Class RedisSnapshotStore
- Namespace
- Excalibur.EventSourcing.Redis
- Assembly
- Excalibur.EventSourcing.Redis.dll
Redis Hash-based implementation of ISnapshotStore.
public sealed class RedisSnapshotStore : ISnapshotStore
- Inheritance
-
RedisSnapshotStore
- Implements
- Inherited Members
Remarks
Stores snapshots in Redis Hash keys: snap:t:{tenantId}:{aggregateType}:{aggregateId}. The
tenant segment is always present — a single-tenant host resolves the framework's single-tenant
identity, never an omitted segment — so the key shape does not vary between single- and multi-tenant
hosts.
Each hash contains the snapshot data, version, and metadata.
Only the latest snapshot is stored per aggregate and tenant. A save replaces the stored snapshot
only when it carries a HIGHER version, so the stored snapshot never moves backwards: a save at an
older version is refused rather than applied, and a save at the same version is a no-op. Saves may
therefore arrive concurrently and in any order without the reader observing a snapshot that is not
the latest one written.
Constructors
RedisSnapshotStore(ConnectionMultiplexer, IOptions<RedisSnapshotStoreOptions>, ILogger<RedisSnapshotStore>, ITenantContext)
Initializes a new instance of the RedisSnapshotStore class.
public RedisSnapshotStore(ConnectionMultiplexer connection, IOptions<RedisSnapshotStoreOptions> options, ILogger<RedisSnapshotStore> logger, ITenantContext tenantContext)
Parameters
connectionConnectionMultiplexerThe Redis connection multiplexer.
optionsIOptions<RedisSnapshotStoreOptions>The snapshot store options.
loggerILogger<RedisSnapshotStore>The logger instance.
tenantContextITenantContextThe ambient tenant context. Required: this store partitions rows by tenant, and it resolves that partition from here, so there is no state in which the partition is undecided. A single-tenant host receives the framework default context and operates as the one canonical tenant.
Methods
DeleteSnapshotsAsync(string, string, CancellationToken)
Deletes all snapshots for an aggregate.
public ValueTask DeleteSnapshotsAsync(string aggregateId, string aggregateType, CancellationToken cancellationToken)
Parameters
aggregateIdstringThe aggregate identifier.
aggregateTypestringThe aggregate type name.
cancellationTokenCancellationTokenCancellation token.
Returns
- ValueTask
A task that represents the asynchronous delete operation.
DeleteSnapshotsOlderThanAsync(string, string, long, CancellationToken)
Deletes snapshots older than a specified version.
public ValueTask DeleteSnapshotsOlderThanAsync(string aggregateId, string aggregateType, long olderThanVersion, CancellationToken cancellationToken)
Parameters
aggregateIdstringThe aggregate identifier.
aggregateTypestringThe aggregate type name.
olderThanVersionlongDelete snapshots with version less than this value.
cancellationTokenCancellationTokenCancellation token.
Returns
- ValueTask
A task that represents the asynchronous delete operation.
GetLatestSnapshotAsync(string, string, CancellationToken)
Gets the latest snapshot for an aggregate.
public ValueTask<ISnapshot?> GetLatestSnapshotAsync(string aggregateId, string aggregateType, CancellationToken cancellationToken)
Parameters
aggregateIdstringThe aggregate identifier.
aggregateTypestringThe aggregate type name.
cancellationTokenCancellationTokenCancellation token.
Returns
Remarks
Confined to the ambient tenant established for this store instance: a snapshot stored under
another tenant for this aggregateId is reported as not found, the same as one
that was never taken.
SaveSnapshotAsync(ISnapshot, CancellationToken)
Saves a snapshot for an aggregate, keeping the highest version stored for it.
public ValueTask SaveSnapshotAsync(ISnapshot snapshot, CancellationToken cancellationToken)
Parameters
snapshotISnapshotThe snapshot to save.
cancellationTokenCancellationTokenCancellation token.
Returns
- ValueTask
A task that represents the asynchronous save operation.
Remarks
Total. Every well-formed snapshot is an acceptable argument. There is no version, payload size, or identifier shape a caller must avoid, and no state of the store in which a save is illegal.
Monotone upsert, scoped to the interval since the last deletion. After
SaveSnapshotAsync returns normally, the version readable for
(aggregateId, aggregateType) within the ambient tenant is at least the version passed. A
save carrying a version lower than or equal to the stored one is a successful no-op. Monotonicity
does NOT hold across a deletion: DeleteSnapshotsAsync,
DeleteSnapshotsOlderThanAsync and provider-level expiry all remove stored versions, after
which a save may legitimately make a LOWER version readable than was readable before. Saves and
deletions on the same key are not ordered with respect to one another, so a save superseded by a
concurrent deletion has not dropped a snapshot. A caller that requires a specific version to be
readable must re-read it; a normal return does not entitle the caller to infer which version is
readable now, only that the store established one at least as high at the moment it completed. A
store that returns normally having neither stored the snapshot nor established that is in
violation; unresolvable contention is reported as ConcurrencyException, never as success.
Modifies only its own key. No other aggregate, aggregate type, or tenant is observably affected.
Faults. Throws only for an invalid argument, for cancellation, or for an infrastructure failure that prevented the store from establishing the outcome above. Implementations MUST NOT report a lost race as a fault, and MUST NOT return normally having neither stored the snapshot nor established that a version at least as high is stored — a silent drop reports success for a snapshot nobody holds.
Exceptions
- ConcurrencyException
May be thrown by an implementation that retries a contended write when it exhausts its attempts without either storing the snapshot or establishing that a version at least as high is already stored. This is the infrastructure-failure case above, reported in the abstraction's own currency: the outcome is UNKNOWN, which is why it is a fault rather than the successful no-op that a proven supersede produces. Implementations that write without retrying cannot reach this state and do not throw it.