- added some missing increments from RWI results
- decrement relevant navigator counts when solr or RWI results are
evicted because duplicates detection or constraints checked belatedly
- do not compute facets when unnecessary to avoid unwanted CPU load
- do not increment from facets when already done
- do not rely on facets on remote solr peers requests, as most of the
time only a limited part of their total results if fetched (thus also
preventing unnecessary load on remote peers)
- use a concurrency friendly score map for the dates navigators to
prevent unwanted ConcurrentModificationExceptions
This improves the situation for the most obvious inconsistencies in
search navigators counts, but more has to be done for a true accuracy
(notably when query modifiers constraints are applied belatedly - after
the solr or RWI retrieval request - such as the content domain
constraint)
used for rwi ranking.
Main changes:
- introduce a posintext() to access the stored value. This reduces also mem alloc of position array for WordReferenceRow (index access)
- use the positions() array for joined references on multi-word queries if needed (otherwise allow positions() to be null
- adjust assignments and the min() max() and distance() calculation accordingly
Issue was the calculation in AbstractReference with positions.clear() call,
this made distance result always 0 (distance needs min 2 positions) and created concurrency issues.
+ unit test of changes
- correct WordReferenceVars.toRowEntry posintext parameter
to set expected min posintext (the difference is on multi-word queries,
while positions are ordered by search word order).
- modified posofphrase/posinphrase join operation
- to set min posofphrase
- and keep posinphrase if not same posofphrase (was set to 0, no differentiation during ranking)
+ fix compiler msg (missing type declaration)
Purpose of the test case is to be able to (controlled) analyse the rwi ranking for
multi word searches (with focus on posintext and word-distance ranking)
including a small change, word posintext counting.
We remember/store 1st posintext. Previously following words got a handle (posintext)
excluding found. Now it just counts and assigns true posintext as handle (posintext)
This is needed and enables existing word position ranking for RWI.
The upcoming concurrency issue in word position min/max calculation were eliminated
by iterator.hasHext check before next() access.
instead of TreeMaps)
- enhanced memory footprint of database indexes (by introduction of
optimize calls)
- optimize calls shrink the amount of used memory for index sets if they
are not changed afterwards any more
- refactored all code which uses URIMetadataRow as standard for word
hash length and word hash ordering and moved that to the class 'Word',
becuase the class URIMetadataRow defined the old metadata data structure
and should be superfluous in the future
- removed unused methods from URIMetadataRow as preparation for further
removal of that class
all unique links! This made it necessary, that a large portion of the
parser and link processing classes must be adopted to carry a different
type of link collection which carry a property attribute which are
attached to web anchors.
- introduction of a new URL class, AnchorURL
- the other url classes, DigestURI and MultiProtocolURI had been renamed
and refactored to fit into a new document package schema, document.id
- cleanup of net.yacy.cora.document package and refactoring
jdk-based logger tend to block
at java.util.logging.Logger.log(Logger.java:476) in concurrent
environments. This makes logging a main performance issue. To overcome
this problem, this is a add-on to jdk logging to put log entries on a
concurrent message queue and log the messages one by one using a
separate process.
- FTPClient uses the concurrent logging instead of the log4j logger
adjusted to smaller and 1-core devices.
- the workflow processor now starts no process at all. these are started
as soon as parser/condenser/indexing queues are filled.
- better abstraction
superfluous. The target is to make a solr document as the core of YaCy
documents which would cause that many conversions can be removed. On the
way to this target the Equivalence of URIMetadataRow and URIMetadataNode
had to be removed to expose the usage of the old URIMetadataRow data
structure.
This refactoring already removes unneccessary conversions and should
make memory usage during indexing lower.
which had been then mixed with remote RWIs. Now these Solr documents are
feeded into the result set as they appear during local and remote
search. That makes the search much faster.
URIMetadataNode which creates the opportunity to access Solr objects
directly and use their information richness
- lazy initialization of the URIMetadataNode object - should cause less
computation and memory usage during search.
- removed dead code
MultiProtocolURI during normalform computation because that should
always be done and also be done during initialization of the
MultiProtocolURI Object. The new normalform method takes only one
argument which should be 'true' unless you know exactly what you are
doing.
metadata representation from the solr index. This shall replace metadata
from the built-in database in the future.
- added the Solr-driven metadata into the search index of YaCy which
makes it now possible to run YaCy without the old metadata index. This
is a major stept forward to a full migration to Solr.
- added log warnings in case that search processes run into time-out
situations
- better concurrency for Integer formatter (used a non-synchronized
formatter before)
- bugfix for search termination (a poison pill was missing)
- added timeout parameters for search (again) -> target is, that they
are never reached.
luccioman