Storage integrations

Storage integrations connect OptunaSharp studies to SQLite, PostgreSQL, MySQL, Redis, or a remote gRPC storage server. They persist trial parameters, values, intermediate reports, and study metadata. Model files and other large outputs belong in an artifact store.

Choose the storage connection

Integration	Package	Data location	Connection model
SQLite	OptunaSharp.Storage.Sqlite	Local SQLite database	One connection, serialized operations
PostgreSQL / MySQL through EF Core	OptunaSharp.Storage.EFCore	Server database using Optuna's schema	Short-lived contexts and trial-update row locks; trial-creation concurrency limitation
StackExchange.Redis	OptunaSharp.Storage.StackExchange.Redis	Redis journal keys	StackExchange.Redis multiplexer
NRedisStack	OptunaSharp.Storage.NRedisStack	The same Redis journal format	NRedisStack's underlying multiplexer
ServiceStack.Redis	OptunaSharp.Storage.ServiceStack.Redis	The same Redis journal format	ServiceStack client manager / pool
Redis shared primitives	OptunaSharp.Storage.Redis	No standalone storage	URL parser, key layout, and Lua script shared by the three adapters
gRPC	OptunaSharp.Storage.Grpc	A server's backing storage	Managed gRPC client and ASP.NET Core server

These packages target .NET 8 and .NET 10. Install the integration package you use; its project references bring in the core OptunaSharp package. In-memory storage and the file journal are already part of core; see the storage guide.

SQLite is convenient for local persistent studies. PostgreSQL and MySQL support transactional server storage, although the current EF Core trial-creation path has a concurrency limitation. Redis provides an operation journal that each worker replays locally. gRPC centralizes access to a backing storage through a server; it does not replace that storage's durability or concurrency behavior.

Pass a storage object

Construct the integration's storage and pass it to Optuna.CreateStudy or Optuna.LoadStudy. Core's StorageFunctions.GetStorage(string) does not select installed database packages from a URL. The URL belongs in the integration constructor.

using OptunaSharp;
using SqliteStorage = OptunaSharp.Storages.RDBStorage;

using var storage = new SqliteStorage("sqlite:///studies.db");
var study = Optuna.CreateStudy(
    storage: storage,
    studyName: "quadratic",
    loadIfExists: true);

study.Optimize(trial =>
{
    double x = trial.SuggestFloat("x", -10, 10);
    return (x - 2) * (x - 2);
}, nTrials: 20);

Both RDB implementations implement ICachedStorageBackend, so study creation wraps them in CachedStorage automatically. Redis backends must be wrapped in JournalStorage. GrpcStorageProxy already implements BaseStorage and maintains its own trial cache.

Sharing data with Python

The RDB packages map Optuna's SQLAlchemy schema; Redis shares its journal operation format; gRPC shares its protobuf service contract. These are separate compatibility boundaries. Redis replay snapshots are runtime-specific, gRPC does not transport heartbeat operations, and schema upgrades differ between the two RDB packages. Each guide explains its boundary and links to the corresponding interoperability tests.

Storage shares study history, not the in-memory state of a sampler or a running objective. Configure each worker's sampler, pruner, and objective as part of the worker application.

Implementation: storage selection and automatic cache wrapping.