Struct WalFileShared 

pub struct WalFileShared {
    pub wal_header: Arc<SpinLock<WalHeader>>,
    pub min_frame: AtomicU64,
    pub max_frame: AtomicU64,
    pub nbackfills: AtomicU64,
    pub frame_cache: Arc<SpinLock<HashMap<u64, Vec<u64>>>>,
    pub pages_in_frames: Arc<SpinLock<Vec<u64>>>,
    pub last_checksum: (u32, u32),
    pub file: Arc<dyn File>,
    pub read_locks: [LimboRwLock; 5],
    pub write_lock: LimboRwLock,
    pub loaded: AtomicBool,
}

WalFileShared is the part of a WAL that will be shared between threads. A wal has information that needs to be communicated between threads so this struct does the job.

Fields

wal_header: Arc<SpinLock<WalHeader>>

min_frame: AtomicU64

max_frame: AtomicU64

nbackfills: AtomicU64

frame_cache: Arc<SpinLock<HashMap<u64, Vec<u64>>>>

pages_in_frames: Arc<SpinLock<Vec<u64>>>

last_checksum: (u32, u32)

file: Arc<dyn File>

read_locks: [LimboRwLock; 5]

read_locks is a list of read locks that can coexist with the max_frame number stored in value. There is a limited amount because and unbounded amount of connections could be fatal. Therefore, for now we copy how SQLite behaves with limited amounts of read max frames that is equal to 5

write_lock: LimboRwLock

There is only one write allowed in WAL mode. This lock takes care of ensuring there is only one used.

loaded: AtomicBool

Implementations

impl WalFileShared

pub fn open_shared( io: &Arc<dyn IO>, path: &str, page_size: u32, ) -> Result<Arc<UnsafeCell<WalFileShared>>>

pub fn open_shared_inner( io: &Arc<dyn IO>, path: &str, page_size: u32, db_freshly_created: bool, ) -> Result<Arc<UnsafeCell<WalFileShared>>>

Open the shared WAL state for path.

When db_freshly_created is true, the accompanying main database file did not exist (or was empty) and was just bootstrapped by crate::maybe_init_database_file. Any -wal file present on disk is therefore an orphan left behind by a previous database incarnation (e.g. the main .db was deleted while its -wal survived). Replaying such a WAL would resurrect stale committed pages on top of the fresh database, corrupting it (the classic symptom is row counts that grow by the previous content on every reopen). In that case we discard the orphaned WAL and start a fresh one instead of recovering from it.

Trait Implementations

impl Debug for WalFileShared

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.