Struct Builder 

pub struct Builder { /* private fields */ }

Builds a Client. dsn is required up front (this crate never parses pylon.toml — see the crate-level docs); everything else has a sensible default.

Implementations

impl Builder

pub fn new(dsn: impl Into<String>) -> Self

pub fn max_pool_size(self, max_pool_size: usize) -> Self

pub fn cache(self, path: impl Into<PathBuf>, max_size_mb: usize) -> Self

Opts into read-through result caching at an LMDB-backed directory — mirrors pylon.toml’s [cache] section, minus per-type ([cache.sets.<Name>]) overrides: caching here is a single global on/off. Omit this entirely for no caching (today’s default behavior). Only Client’s own query methods read/write the cache — Transaction never populates it, since rows read inside a transaction aren’t committed, matching pylon/client.py’s AsyncTransaction.

This client’s own writes evict the tags they touch, from inside a transaction too, so a program that writes and then re-reads sees its own change. Writes from other processes still need a listener on the same directory (e.g. pylon worker start) to invalidate.

Opens its own LMDB handle for path — heed (the LMDB binding this crate uses) refuses a second Env::open on the same canonicalized path while an earlier handle onto it is still alive within the same process, so a caller building more than one Client that should share one cache directory (e.g. one per named pylon.toml connection) must use Builder::cache_handle with one already-open Arc<pylon_cache::Cache> instead of calling this per client.

pub fn cache_handle(self, cache: Arc<Cache>) -> Self

Like Builder::cache, but attaches to an already-open handle instead of opening a new one — see that method’s doc comment for why this exists.

pub fn build(self) -> Result<Client>

Builds the client without touching the database: the connection pool and the schema snapshot are opened together on first use, or on an explicit Client::ensure_connected. A Client can therefore be constructed outside an async context, and a configured-but-never-queried connection costs nothing.

The cache, when Builder::cache configured one, is opened here — it’s a local LMDB directory rather than a network resource, and a bad path is worth failing on at construction rather than on whichever query happens to run first.

A process that wants a bad DSN/host/credentials — or a database where neither pylon migration apply nor pylon migration watch has ever run, so there is no schema snapshot to fetch — to fail at startup instead of on its first query should call Client::ensure_connected right after this.