Presentation with comments (PPT Version)

Report
APACHE SLING & FRIENDS TECH MEETUP
BERLIN, 22-24 SEPTEMBER 2014
Oak, the Architecture of the new Repository
Michael Dürig, Adobe Research
Design goals





adaptTo() 2014
Scalable
Big repositories
Clustering
Customisable, flexible
OSGi friendly
2
Outline
 CRUD
 Changes
 Search
adaptTo() 2014
3
Tree model
a
b
adaptTo() 2014
d
c
4
Updating
?
a
b
adaptTo() 2014
d
c
x
5
MVCC
HEAD
r1: /
a
r2: /
d
r1: /d
b
r1: /a/b
r2: /a/b
adaptTo() 2014
r2: /d
c
r2: /d/x
6
Refresh and Garbage Collection
adaptTo() 2014
7
Refresh
garbage
adaptTo() 2014
8
Garbage collection
garbage
adaptTo() 2014
9
Concurrency and Conflicts
adaptTo() 2014
10
Concurrent updates
r2a
adaptTo() 2014
r1
r2b
11
Merging
r2a
merge
upates
r1
r3
r2b
adaptTo() 2014
12
Conflict handling: serialisation
 Fully serialised
 Fail, no concurrent update
 Partially serialised
 Concurrent conflict free updates
adaptTo() 2014
13
Conflict handling strategies: merging
 Partial merge
 Conflict markers, deferred resolution
 Full merge
 Need to choose victim
adaptTo() 2014
14
Replicas and Sharding
adaptTo() 2014
15
Replica and caches
master copy
adaptTo() 2014
full replica
cache
16
Sharding strategies
by path
adaptTo() 2014
by level
by hash
with caching
17
Implementations
adaptTo() 2014
18
MicroKernel / NodeStore
 Tree / Revision model implementation
Responsible for
Not responsible for
Clustering
Validation
Sharding
Access control
Caching
Search
Conflict handling
Versioning
adaptTo() 2014
19
Current implementations
DocumentMK
TarMK (SegmentMK)
Persistence
MongoDB, JDBC
Local FS
Conflict handling
Partial serialisation
Full serialisation
Clustering
MongoDB clustering
Simple failover
Sharding
MongoDB sharding
N/A
Node Performance
Moderate
High
Key use cases
Large deployments (>1TB), concurrent
Small/medium deployments, mostly read
deployments, mostly read
adaptTo() 2014
20
Access Control
adaptTo() 2014
21
Accessible paths
a
b
adaptTo() 2014
d
c
22
xistentialism
 All paths traversable
 Node may not exist
 Decorator on NodeStore
root.getChildNode("a").exists();
⟹ false
root.getChildNode("a")
.getChildNode("b").exists();
⟹ true
adaptTo() 2014
23
Comparing Revisions
adaptTo() 2014
24
Content diff
 What changed between trees
 Cornerstone for




adaptTo() 2014
Validation
Indexing
Observation
…
25
What changed?
∆
adaptTo() 2014
26
Example: merging
r2a
r1
r2b
adaptTo() 2014
∆ r2a
r1 ➞
“a” modified
“b” removed
r3
r1 ➞∆ r2b
“d” modified
“x” added
27
Commit Hooks
adaptTo() 2014
28
Commit hooks
 Key plugin mechanism
 Higher level functionality



adaptTo() 2014
Validation (node type, access control, …)
Trigger (auto create, defaults, …)
Updates (index, …)
29
Editing a commit
∆
adaptTo() 2014
∆+x
30
Commit hooks
 Based on content diff
 pass a commit
 fail a commit
 edit a commit
 Applied in sequence
adaptTo() 2014
31
Type of hooks
CommitHook
Editor
Validator
Content diff
Optional
Always
Always
Can modify
Yes
Yes
No
Programming model
Simple
Callbacks
Callbacks
Performance impact
High
Medium
Low
adaptTo() 2014
32
Observers
adaptTo() 2014
33
Observers
 Observe changes




After commit
Often does a content diff
Asynchronous
Optionally synchronous

adaptTo() 2014
Local cluster node only
34
Examples




JCR observation
External index update
Cache invalidation
Logging
adaptTo() 2014
35
Search
adaptTo() 2014
36
Query Engine
SELECT
WHERE x=y
parse
execute
Parse
r
Index
Parser
Index
post process
/a//*
Parse
r
adaptTo() 2014
Parser
Index
Travers
e
37
Index Implementations
 Property (ordered)
 Reference
 Lucene
 In-content or file system
 Solr
 Embedded or external
adaptTo() 2014
38
Big Picture
adaptTo() 2014
39
Big picture
JCR API
Oak JCR
Oak API
Plugins
Oak Core
NodeStore API
MicroKernel
adaptTo() 2014
40
Resources
http://jackrabbit.apache.org/oak/
adaptTo() 2014
41
Appendix
adaptTo() 2014
42
Resources
http://jackrabbit.apache.org/oak/
http://jackrabbit.apache.org/oak/docs/
https://svn.apache.org/repos/asf/jackrabbit/oak/trunk/
adaptTo() 2014
43

similar documents