Interface MapStore<K,V>
- Type Parameters:
K- type of the MapStore keyV- type of the MapStore value
- All Superinterfaces:
MapLoader<K,V>
- All Known Subinterfaces:
EntryStore<K,V>
- All Known Implementing Classes:
GenericMapStore,MapStoreAdapter
IMap.put(key, value) normally stores the entry into JVM's memory. If the MapStore implementation is provided then Hazelcast will also call the MapStore implementation to store the entry into a user-defined storage, such as RDBMS or some other external storage system. It is completely up to the user how the key-value will be stored or deleted.
Same goes for IMap.remove(key).
Implementations must assume:
- Operations may be retried, reordered, or replayed after failures or migrations.
- Batch methods (
storeAll/deleteAll) may partially succeed and be retried as per-entry operations. - At-least-once delivery semantics; exactly-once or atomic multi-row guarantees are not provided.
Implementations must therefore ensure:
- Idempotent writes and deletes.
- Tolerance to partial failure and replay.
- No reliance on global ordering, transactional isolation, or side effects.
MapStore is not a durability layer in the database sense.
It exists to make in-memory state recoverable, not to provide transactional persistence.
-
Method Summary
Modifier and TypeMethodDescriptionvoidDeletes the entry with the given key from the external store.voiddeleteAll(Collection<K> keys) Deletes multiple entries from the external store.voidStores the key-value pair in the external store.voidStores multiple entries in the external store.Methods inherited from interface com.hazelcast.map.MapLoader
load, loadAll, loadAllKeys
-
Method Details
-
store
Stores the key-value pair in the external store.Invocation semantics depend on configuration:
- Write-through (
writeDelaySeconds == 0): this method is invoked synchronously as part of theIMap.put()call. If this method throws an exception, the map operation fails. - Write-behind (
writeDelaySeconds > 0): this method may be invoked asynchronously on a background thread, potentially long after the original map operation has completed.
Under normal circumstances Hazelcast provides at-least-once delivery semantics for calls to this method. Implementations must therefore be idempotent and tolerate retries, reordering, and duplicate invocations.
- Parameters:
key- key of the entry to storevalue- value of the entry to store
- Write-through (
-
storeAll
Stores multiple entries in the external store.This method is primarily used with write-behind enabled (
writeDelaySeconds > 0) when batching is applicable. Hazelcast may still invokestore(Object, Object)instead of this method depending on runtime conditions (for example, batch size, write coalescing, or number of distinct keys in the current write window). Implementations must not rely on this method being invoked for every flush.Error handling and retries:
- If this method throws an exception, Hazelcast treats the batch as failed and starts retry handling.
- If the provided
maphas not been modified (no entries removed to signal partial success), Hazelcast retriesstoreAll(map)up to three times, waiting about 1 second between attempts. - After the final batch attempt fails, Hazelcast falls back to calling
store(Object, Object)for the remaining entries, one by one. - If the implementation removes entries from
mapas they are stored (before throwing), Hazelcast treats removed entries as successful and does not retry them; only the entries still present inmapare retried and/or passed to the per-entry fallback.
This allows implementations to signal partial success explicitly. Entries left in the
mapafter an exception will be retried individually; removed entries will not be passed to subsequent calls.Hazelcast does not provide transactional guarantees across entries. Batch boundaries do not imply atomicity.
- Parameters:
map- map of entries to store
-
delete
Deletes the entry with the given key from the external store.Invocation semantics depend on configuration:
- Write-through: invoked synchronously as part of
IMap.remove(). - Write-behind: invoked asynchronously on a background thread.
Hazelcast provides at-least-once delivery semantics. Implementations must tolerate retries and duplicate invocations. Deleting a non-existent entry should be treated as a successful no-op.
- Parameters:
key- the key to delete from the store
- Write-through: invoked synchronously as part of
-
deleteAll
Deletes multiple entries from the external store.This method is primarily used with write-behind enabled (
writeDelaySeconds > 0) when batching is applicable. Hazelcast may still invokedelete(Object)instead of this method depending on configuration and runtime conditions. Implementations must not rely on this method being invoked for every flush.Error handling and retries:
- If this method throws an exception, Hazelcast initiates retry handling.
- On retry, Hazelcast invokes
delete(Object)one-by-one for the remaining keys. - If the implementation removes successfully deleted keys from the
provided
keyscollection before throwing, those keys are treated as successful and are not retried.
This allows implementations to signal partial success explicitly. Keys left in the
keyscollection after an exception will be retried individually; removed keys will not be passed to subsequent calls.Batch deletion does not imply atomicity or ordering guarantees.
- Parameters:
keys- the keys of the entries to delete
-