
Closes#3687 .
Previously, the `try_fold_expr_to_i64` function casted `NULL` as `0`
when evaluating expressions in `LIMIT` or `OFFSET` clauses. I removed
this function since evaluating the expression directly and relying on
the MustBeInt operation for casting seems to handle everything.
Closes#3695
We merged two concurrent fixes to `nchange` handling last night and
AFAICT the fix in #3692 was incorrect because it doesn't count UPDATEs
in cases where the original row was DELETEd as part of the UPDATE
statement.
The correct fix was in 87434b8
EDIT: okay, it's not strictly _incorrect_ in #3692 I guess, I just think
it's more intuitive to increment the change for UPDATE in the `insert`
opcode because that's what performs the actual update.
Closes#3735
We merged two concurrent fixes to `nchange` handling last night and
AFAICT the fix in #3692 was incorrect because it doesn't count UPDATEs
in cases where the original row was DELETEd as part of the UPDATE
statement.
The correct fix was in 87434b8
