Cache

Cache (v5.4.0 onwards)

Purpose

To provide results from cache where available.

Premise: 'Some proportion of requests may be similar'

The Polly CachePolicy is an implementation of read-through cache, also known as the cache-aside pattern. Providing results from cache where possible reduces overall call duration and can reduce network traffic.

Retrieving a result from an in-memory cache can eliminate a downstream call entirely. A distributed cache can be used to provide a shared cache across upstream nodes; to retrieve values from a nearer-by network resource than the underlying called system might be; or where caching requirements exceed in-memory storage.

Working with CacheProviders

Polly CachePolicy operates in conjunction with an ISyncCacheProvider or IAsyncCacheProvider implementation.

The following existing implementations are available via separate nuget packages:

Package	Description	Supported targets
`Polly.Caching.Memory` _{(nuget for Polly >=6.0.1 ; nuget for Polly <=5.9.0 ; github and doco)}	An in-memory cache implementation using the standard .NET Framework / .NET Core `MemoryCache` providers.	.NET 4.0 .NET 4.5 .NET Standard 1.1 (supports .NET Core and Xamarin) .NET Standard 2.0
`Polly.Caching.Distributed` _{(nuget ; github and doco)}	Supports any implementation of .NET Core's `IDistributedCache`, including the Redis implementation and SQL-server-based implementations that Microsoft provides.	.NET Standard 1.1 (supports .NET Core and Xamarin) .NET Standard 2.0

Use the below compatibility grid to select the correct version of cache providers to use with your version of Polly:

Version of Polly	Version of above cache providers to use
to V6	v2
v7 onwards	v3 onwards

New cache providers can also be implemented against the easy to fulfil ISyncCacheProvider and IAsyncCacheProvider interfaces.

Operation

The cache key to use is determined according to the supplied (or default) cache key strategy.
Where the cache holds a value under the corresponding key:
- the delegate passed to .Execute(...) or similar is not called
- the value from cache is returned instead.
Where the cache does not hold a result under the corresponding key:
- the delegate passed to .Execute(...) or similar is called as usual
- the retrieved value is put in the cache, using the configured time-to-live strategy
- the retrieved value is returned.

Syntax

Defining CachePolicy

CachePolicy cache = Policy
  .Cache(ISyncCacheProvider cacheProvider
           , TimeSpan ttl | ITtlStrategy ttlStrategy
          [, ICacheKeyStrategy cacheKeyStrategy | Func<Context, string> cacheKeyStrategy]
              [, Action<Context, string, Exception> onCacheError]
                   |
              [, Action<Context, string> onCacheGet
               , Action<Context, string> onCacheMiss
               , Action<Context, string> onCachePut
               , Action<Context, string, Exception> onCacheGetError 
               , Action<Context, string, Exception> onCachePutError]
           );

CachePolicy cache = Policy
  .CacheAsync(IAsyncCacheProvider cacheProvider
           , TimeSpan ttl | ITtlStrategy ttlStrategy
          [, ICacheKeyStrategy cacheKeyStrategy | Func<Context, string> cacheKeyStrategy]
              [, Action<Context, string, Exception> onCacheError]
                   |
              [, Action<Context, string> onCacheGet
               , Action<Context, string> onCacheMiss
               , Action<Context, string> onCachePut
               , Action<Context, string, Exception> onCacheGetError 
               , Action<Context, string, Exception> onCachePutError]
           );

CachePolicy<TResult> cache = Policy
  .Cache<TResult>(ISyncCacheProvider<TResult> cacheProvider, /* etc */);

CachePolicy<TResult> cache = Policy
  .CacheAsync<TResult>(IAsyncCacheProvider<TResult> cacheProvider, /* etc */);

Using CachePolicy

Like any other policy, a CachePolicy may be re-used across multiple call sites. Specify the key to be used for caching for the particular execution by passing a Context instance in to the execution: the OperationKey (prior v6: ExecutionKey) on the Context instance is the key that will be used for caching.

TResult result = await cachePolicy.ExecuteAsync(context => getFoo(), new Context("FooKey"));

The above pattern assumes you are using the default CacheKeyStrategy.

You can also specify a custom cache key strategy when configuring the policy, to vary this behaviour.

Parameters

cacheProvider

cacheProvider: The underlying cache provider to use.

CachePolicy must be used in conjunction with an ISyncCacheProvider or IAsyncCacheProvider implementation: existing providers are available via Nuget (see above) or you may implement your own.

Serializers (see below) can also be used with cache providers, to serialize execution TResult types to TCache types required by cache providers.

The same cacheProvider and serializer instances may be used across multiple call sites.

ttl

TimeSpan ttl: Time-to-live (ttl) for the cache item, as a relative, non-sliding duration from the moment the item is put in the cache.

For example, if TimeSpan.FromMinutes(5) is passed, the cacheProvider should consider the item valid for 5 minutes.

ttlStrategy (alternative to ttl above)

ITtlStrategy ttlStrategy: offers ttl strategies beyond the simple TimeSpan ttl above.

RelativeTtl

RelativeTtl(TimeSpan ttl): equivalent to ttl above.

AbsoluteTtl

AbsoluteTtl(DateTimeOffset absoluteExpirationTime): indicates that the cacheProvider should make the cached item expire at the absolute time given.

SlidingTtl

SlidingTtl(TimeSpan slidingTtl): indicates that the cacheProvider should treat the cached item as having a sliding ttl of the specified timespan. For instance, if TimeSpan.FromMinutes(5) is passed, the cacheProvider should consider the item valid for a further 5 minutes, each time the cache item is touched.

ContextualTtl

ContextualTtl: specifies that the execution should take the ttl from a property on the Context passed to execution, context[ContextualTtl.TimeSpanKey].

This allows you to define a central cache policy that will use varying ttls in different call sites, by placing the desired ttl on Polly's execution context. For example:

context[ContextualTtl.TimeSpanKey] = TimeSpan.FromMinutes(5);
context[ContextualTtl.SlidingExpirationKey] = true; // if desired; if not set, false is assumed

ResultTtl

ResultTtl: specifies a function that will be used to calculate the ttl based on the TResult item to be cached. This is useful in any scenario where the result indicates (perhaps via a header) how long the result should be cached for. An example can be where a call obtains an authorisation token and the call result also tells you how long that token is valid.

ResultTtl(Func<TResult, Ttl>): specifies a function to calculate the Ttl based on the TResult item being cached.

ResultTtl(Func<Context, TResult, Ttl>): specifies a function to calculate the Ttl based on the TResult item being cached and the execution Context.

Default cacheKeyStrategy (if none specified)

If no cacheKeyStrategy is specified, the cache key to use is taken as the OperationKey property on the execution Context, ie context.OperationKey. For example:

TResult result = await cache.ExecuteAsync(async context => await getFooAsync(), new Context("FooKey")); // "FooKey" is the cache key to use in this execution.

If context.OperationKey is not specified (no Context is passed to the execution, or context.OperationKey is not set), caching behaviour is ignored, and the underlying delegate passed to .Execute(...) (or similar) is called.

Prior v6, OperationKey was named ExecutionKey.

cacheKeyStrategy (optional)

Func<Context, string> cacheKeyStrategy: allows the specification of a custom strategy for using a more specific cache key in the execution. For instance, to cache items obtained through the execution by some guid:

// configuration
CachePolicy cache = Policy.CacheAsync(cacheProvider, TimeSpan.FromMinutes(5), context => context.OperationKey + context["guid"]);

// usage, elsewhere
Guid guid = ... // from somewhere
Context policyExecutionContext = new Context("GetResource-");
policyExecutionContext["guid"] = guid.ToString();
TResult result = await cache.ExecuteAsync(async context => await getResourceAsync(guid), policyExecutionContext); // "Resource-SomeGuid" is the key used in this execution, if guid == SomeGuid.

ICacheKeyStrategy cacheKeyStrategy: is available as a parameter in some overloads, for more complex funcs.

Working with serializers

Some cache providers (such as Redis) store items as specific types (eg string or byte[]), requiring you to serialize more complex types to those.

The following existing serializers are available for use with Polly policies:

Package	Description
`Polly.Caching.Serialization.Json` _{(nuget ; github and doco)}	A Newtonsoft.Json-based serializer for serializing any type to JSON string.

See here for notes on using serializers with Polly CachePolicy. New serializers are also easy to implement.

Interacting with policy operation

onCacheGet

An optional onCacheGet delegate allows specific code to be executed (for example for logging), when a value is retrieved from cache.

onCacheMiss

An optional onCacheMiss delegate allows specific code to be executed (for example for logging), when a cache-miss occurs (a value is not found in the cache for the given key).

onCachePut

An optional onCachePut delegate allows specific code to be executed (for example for logging), after a value has been put to the cache.

onCacheError

An optional onCacheError delegate allows specific code to be executed (for example for logging), if any call to the underlying cacheProvider throws an exception. If the onCacheError delegate is configured, it is used for both onCacheGetError and onCachePutError.

onCacheGetError

The alternative, optional onCacheGetError delegate is a more specific version of onCacheError, executed only if get calls to the underlying cacheProvider throw an exception.

onCachePutError

The alternative, optional onCachePutError delegate is a more specific version of onCacheError, executed only if put calls to the underlying cacheProvider throw an exception.

Input parameters for all above delegates

All delegates above take as input parameters the execution Context and the string cache key. Error-capturing delegates also take the Exception thrown by the cache provider.

Throws

No exceptions due to caching operations are thrown. If the underlying cacheProvider throws an exception during a cache operation:

the exception is passed to the relevant onCacheError, onCacheGetError or onCachePutError delegate, if configured.
the execution continues. For example, if the underlying cacheProvider throws while checking if the cache contained a value for the given key, the execution treats this as a cache-miss, and calls the delegate passed to .Execute(...).
- In other words, the execution intentionally swallows the exception after having passed the exception to onCacheError, onCacheGetError or onCachePutError for (say) logging. This is so that caching in itself can never bring the application down.

Usage recommendations

Placement of cache policy within `PolicyWrap`

See guidance on ordering the available policy types in a wrap. CachePolicy should be usually be placed outermost in a PolicyWrap, with only FallbackPolicy outside.

`void` executions

If an execution returning void is placed through a CachePolicy, caching operation is silently bypassed (there is no result to cache) rather than an exception thrown. This allows for a CachePolicy to be included in a PolicyWrap which might sometimes be used for TResult-returning executions, sometimes for void-returning, without exceptions being thrown.

Caching responses only selectively

There can be occasions where you want to cache only certain responses of the execution and not others. A typical case is when a CachePolicy governs executions returning HttpResponseMessage and you want to cache only when HttpResponseMessage.StatusCode == HttpStatusCode.OK.

This can be achieved by using a ResultTtl strategy which returns a Ttl representing TimeSpan.Zero as the ttl for results you do not wish to cache:

Func<Context, HttpResponseMessage, Ttl> cacheOnly200OKfilter =
    (context, result) => new Ttl(
        timeSpan: result.StatusCode == HttpStatusCode.OK ? TimeSpan.FromMinutes(5) : TimeSpan.Zero, 
        slidingExpiration: true
    );

IAsyncPolicy<HttpResponseMessage> cacheOnly200OKpolicy = 
    Policy.CacheAsync<HttpResponseMessage>(
        cacheProvider: /* the cache provider you are using */,
        ttlStrategy: new ResultTtl(cacheOnly200OKfilter),
        onCacheError: /* whatever cache error logging */
    ); // (or other richer CacheAsync overload taking an ITtlStrategy ttlStrategy)

When the ITtlStrategy returns TimeSpan.Zero, the policy skips putting that item to the cache.

Note however some cautions about caching at the HttpResponseMessage level: see the discussion: Is caching at the HttpResponseMessage level the right fit?.

Caching `default(TResult)` for value types

At Polly v6, cache policies do not cache result values of default(TResult), by analogy with an original convention from .NET Framework that a cache returning null means no value was found for that key in the cache. An undesirable side effect is that this prevents caching default(TResult) for value types (for which default(TResult) != null). At Polly v6, to ensure that you can cache default(TResult) for a value-type, change the execution type to TResult?.

From Polly v7.0.0 (with cache providers >=v3.0.0), cache policies permit caching null and default(TResult) for all value and reference types.

Thread safety and policy reuse

Thread safety

The internal operation of CachePolicy is thread-safe: multiple calls may safely be placed concurrently through a policy instance (assuming the configured cacheProvider implementation is also thread-safe).

Policy reuse

CachePolicy instances may be re-used across multiple call sites.

cacheProvider instances may be re-used across multiple CachePolicys and call sites.

Serializer instances may be re-used across multiple CachePolicys and call sites.

When reusing policies, use differing OperationKey to specify cache key (if DefaultCacheKeyStrategy is used), and to distinguish different call-site usages within logging and metrics.