Merge pull request #3089 from vespa-engine/balder/flip-visit-defaults-for-visiting

Flip defaults to optimize for streaming search by default.

3 files changed, 18 insertions, 11 deletions

diff --git a/searchcore/src/vespa/searchcore/config/proton.def b/searchcore/src/vespa/searchcore/config/proton.def
index 0c4160640eb..ac17efcab3f 100644
--- a/searchcore/src/vespa/searchcore/config/proton.def
+++ b/searchcore/src/vespa/searchcore/config/proton.def
@@ -202,7 +202,7 @@ summary.cache.maxbytes long default=0 restart
 
 ## Include visits in the cache, if the visitoperation allows it.
 ## This will enable another separate cache of summary.cache.maxbytes size.
-summary.cache.allowvisitcaching bool default=false restart
+summary.cache.allowvisitcaching bool default=true restart
 
 ## Control number of cache entries preallocated.
 ## Default is no preallocation
@@ -213,20 +213,26 @@ summary.cache.initialentries long default=0 restart
 summary.cache.compression.type enum {NONE, LZ4, ZSTD} default=LZ4 restart
 
 ## Control compression level of the summary while in cache.
+## LZ4 has normal range 1..9 while ZSTD has range 1..19
+## 9 is a reasonable default for both
 summary.cache.compression.level int default=9 restart
 
 ## Control compression type of the summary while in memory during compaction
 ## NB So far only stragey=LOG honours it.
-summary.log.compact.compression.type enum {NONE, LZ4, ZSTD} default=LZ4 restart
+summary.log.compact.compression.type enum {NONE, LZ4, ZSTD} default=ZSTD restart
 
 ## Control compression level of the summary while in memory during compaction
+## LZ4 has normal range 1..9 while ZSTD has range 1..19
+## 9 is a reasonable default for both
 summary.log.compact.compression.level int default=9 restart
 
 ## Control compression type of the summary
-## NB So far only stragey=LOG honours it.
-summary.log.chunk.compression.type enum {NONE, LZ4, ZSTD} default=LZ4 restart
+summary.log.chunk.compression.type enum {NONE, LZ4, ZSTD} default=ZSTD restart
 
 ## Control compression level of the summary
+## LZ4 has normal range 1..9 while ZSTD has range 1..19
+## 9 is a reasonable default for both. Going above for ZSTD can give an improvement,
+## but is better done in conjunction with increasing chunk size.
 summary.log.chunk.compression.level int default=9 restart
 
 ## Max size in bytes per chunk.
@@ -366,12 +372,12 @@ maxvisibilitydelay double default=1.0
 ## You can set this to a number above zero for visit to shortcut expensive serialize size computation.
 ## This value will be provided instead.
 ## negative number will compute it accurately.
-## Default will be flipped shortly. (Once image personal index has gone live.)
-visit.defaultserializedsize long default=-1
+## This is only used for weakly consistent visiting, like streaming search.
+visit.defaultserializedsize long default=1
 
 ## This will ignore the maxbytes limit advised from above.
-## default will be flipped shortly. (Once image personal index has gone live.)
-visit.ignoremaxbytes bool default=false
+## This is only used for weakly consistent visiting, like streaming search.
+visit.ignoremaxbytes bool default=true
 
 ## Number of initializer threads used for loading structures from disk at proton startup.
 ## The threads are shared between document databases when value is larger than 0.
diff --git a/searchcore/src/vespa/searchcore/proton/persistenceengine/document_iterator.cpp b/searchcore/src/vespa/searchcore/proton/persistenceengine/document_iterator.cpp
index 5050864e25c..8260c0209a9 100644
--- a/searchcore/src/vespa/searchcore/proton/persistenceengine/document_iterator.cpp
+++ b/searchcore/src/vespa/searchcore/proton/persistenceengine/document_iterator.cpp
@@ -80,10 +80,10 @@ DocumentIterator::DocumentIterator(const storage::spi::Bucket &bucket,
       _selection(selection),
       _versions(versions),
       _fields(fields.clone()),
-      _defaultSerializedSize(defaultSerializedSize),
+      _defaultSerializedSize((readConsistency == ReadConsistency::WEAK) ? defaultSerializedSize : -1),
       _readConsistency(readConsistency),
       _metaOnly(fields.getType() == document::FieldSet::NONE),
-      _ignoreMaxBytes(ignoreMaxBytes),
+      _ignoreMaxBytes((readConsistency == ReadConsistency::WEAK) && ignoreMaxBytes),
       _fetchedData(false),
       _sources(),
       _nextItem(0),
@@ -277,7 +277,7 @@ DocumentIterator::fetchCompleteSource(const IDocumentRetriever & source, Iterate
         }
     } else {
         MatchVisitor visitor(matcher, metaData, lidIndexMap, _fields.get(), list, _defaultSerializedSize);
-        visitor.allowVisitCaching(_readConsistency == ReadConsistency::WEAK);
+        visitor.allowVisitCaching(isWeakRead());
         source.visitDocuments(lidsToFetch, visitor, _readConsistency);
     }
 
diff --git a/searchcore/src/vespa/searchcore/proton/persistenceengine/document_iterator.h b/searchcore/src/vespa/searchcore/proton/persistenceengine/document_iterator.h
index 1ceee910d91..2b2679a12ee 100644
--- a/searchcore/src/vespa/searchcore/proton/persistenceengine/document_iterator.h
+++ b/searchcore/src/vespa/searchcore/proton/persistenceengine/document_iterator.h
@@ -36,6 +36,7 @@ private:
     bool checkDoc(const document::Document &doc) const;
     bool checkDoc(const SelectContext &sc) const;
     void fetchCompleteSource(const IDocumentRetriever & source, storage::spi::IterateResult::List & list);
+    bool isWeakRead() const { return _readConsistency == ReadConsistency::WEAK; }
 
 public:
     DocumentIterator(const storage::spi::Bucket &bucket,