If we don't clear the dirty pages, we will initiate a rollback. In the
rollback, we will attempt to clear the whole page cache, but it will
then panic because there will still be dirty pages from the failed
writev
Reviewed-by: Jussi Saurio <jussi.saurio@gmail.com>
Closes#3189
This PR introduces support for foreign key constraints, and the `PRAGMA
foreign_keys;`, and relevant opcodes: `FkCounter` and `FkIfZero`.
Extensive fuzz tests were added both for regular and composite
PK/rowid/unique index constraints, as well as some really weird
edgecases to make sure we our affinity handling is correct as well when
we trigger the constraints.
Foreign-key checking is driven by two VDBE ops: `FkCounter` and
`FkIfZero`, and
`FkCounter` is a running meter on the `Connection` for deferred FK
violations. When an `insert/delete/update` operation creates a potential
orphan (we insert a child row that doesn’t have a matching parent, or we
delete/update a parent that children still point at), this counter is
incremented. When a later operation fixes that (e.g. we insert the
missing parent or re-target the child), we decrement the counter. If any
is remaining at commit time, the commit fails. For immediate
constraints, on the violation path we emit Halt right away.
`FkIfZero` can either be used to guard a decrement of FkCounter to
prevent underflow, or can potentially (in the future) be used to avoid
work checking if any constraints need resolving.
NOTE: this PR does not implement `pragma defer_foreign_keys` for global
`deferred` constraint semantics. only explicit `col INT REFERENCES t(id)
DEFERRABLE INITIALLY DEFERRED` is supported in this PR.
This PR does not add support for `ON UPDATE|DELETE CASCADE`, only for
basic implicit `DO NOTHING` behavior.
~~NOTE: I did notice that, as referenced here: #3463~~
~~our current handling of unique constraints does not pass fuzz tests, I
believe only in the case of composite primary keys,~~ ~~because the fuzz
test for FK referencing composite PK is failing but only for UNIQUE
constraints, never (or as many times as i tried) for foreign key
constraints.~~
EDIT: all fuzzers are passing, because @sivukhin fixed the unique
constraint issue.
The reason that the `deferred` fuzzer is `#[ignore]`'d is because sqlite
uses sub-transactions, and even though the fuzzing only does 1 entry per
transaction... the fuzzer can lose track of _when_ it's in a transaction
and when it hits a FK constraint, and there is an error in both DB's, it
can just continue to do run regular statements, and then the eventual
ROLLBACK will revert different things in sqlite vs turso.. so for now,
we leave it `ignore`d
Reviewed-by: Jussi Saurio <jussi.saurio@gmail.com>
Closes#3510
- add index root pages to list of root pages to check
- check for dangling (unused) pages
```sql
$ cargo run wut.db
turso> .mode list
turso> pragma integrity_check;
Page 3: never used
Page 4: never used
Page 7: never used
Page 8: never used
```
```sql
$ sqlite3 wut.db 'pragma integrity_check;'
*** in database main ***
Page 3: never used
Page 4: never used
Page 7: never used
Page 8: never used
```
Closes#3613
op_destroy was assuming we never yield IO from
BTreeCursor::btree_destroy(), so every so often it would just not
complete the procedure and leave dangling pages in the database
Closes#3608Closes#3618
op_destroy was assuming we never yield IO from BTreeCursor::btree_destroy(),
so every so often it would just not complete the procedure and leave
dangling pages in the database
This PR add proper program abort in case of unfinished statement reset
and interruption.
Also, this PR makes rollback methods non-failing because otherwise of
their callers usually unclear (if rollback failed - what is the state of
statement/connection/transaction?)
Reviewed-by: Preston Thorpe <preston@turso.tech>
Closes#3591
There is currently a bug found in our materialized view implementation
that happens when we delete a row, and then re-insert another row with
the same primary key.
Our insert code needs to detect updates and generate a DELETE +
INSERT. But in this case, after the initial DELETE, the fresh insert
generates another delete.
We ended up with the wrong response for aggregations (and I am pretty
sure even filter-only views would manifest the bug as well), where
groups that should still be present just disappeared because of the
extra delete.
A new test case is added that fails without the fix.
On reaching 8 MiB checkpoint threshold we perform a blocking checkpoint
on the logical log. These changes modified how transaction_state is
tracked so that on a regular transaction we don't update it checkpoint's
state machine.
I wonder if checkpoint should stay inside commit's transaction locks or
like I did, where checkpoint happens right after committing transaction
but this happens on the same query during halt.
Closes#3565
Fixes the following problems with COLLATE:
- Fix: incorrectly used e.g. `x COLLATE NOCASE = 'fOo'` as index
constraint on an index whose column was not case-insensitively collated
- Fix: various ephemeral indexes (in GROUP BY, ORDER BY, DISTINCT) and
subqueries did not retain proper collation information of columns
- Fix: collation of a given expression was not determined properly
according to SQLite's rules
Adds TCL tests and fuzz test
Closes#3476Closes#1524Closes#3305
Reviewed-by: Preston Thorpe <preston@turso.tech>
Closes#3538
We don't need to clear the cursors explicitly because OpenRead and
OpenWrite will anyway replace them.
Reviewed-by: Preston Thorpe <preston@turso.tech>
Closes#3526
The VDBE step() function was taking Arc<MvStore> by value, causing it to
be cloned on every single step of query execution. This resulted in
thousands of atomic reference count increments/decrements per query,
showing up as a major hotspot in profiling.
Changed step() and related functions to take Option<&Arc<MvStore>>
instead, passing a reference rather than cloning the Arc. This eliminates
the unnecessary atomic operations while maintaining the same semantics.
essentially after the first runthrough of `op_transaction` per a
given `ProgramState`, we weren't resetting the instruction state
to `Start´ at all, which means we didn't do any transaction state
checking/updating.
PR includes a rust bindings regression test that used to panic before
this change, and I bet it also fixes this issue in turso-go:
https://github.com/tursodatabase/turso-go/issues/28
Fixes:
- `start_value` and `length_value` should be casted to integers
- proper handling of utf-8 characters
- do not need to cast blob to string, as substr in blobs refers to byte
indexes and not char-indexes
Closes#3465
Consolidates the `exec_trim`, `exec_rtrim`, `exec_ltrim` code and only
pattern matches on whitespace character.
Fixes#3319
Reviewed-by: Jussi Saurio <jussi.saurio@gmail.com>
Closes#3437
There were 2 problems:
1. The SELECT wasn't propagating which register it used for its results,
so sometimes the INSERT read bad data.
2. `TableReferences::contains_table` was only checking the top-level
tables, not the nested tables in FROM queries. This condition is used to
emit "template 4", the bytecode template for self-inserts.
Closes https://github.com/tursodatabase/turso/issues/3312
Reviewed-by: Jussi Saurio <jussi.saurio@gmail.com>
Closes#3436
**Handle table ID / rootpages properly for both checkpointed and non-
checkpointed tables**
Table ID is an opaque identifier that is only meaningful to the MV
store.
Each checkpointed MVCC table corresponds to a single B-tree on the
pager,
which naturally has a root page.
**We cannot use root page as the MVCC table ID directly because:**
- We assign table IDs during MVCC commit, but
- we commit pages to the pager only during checkpoint
which means the root page is not easily knowable ahead of time.
**Hence:**
- MVCC table ids are always negative
- sqlite_schema rows will have a negative rootpage column if the
table has not been checkpointed yet.
- on checkpoint when the table is allocated a real root page, we update
the row in sqlite_schema and in MV store's internal mapping
**On recovery:**
- All sqlite_schema tables are read directly from disk and assigned
`table_id = -1 * root_page` -- root_page on disk must be positive
- Logical log is deserialized and inserted into MV store
- Schema changes from logical_log are captured into the DB's global
schema
**Note about recovery:**
I changed MVCC recovery to happen on DB initialization which should
prevent any races, so no need for `recover_lock`, right @pereman2 ?
Closes#3419
As per SQLite API, sqlite3_reset() does *not* clear bind parameters.
Instead they're persistent across statement reset and only cleared with
sqlite3_clear_bindings().
Table ID is an opaque identifier that is only meaningful to the MV store.
Each checkpointed MVCC table corresponds to a single B-tree on the pager,
which naturally has a root page.
We cannot use root page as the MVCC table ID directly because:
- We assign table IDs during MVCC commit, but
- we commit pages to the pager only during checkpoint
which means the root page is not easily knowable ahead of time.
Hence, we:
- store the mapping between table id and btree rootpage
- sqlite_schema rows will have a negative rootpage column if the
table has not been checkpointed yet.
