Fork me on GitHub

Segment Storage Design Overview

NOTE: The information on this page applies to an older version of the TarMK and is mainly of historical interest. For the documentation of the current versions see Oak Segment Tar.

The SegmentNodeStore is an Oak storage backend that stores content as various types of records within larger segments. One or more journals are used to track the latest state of the repository. In the Tar implementation only one “root” journal is used.

The SegmentNodeStore was designed from the ground up based on the following key principles:

  • Immutability. Segments are immutable, which makes is easy to cache frequently accessed segments. This also makes it less likely for programming or system errors to cause repository inconsistencies, and simplifies features like backups or master-slave clustering.

  • Compactness. The formatting of records is optimized for size to reduce IO costs and to fit as much content in caches as possible. A node stored in SegmentNodeStore typically consumes only a fraction of the size it would as a bundle in Jackrabbit Classic.

  • Locality. Segments are written so that related records, like a node and its immediate children, usually end up stored in the same segment. This makes tree traversals very fast and avoids most cache misses for typical clients that access more than one related node per session.

This document describes the overall design of the SegmentNodeStore. See the source code and javadocs in org.apache.jackrabbit.oak.plugins.segment for full details.

Segments

The content tree and all its revisions are stored in a collection of immutable segments. Each segment is identified by a UUID and typically contains a continuous subset of the content tree, for example a node with its properties and closest child nodes. Some segments might also be used to store commonly occurring property values or other shared data. Segments can be to up to 256KiB in size.

Segments come in two types: data and bulk segments. The type of a segment is encoded in its UUID and can thus be determined already before reading the segment. The following bit patterns are used (each x represents four random bits):

  • xxxxxxxx-xxxx-4xxx-axxx-xxxxxxxxxxxx data segment UUID
  • xxxxxxxx-xxxx-4xxx-bxxx-xxxxxxxxxxxx bulk segment UUID

(This encoding makes segment UUIDs appear as syntactically valid version 4 random UUIDs specified in RFC 4122.)

Bulk segments

Bulk segments contain raw binary data, interpreted simply as a sequence of block records with no headers or other extra metadata:

[block 1] [block 2] ... [block N]

A bulk segment whose length is n bytes consists of n div 4096 block records of 4KiB each followed possibly a block record of n mod 4096 bytes, if there still are remaining bytes in the segment. The structure of a bulk segment can thus be determined based only on the segment length.

Data segments

A data segment can contain any types of records, may refer to content in other segments, and comes with a segment header that guides the parsing of the segment. The overall structure of a data segment is:

[segment header] [record 1] [record 2] ... [record N]

The header and each record is zero-padded to make their size a multiple of four bytes and to align the next record at a four-byte boundary.

The segment header consists of the following fields:

+--------+--------+--------+--------+--------+--------+--------+--------+
| magic bytes: "0aK" ASCII |version |reserved|idcount |rootcount        |
+--------+--------+--------+--------+--------+--------+--------+--------+
| blobrefcount    | reserved (set to 0)                                 |
+--------+--------+--------+--------+--------+--------+--------+--------+
| Referenced segment identifiers  (idcount x 16 bytes)                  |
|                                                                       |
|                            ......                                     |
|                                                                       |
+--------+--------+--------+--------+--------+--------+--------+--------+
| Root record references  (rootcount x 3 bytes)                         |
|                                                                       |
|                            ......          +--------+--------+--------+
|                                            |                          |
+--------+--------+--------+--------+--------+                          +
| External blob record references  (blobrefcount x 2 bytes)             |
|                                                                       |
|          ......                            +--------+--------+--------+
|                                            | padding (set to 0)       |
+--------+--------+--------+--------+--------+--------+--------+--------+

The first three bytes of a segment always contain the ASCII string “0aK”, which is intended to make the binary segment data format easily detectable. The next byte indicates the version of segment format, and is set to 10 for all segments that follow the format described here.

The idcount byte indicates how many other segments are referenced by records within this segment. The identifiers of those segments are listed starting at offset 16 of the segment header. This lookup table of up to 255 segment identifiers is used to optimize garbage collection and to avoid having to repeat the 16-byte UUIDs whenever references to records in other segments are made.

The 16-bit rootcount field indicates the number of root record references that follow after the segment identifier lookup table. The root record references are a debugging and recovery aid, that are not needed during normal operation. They identify the types and locations of those records within this segment that are not accessible by following references in other records within this segment. These root references give enough context for parsing all records within a segment without any external information. See OAK-2498.

The 16-bit blobrefcount field indicates the number of external blob record references that follow after the root record references. External blobs are binary values stored in an external data store, and the blobref list makes it possible to quickly find all such references without having to do a full traversal of all repository content. The list of active references is given to data store garbage collection so it won't collect the referenced binaries.

Segment Compaction

Segment compaction is needed when the repository size grows to an unmanageable size. The growth can happen when there are a lot of repository write operations which might create a considerable amount of content revisions that need to get garbage collected. The simplest solution is to schedule the revision garbage collection exposed as a jmx operation via the RevisionGarbageCollection MBean. This is a non-blocking operation, it will set an internal flag for compaction during the next flush call.

As a workaround in the case where the size is already too big, you might want to run a manual compaction using the Oak run tool.

Journals

Journals are special, atomically updated documents that record the state of the repository as a sequence of references to successive root node records.

A small system (like Tar) could use just a single journal and would serialize all repository updates through atomic updates of that journal. A larger system that needs more write throughput can have more journals, linked to each other in a tree hierarchy. Commits to journals in lower levels of the tree can proceed concurrently, but will need to be periodically merged back to the root journal. Potential conflicts and resulting data loss or inconsistency caused by such merges can be avoided by always committing against the root journal.

The root node references stored in journals are used as the starting point for garbage collection. It is assumed that all content currently visible to clients must be accessible through at least one of the journals. If a client wants to keep a reference to some old content revision that's no longer referenced by one of the main journals, it should create an empty private branch based on that revision and keep refreshing the branch until that content is no longer needed.

Records

The content inside a segment is divided in records of different types: blocks, lists, maps, values, templates and nodes. These record types and their internal structures are described in subsections below.

Each record is uniquely addressable by its location within the segment and the UUID of that segment. A single segment can contain up to 256KiB of data and and references to up to 256 segments (including itself). Since all records are aligned at four-byte boundaries, 16 bits are needed to address all possible record locations within a segment. Thus only three bytes are needed to store a reference to any record in any segment (1 byte to identify the segment, 2 bytes for the record offset):

+--------+--------+--------+
| refid  | offset          |
+--------+--------+--------+

The refid filed is the number of the referenced segment identifier, with refid zero interpreted as a reference to the current segment and refids 1-255 the segment identifiers stored in the lookup table in the segment header.

Block records

Blocks are binary records of up to 4kB. They're used as building blocks of large binary (or string) values and stored as-is with no extra metadata or structure. Blocks are the only record type that can't contain references to other records. Block records are typically stored in bulk segments that consist only of block records and are thus easily identifiable as containing zero references to other segments.

List records

List records are used as components of more complex record types. Lists are used for storing arrays of values for multi-valued properties and sequences of blocks for large binary values.

The list of references is split into pieces of up to 255 references each and those pieces are stored as records. If there are more than 255 pieces like that, then a higher-level list is created of references to those pieces. This process is continued until the resulting list has less than 255 entries.

+--------+--------+--------+-----+
| sub-list ID 1            | ... |
+--------+--------+--------+-----+
  |
  v
+--------+--------+--------+-----+--------+--------+--------+
| record ID 1              | ... | record ID 255            |
+--------+--------+--------+-----+--------+--------+--------+

The result is a hierarchically stored immutable list where each element can be accessed in O(log N) time and the size overhead of updating or appending list elements (and thus creating a new immutable list) is also O(log N).

Map records

Like lists, maps are components of more complex record types. Maps store unordered sets of key-value pairs of record references and are used for nodes with a large number of properties or child nodes.

Maps are stored using the hash array mapped trie (HAMT) data structure. The hash code of each key is split into pieces of 5 bits each and the keys are sorted into 32 (2^5) buckets based on the first 5 bits. If a bucket contains less than 32 entries, then it is stored directly as a list of key-value pairs. Otherwise the keys are split into sub-buckets based on the next 5 bits of their hash codes. When all buckets are stored, the list of top-level bucket references gets stored along with the total number of entries in the map.

The result is a hierarchically stored immutable map where each element can be accessed in O(log N) time and the size overhead of updating or inserting list elements is also O(log N).

TODO: Links to HAMT documentation

Value records

Value records are byte arrays used for storing all names and values of the content tree. Since item names can be thought of as name values and since all JCR and Oak values can be expressed in binary form (strings encoded in UTF-8), it is easiest to simply use that form for storing all values. The size overhead of such a form for small value types like booleans or dates is amortized by the facts that those types are used only for a minority of values in typical content trees and that repeating copies of a value can be stored just once.

There are four types of value records: small, medium, long and external. The small- and medium-sized values are stored in inline form, prepended by one or two bytes that indicate the length of the value. Long values of up to two exabytes (2^61) are stored as a list of block records. Finally an external value record contains the length of the value and a string reference (up to 4kB in length) to some external storage location.

The type of a value record is encoded in the high-order bits of the first byte of the record. These bit patterns are:

  • 0xxxxxxx: small value, length (0 - 127 bytes) encoded in 7 bits
  • 10xxxxxx: medium value length (128 - 16511 bytes) encoded in 6 + 8 bits
  • 110xxxxx: long value, length (up to 2^61 bytes) encoded in 5 + 7*8 bits
  • 1110xxxx: external value, reference string length encoded in 4 + 8 bits

Template records

A template record describes the common structure of a family of related nodes. Since the structures of most nodes in a typical content tree fall into a small set of common templates, it makes sense to store such templates separately instead of repeating that information separately for each node. For example, the property names and types as well as child node names of all nt:file nodes are typically the same. The presence of mixins and different subtypes increases the number of different templates, but they're typically still far fewer than nodes in the repository.

A template record consists of a set of up to N (exact size TBD, N ~ 256) property name and type pairs. Additionally, since nodes that are empty or contain just a single child node are most common, a template record also contains information whether the node has zero, one or many child nodes. In case of a single child node, the template also contains the name of that node. For example, the template for typical mix:versionable nt:file nodes would be (using CND-like notation):

- jcr:primaryType (NAME)
- jcr:mixinTypes (NAME) multiple
- jcr:created (DATE)
- jcr:uuid (STRING)
- jcr:versionHistory (REFERENCE)
- jcr:predecessors (REFERENCE) multiple
- jcr:baseVersion (REFERENCE)
+ jcr:content

The names used in a template are stored as separate value records and included by reference. This way multiple templates that for example all contain the “jcr:primaryType” property name don't need to repeatedly store it.

Node records

The overall structure of the content tree is stored in node records. Node records hold the actual content structure of the repository.

A typical node record consists of a template reference followed by property value references (list references for multivalued properties) and zero, one or more child node entries as indicated by the template. If the node has more than one child nodes, then those entries are stored as an array of name-node pairs of references.

A node that contains more than N properties or M child nodes (exact size TBD, M ~ 1k) is stored differently, using map records for the properties and child nodes. This way a node can become arbitrarily large and still remain reasonably efficient to access and modify. The main downside of this alternative storage layout is that the ordering of child nodes is lost.

Tar

TODO:

  • tar entry checksums
  • graph and index entries
  • recovery mechanism
  • tar generations / cleanup
  • journal.log
  • compaction
  • cleanup
  • backup
  • slow startup / journal.log