Struct Config

Source

pub struct Config { /* private fields */ }

Expand description

Db open-time configuration. Construct via Config::default and modify with the builder methods.

Implementations§

Source §

impl Config

Source

pub fn cache_size(self, bytes: usize) -> Result<Self>

Set the pager’s LRU cache size, in bytes. Rounded down to the nearest 4 KiB page.

§Errors

Returns obj_core::Error::InvalidArgument if the resulting frame count is zero.

Source

pub fn sync_mode(self, mode: SyncMode) -> Self

Set the durability mode the WAL uses for every commit.

Source

pub fn busy_timeout(self, timeout: Duration) -> Self

Set the cross-process / in-process busy-lock timeout. WriteTxn::begin and ReadTxn::begin return Err(Error::Busy) if the relevant lock cannot be acquired within this budget.

Source

pub fn skip_open_check(self, skip: bool) -> Self

Skip the lightweight open-time integrity check (M11 #91).

By default (false), crate::Db::open / crate::Db::open_with run a fast subset of crate::Db::integrity_check before returning: file-header CRC, catalog root sanity, catalog B-tree CRC + invariants, and per-collection pointer-range validation. The walk is bounded to the catalog tree only — no per-collection deep walk — so the cost is essentially independent of the database’s total size.

Set to true to opt out. The knob exists for narrow use cases — fault-injection harnesses that deliberately open a corrupted DB to exercise downstream error paths, hot-reload tooling that re-opens the same file many times per second, or developer workflows that have just run a full Db::integrity_check and don’t want to repeat the catalog portion. Production callers SHOULD leave it on.

Skipping the open check does NOT bypass detection: a corrupted page surfaces on the first operation that touches it. Note also that the obj Db constructor performs an implicit Catalog::open_or_init that reads the catalog B-tree’s reserved row; a DB whose catalog tree is so corrupted that descend fails will still error out of the open path even with skip_open_check(true). The knob’s guarantee is “no EXTRA walk beyond what was already required to construct the Db handle.”

Source

pub fn cross_process_lock(self, enabled: bool) -> Self

Enable / disable the cross-process file lock layer.

When false, the Db opened with this config does NOT acquire OS-level byte-range locks on the database file. Used by the M6 #49 concurrent stress test where every thread shares one Db (and therefore one file descriptor): POSIX OFD locks are per-fd, so multiple threads on the same fd cannot use the lock to enforce inter-thread exclusion — that’s what the in-process write-serialization mutex is for.

Default: true (cross-process locking enabled).