Eryri.FileDistributedCache 1.0.9

Eryri.FileDistributedCache

Keep more reusable data without giving more RAM to your cache—or running another cache service.

Eryri.FileDistributedCache stores cache values on the local filesystem behind .NET's IDistributedCache and IBufferDistributedCache APIs.

Give it a byte limit and it evicts entries to make room before admitting a new write; expired entries are cleaned up automatically.

Use an ephemeral cache when you only need local disk capacity, or provide a persistent cache directory when you want entries to survive process restarts. Persistent caches use a write-ahead log with periodic snapshots to recover efficiently without keeping the entire cache index in memory.

It is built for a single process, with concurrent access, configurable eviction, synchronous and asynchronous operations, and Native AOT compatibility.

Use it directly when local disk is the right cache, or register it as the secondary provider for HybridCache when you want a memory-first cache with a larger local-disk tier.

Why use it?

• Keep RAM for your application. Store reusable payloads in files rather than retaining the whole disk-backed cache in memory. A local filesystem is useful when your reusable working set is larger than the memory budget you want to assign to caching.
• Keep writes moving near capacity. Set a byte limit and the cache evicts entries when needed to make room before adding a new one. Choose LRU, LFU, FIFO or TTL to decide what goes first.
• Let expired data clean itself up. Absolute and sliding expiration are supported. A background cleanup timer is rescheduled for the next expiry, using a priority queue rather than relying solely on fixed-interval sweeps.
• Fit familiar .NET APIs. Use IDistributedCache or IBufferDistributedCache; the implementation is thread-safe and Native AOT-compatible.
• Avoid a new service for a local need. There is no cache server to deploy just to give one process more disposable cache capacity. The trade-off is that entries are not shared across processes.

Good fit: a single-process service or worker with reproducible cache values and useful local disk space. Not a fit: cross-replica consistency. The cache's configured byte limit controls its own entries; it does not reserve free space for other applications.

Scope at a glance

Install and get started

The NuGet package ID is Eryri.FileDistributedCache.

dotnet add package Eryri.FileDistributedCache

The namespace is Eryri.Extensions.Caching.FileSystem.

using Eryri.Extensions.Caching.FileSystem;
using Microsoft.Extensions.Caching.Distributed;
using Microsoft.Extensions.DependencyInjection;

using var services = new ServiceCollection()
    .AddDistributedFileCache()
    .BuildServiceProvider();

var cache = services.GetRequiredService<IDistributedCache>();

await cache.SetStringAsync(
    "product:42",
    "Cached product data",
    new DistributedCacheEntryOptions
    {
        AbsoluteExpirationRelativeToNow = TimeSpan.FromMinutes(10)
    });

var value = await cache.GetStringAsync("product:42");
Console.WriteLine(value ?? "Cache miss");

Treat a miss as normal: an entry may have expired, been evicted, or disappeared when the process restarted. Retrieve or recompute the value from its source of truth and write it back.

Use with HybridCache

HybridCache can use an IDistributedCache provider as its secondary tier. Register this provider alongside HybridCache to put a local filesystem behind the usual in-memory tier. This is not a guarantee that values spill to disk only after RAM fills: HybridCache manages its tiers according to its own behavior and configuration.

using Eryri.Extensions.Caching.FileSystem;
using Microsoft.Extensions.Caching.Hybrid;
using Microsoft.Extensions.DependencyInjection;

using var services = new ServiceCollection()
    .AddDistributedFileCache()
    .AddHybridCache().Services
    .BuildServiceProvider();

var cache = services.GetRequiredService<HybridCache>();
var options = new HybridCacheEntryOptions
{
    Expiration = TimeSpan.FromMinutes(10),
    LocalCacheExpiration = TimeSpan.FromSeconds(5), // Keep in-memory items for only a short period
};

await cache.SetAsync("product:42", "Cached product data", options);

var value = await cache.GetOrCreateAsync(
    "product:42",
    _ => ValueTask.FromResult("Value to create on a miss"),
    options);

Console.WriteLine(value);

Both tiers remain local to this process. For a shared secondary cache across replicas, use a backend designed for sharing.

Expiration and capacity

Set a byte-based size limit if you want the cache to control its footprint. When a new entry would exceed that limit, the cache evicts existing entries as needed before adding it. The configured eviction policy determines which eligible entries are selected. Expired entries are also removed automatically by a background timer scheduled for the next expiry.

The package supports DistributedCacheEntryOptions:

• AbsoluteExpiration
• AbsoluteExpirationRelativeToNow
• SlidingExpiration.

Expiration controls whether an entry is valid; a capacity policy chooses which entries to remove to admit new data. The two are different mechanisms.

For manual capacity reduction, IFileDistributedCache.Compact(decimal percentage) selects entries according to the configured eviction policy. For example, Compact(0.10) requests removal of at least 10% of existing entries. This is useful even when no size limit is configured.

Configuration reference

Operational boundaries

• One process owns one cache. Do not share the cache directory or assume another process can observe its entries safely.
• Misses are expected. Cached values must be safe to lose and regenerable after expiration, eviction, or storage failure.
• Filesystem access matters. The process needs permission to create, read, write, and delete files in its cache directory. A configured cache-size limit does not protect against another workload filling the underlying filesystem.
• Write failures: If the filesystem rejects a write, the entry is not added and an error is logged. Callers should not treat Set as proof that the entry can later be read back.

Benchmarks

The benchmark that matters most for this package is sustained operation with realistic workloads while the cache is already full: this tests the cost of making room, not just writing into an empty directory. The figures below are useful directional evidence, not a cross-machine performance guarantee.

• PayloadSize=4KiB
• CacheSizeLimit=1MiB
• ParallelOperations=100
• InvocationCount=30
• IterationCount=5
• UnrollFactor=1
• WarmupCount=1

Tests located here: BenchmarkTests.cs

1s == 1000ms, 1ms == 1000us, 1us == 1000ns

Eryri.FileDistributedCache

DamianH.FileDistributedCache

NeoSmart.Caching.Sqlite

LiteDb.Extensions.Caching

Caching.FileBackedDistributedCache

Net.DistributedFileStoreCache