@venizia/ignis-connectors/postgres/pglite
Generated from the built declarations by
vulcan x atlas api gen- do not edit.
3 exports, 3 with TSDoc.
PGliteDriver
class
Postgres compiled to WASM, in-process. Reports PostgreSQL 18.x, so the dialect works unchanged.
One session only, and a second BEGIN neither nests nor errors - it joins the open transaction, so the outer COMMIT commits the inner writer's rows. Hence the 1-slot pool behind acquire().
Consequence: a createConnector() write - the path all repositories take - runs INSIDE any open transaction and dies with its ROLLBACK. Under concurrency, write through acquire().
export declare class PGliteDriver<Schema extends TAnyDataSourceSchema = TAnyDataSourceSchema> implements TRelationalDriver<Schema, PGlite> {
/** The engine this driver speaks; the migration ledger reads it (see {@link RelationalDialects}). */
readonly dialect = "postgres";
private readonly client;
/**
* The pooled resource IS the single client, so `create()` always hands back the same instance.
* The pool contributes the mutual exclusion PGlite lacks, not connection multiplicity.
*/
private readonly slot;
private readonly logger;
/**
* A leaked transaction never releases the one slot, so without a timeout every later `acquire()`
* in the process hangs forever and silently. 30s outlasts any honest transaction on an in-process
* database with no network in it, and turns a leak into a named error instead of a dead process.
*/
static readonly DEFAULT_ACQUIRE_TIMEOUT_MS = 30000;
/**
* PGlite's Emscripten runtime plants this status on the HOST process during ordinary operation -
* measured on `@electric-sql/pglite` 0.5.5, `process.exitCode` is `undefined` before the first
* query and 99 after it, and every later query re-arms it. A server that shuts down cleanly then
* exits 99, which Docker, systemd and CI all read as a crash.
*
* Named rather than inlined so a future PGlite using a different status stops matching and makes
* the problem visible again, instead of {@link end} silently forcing a zero it did not verify.
*/
private static readonly WASM_HOST_EXIT_STATUS;
constructor(opts: TPGliteDriverOptions);
createConnector(opts: {
schema: Schema;
}): TRelationalConnector<Schema>;
acquire(opts: {
schema: Schema;
}): Promise<TRelationalConnection<Schema>>;
getClient(): PGlite;
end(): Promise<void>;
/**
* Clears the exit status PGlite planted on the host - see {@link WASM_HOST_EXIT_STATUS} - and
* only while it still holds exactly that value, so an exit code the application set itself
* survives untouched.
*
* `globalThis.process` is bound to a local and never member-accessed as `globalThis.process.x`:
* this driver has to bundle clean for a browser target, where the property read yields
* `undefined` and the optional chain short-circuits, while the member-access spelling is what the
* purity gate treats as fatal.
*/
private clearWasmHostExitStatus;
/**
* The pool's destroy hook. A failed COMMIT/ROLLBACK may leave the session inside a transaction,
* and the next borrower would inherit it. `ROLLBACK` outside a transaction is a no-op, so this is
* safe whichever state the session is actually in.
*/
private scrubSession;
}Source: packages/connectors/src/relational/postgres/drivers/pglite.ts
TPGliteDataSource
type
A Postgres datasource whose client is PGlite: what PGliteCRUDRepository takes and what its dataSource is.
export type TPGliteDataSource = IPostgresDataSource<{}, TAnyDataSourceSchema, {}, PGlite>;Source: packages/connectors/src/relational/postgres/repositories/core/pglite.ts
TPGliteDriverOptions
type
Pool knobs the driver forwards, minus size: one slot IS the mutual exclusion PGlite lacks.
export type TPGliteDriverOptions = Omit<IPoolControlOptions, 'size'> & {
client: PGlite;
};Source: packages/connectors/src/relational/postgres/drivers/pglite.ts