GraphDB-Enterprise Release Notes

compared with
Current by Nikola Petrov
on Apr 09, 2015 17:47.

Key
This line was removed.
This word was removed. This word was added.
This line was added.

Changes (2)

View Page History
{toc}

h1. GraphDB 6.1-SP3
This Service Pack 3 build addresses some reported problems with 6.1

h2. GraphDB Engine 6.1-b4371143

* OWLIM-1888 Better inference performance for datasets with big number of classes and onProperty restrictions
* Fixed a problem where queries containing UNION and BIND were under certain circumstances returning incorrect results
* New merge command for the storage tool
* OWLIM-1932 Count from graph http://www.ontotext.com/count does not work for describe queries
* OWLIM-1934 Removed some old and unwanted namespaces that were included by default
* OWLIM-1928 Added reversed numeric and date literal indices for better performance
* OWLIM-1990 Worker data is deleted when a plugin fails to initialize
* Using store-backed queue to keep transaction operations on the master - this should prevent out of memory errors when importing huge files


h2. GraphDB Workbench 6.4.1
- See all the release notes here: [https://confluence.ontotext.com/display/GraphDB6/GraphDB-Workbench+Release+Notes]


h2. GraphDB Connectors 3.1.4
- See all the release notes here: [https://confluence.ontotext.com/display/GraphDB6/GraphDB+Connectors+Release+notes]


h1. GraphDB 6.1-SP1
This Service Pack 1 build addresses consistency issue with the GraphDB Storage. The cluster's backup/restore functionality is also fixed.

h2. GraphDB 6.1.8410
- extended new cluster backup/restore to support multiple named backups. Important note: the backup now uses a name instead of an absolute folder
- fixes for the master URL parameter (autodetected in most cases, thus no need to specify manually)
- minor cluster improvements (shutdown on OOM during sync; worker: handling multiple initializations and no reporting of missing fingerprints on init; fixed race condition in replication; new internal URIs (not URLs) for replication)
- fixed: [owlim-1853] LoadRDF under Windows sometimes does not include statements in the PSO index and the indexes are incosistent;
- Added a check/warning to see if there is enough memory for the entity pool when it gets restored from persistence (-Xmx - cache-memory should be >= entity pool size * 1.25, i.e. 25% overhead is left for the datatype index and other in-memory structures included in the entity pool and for other purposes, e.g. for running queries)
- Constraint validation and support for multiple rulesets is out of its experimental stage and now in production. See the docs here: [https://confluence.ontotext.com/display/GraphDB6/GraphDB+Constraint+Validation]

h2. GraphDB Workbench 6.3.2
- See all the release notes here: [https://confluence.ontotext.com/display/GraphDB6/GraphDB-Workbench+Release+Notes]


h1. GraphDB - 6.1.8316
This is an Enterprise release only.

h2. Improvements (replication cluster):
- Tx Log stability improvements & fixes - scrapped optimistic join procedure; scrapped shared horizon & completed queue; fixed status patching base
- Improved logging detail
- Updates are dispatched to the other peers immediately once accepted and enqueued
- dedicated peer dispatch threads
- fixed split-brain recovery
- Client API: Changed the master retry policy (do not check every master too often; but still make sure that when previously unavailable master becomes available we detect that)

h2. Replication improvements:
- cleanup target directory on accepting replication data
- 60s timeout in replication client
- limited retries in the replication server
- replication client startup delay patch
- retry serving replication data on failure
- exposed exceptions thrown from the replication server

h2. Other improvements:
- The inferencer debug statistics were not displayed at shutdown because we used the SwitchableInferencer's ones instead of those in currentInferencer that did the job
- Created StorageTool app, used to Scan/Rebuild indexes in case of missing statements

h3. Fixes:
- [OWLIM-1838] Custom NTriples/NQuads parser: ArrayOutOfBoundsException can be thrown on empty lines while parsing and then NPE while trying to process t.getMessage() which may be null
- [OWLIM-1834] Fixed: LoadRDF Tool doesn't work with GraphDB Enterprise
- [W-44] Introduced a parameter 'in-clause-max-members' (defaults to 16) which limits the number of entities specified in the IN clause. If the IN clause contains more elements then it won't be optimized to a UNION query
- [W-55] Fixed the ASC/DESC issue in ORDER BY clause when the variable in the ORDER BY is in an OPTIONAL clause (the issue was different number of results returned in ASC/DESC cases).


h1. GraphDB version 6.1-RC1


- *Much faster write transactions* for small insert/update/delete operations on large repositories. Results on LDBC [Semantic Publishing Benchmark|http://ldbcouncil.org/developer/spb] (SPB) at 50M went up from 32 read and 12 write queries per second in ver. 6.0 to 40 reads/s and 31 writes/s in ver. 6.1. The improvement gets even more visible and SPB at 1B scale: from 10 reads/s and 2 writes/s in ver. 6.0 to 11 reads/s and 10 writes/s in ver. 6.1. In summary, GraphDB 6.1 is able to handle twice more updates at 50M scale and 5 times more updates at scale of 1 billion statements. This way GraphDB 6.1 is already capable to deal with true Dynamic Semantic Publishing scenario, like the one of [BBC|http://www.bbc.co.uk/blogs/legacy/bbcinternet/2012/04/sports_dynamic_semantic.html], at a scale of 1 billion statements and higher.
See more: http://www.ontotext.com/graphdb-benchmark-results/

- *Improved load capabilities for large new datasets in live databases* instances. Scenario description: there is a production cluster with average load - running LDBC-50m (SPB) with 4 reading threads (doing select queries) and 1 writing thread (doing update queries). We need to add a large dataset (e.g. DBPedia) with hundreds of millions statements -- as fast as possible, but without disrupting the overall cluster speed too much (not introducing write latency of more than 1-2s). The data set doesn't need inference, so it is loaded with the empty rule set.