Struct Cache 

pub struct Cache { /* private fields */ }

Implementations

impl Cache

pub fn open(path: &Path, max_size_mb: usize) -> Result<Self>

Opens the cache, without fsyncing on every commit.

A durable commit costs ~2.0 ms for a 49-row entry against ~55 µs without one — 97% of a cache write, and far more than the query whose result is being stored. Paying it makes populating the cache several times more expensive than not having a cache at all.

MDB_NOSYNC drops the D of ACID and keeps the rest: LMDB still writes whole pages and flips the meta page atomically, so the database stays structurally valid — a system crash can cost the most recent commits, not the store. That is the right trade for this data, because every entry is a copy of something Postgres can produce again, so losing one costs a cache miss. Drop flushes, so an ordinary shutdown keeps everything anyway; only a power loss or a kill -9 forfeits the last writes.

Deliberately not combined with MDB_WRITEMAP. The two together do risk real corruption on a system crash, and the writable mmap buys nothing once the fsync is gone.

pub fn get(&self, key: &str) -> Result<Option<CachedEntry>>

pub fn put( &self, key: &str, rows: Vec<DecodedValue>, tags: Vec<String>, ) -> Result<()>

pub fn flush(&self) -> Result<()>

Forces everything committed so far to disk. Called on Drop, so an ordinary shutdown loses nothing; call it directly to checkpoint sooner.

pub fn invalidate(&self, tags: &[String]) -> Result<()>

Evicts every cache entry tagged with any of tags — called from the LISTEN/NOTIFY invalidation path after a write commits.

pub fn stat(&self) -> Result<CacheStats>

Current size of the cache — see CacheStats.

pub fn clear(&self) -> Result<()>

Evicts every cache entry, unconditionally — used by the pylon cache purge CLI command. Unlike invalidate, this doesn’t require knowing any tags up front.

Trait Implementations

impl Drop for Cache

fn drop(&mut self)

Flush on the way out, so the deferred sync open selects costs nothing on an ordinary shutdown — only an abrupt one (power loss, SIGKILL) forfeits the most recent writes.

Errors are swallowed deliberately: this runs during teardown, where there is nobody left to report to, and a cache that failed to persist is a cache miss rather than a fault.

fn pin_drop(self: Pin<&mut Self>)

🔬This is a nightly-only experimental API. (pin_ergonomics)

Execute the destructor for this type, but different to Drop::drop, it requires self to be pinned.