link:../src/crux/db.clj[role=include]Liking cljdoc? Tell your friends :D
Design
Philosophy
Crux is an unbundled database, where parts are pluggable and can be swapped out for alternative implementations. We are attempting to follow the Unix philosophy of each part doing one thing well.
This document walks through some of the high-level Crux component.
ObjectStore
Crux contains an ObjectStore for storing and retreiving documents.
The main implementation is KvObjectStore wrapped by
CachedObjectstore.
| Objects are stored against used their hashed value. |
The KvObjectStore wraps the lower-level KvStore
|
KvStore
link:../src/crux/kv.clj[role=include]The KvStore exposes the operations Crux needs to use a Key Value store for
indexing purposes, and for storing Objects.
Implementations of the KvStore include:
| Implementation | Description |
|---|---|
| Uses RocksDB and the standard Java API that ships with RocksDB |
| Uses RocksDB and a custom built JNR bridge |
| Uses LMBD via |
| Uses LMDB via |
| An in-memory KvStore |
TxLog
Crux uses a replayable transaction log that is used to derive indicies.
link:../src/crux/db.clj[role=include]The transaction log is used to write both transaction and documents to an appendable transaction log.
| Implementation | Description |
|---|---|
| (default), using Kafka with separate Kafka topics for both documents and transactions |
| Uses the KvStore as the transaction log during testing only |
| Uses the KvStore for the transaction log in standalone mode, backed by |
Indexer
The indexer indexes documents and transactions for query purposes. This might be on the back of subscriptions to Kafka topics, or by direct calls made in a single node topological setup.
link:../src/crux/db.clj[role=include]The single implememention is crux.tx.KvIndexer, which makes use of
the KvStore to persist indices.
Query Design
Index
Crux has the fundamental notion of an index.
link:../src/crux/db.clj[role=include]The two operations are seek and next.
Layered Index
The layered index exists to faciliate the idea of navigating up and down an index, in a tree like manner.
link:../src/crux/db.clj[role=include]For example the index attribute+value+entity+content-hash is the following tree:
digraph G
{
attribute->value
value->entity
entity->content
}
open-level gives instructions to open and move down a level. In the
above example if could be moving the index down to point at the values
within a given attribute. That is to say that if we have an attribute
:name, the index will iterate across all values for that attribute,
until there are no more name values.
close-level moves the index back-up, so in the above example, we can
iterate at the higher level of attribute.
Virtual Index
A Virtual Index comprises together multiple child indices. This is to join indices together, returning key/value pairs on where they match.
A join condition in a query could be reflected by a Virtual Index. A Virtual Index will maintain state as to where Index is currently positioned.
UnaryJoinVirtualIndex comprises of multiple child indices and
implements both Index and LayeredIndex. Calling seek-values on
it will advance all the child indices internally until they all
contain the same key. This involves calling seek-values on each
child index until the indices match at the same level. Calling
next-values would move all the indices along until the next common
key that all the indices share.
| 0 | 1 | 3 | 4 | 5 | 6 | |
| 0 | 2 | 3 | 5 | |||
| 2 | 3 | 4 | 5 | 6 |
In the above example, where UnaryJoinVirtualIndex joins three
RelationVirtualIndexs (a,b,c). Calling seek-values would return 3
as the first value found.
Calling next-values would jump ahead to the value 5.
Can you improve this documentation? These fine people already did:
Jon Pither, Jeremy Taylor & Håkan RåbergEdit on GitHub
cljdoc is a website building & hosting documentation for Clojure/Script libraries
× close