From a9ad8636866dc1957a2fc295df1adce99899b48e Mon Sep 17 00:00:00 2001 From: orbiter Date: Mon, 16 Feb 2009 16:13:48 +0000 Subject: [PATCH] second part of 'doubles' fix - better handling of doubles in RAMIndex. More logging. still missing: deletion of double entries in collections git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@5613 6c8d7289-2bf4-0310-a012-ef5d649a1542 --- source/de/anomic/index/indexRI.java | 1 + source/de/anomic/kelondro/index/RAMIndex.java | 12 +++- .../kelondro/kelondroCollectionIndex.java | 2 +- source/de/anomic/kelondro/table/EcoTable.java | 71 ++++++++++--------- 4 files changed, 48 insertions(+), 38 deletions(-) diff --git a/source/de/anomic/index/indexRI.java b/source/de/anomic/index/indexRI.java index 5c4ee60a1..896a0aeef 100644 --- a/source/de/anomic/index/indexRI.java +++ b/source/de/anomic/index/indexRI.java @@ -44,6 +44,7 @@ public interface indexRI { public boolean removeEntry(String wordHash, String urlHash) throws IOException; public int removeEntries(String wordHash, Set urlHashes) throws IOException; public void addEntries(indexContainer newEntries) throws IOException; + public int sizeEntry(final String key); public void clear() throws IOException; public void close(); diff --git a/source/de/anomic/kelondro/index/RAMIndex.java b/source/de/anomic/kelondro/index/RAMIndex.java index ced332feb..aa3b8d5cf 100644 --- a/source/de/anomic/kelondro/index/RAMIndex.java +++ b/source/de/anomic/kelondro/index/RAMIndex.java @@ -128,9 +128,15 @@ public class RAMIndex implements ObjectIndex { public synchronized ArrayList removeDoubles() { // finish initialization phase explicitely - if (index1 == null) index1 = new RowSet(rowdef, 0); - index0.sort(); - return index0.removeDoubles(); + index0.sort(); + if (index1 == null) { + return index0.removeDoubles(); + } + index1.sort(); + ArrayList d0 = index0.removeDoubles(); + ArrayList d1 = index1.removeDoubles(); + d0.addAll(d1); + return d0; } public synchronized Row.Entry remove(final byte[] key) { diff --git a/source/de/anomic/kelondro/kelondroCollectionIndex.java b/source/de/anomic/kelondro/kelondroCollectionIndex.java index 4b084fb16..7403f8d4a 100644 --- a/source/de/anomic/kelondro/kelondroCollectionIndex.java +++ b/source/de/anomic/kelondro/kelondroCollectionIndex.java @@ -263,7 +263,7 @@ public class kelondroCollectionIndex { Row.Entry maxentry; int doublecount = 0; ArrayList doubles = index.removeDoubles(); - if (doubles.size() > 0) Log.logWarning("COLLECTION INDEX STARTUP", "found " + doubles + " doubles in collections, removing them in arrays"); + if (doubles.size() > 0) Log.logWarning("COLLECTION INDEX STARTUP", "found " + doubles.size() + " doubles in collections, removing them in arrays"); for (final RowCollection doubleset: doubles) { // for each entry in doubleset choose one which we want to keep maxentry = null; diff --git a/source/de/anomic/kelondro/table/EcoTable.java b/source/de/anomic/kelondro/table/EcoTable.java index d17b887e0..b8ebf76d6 100644 --- a/source/de/anomic/kelondro/table/EcoTable.java +++ b/source/de/anomic/kelondro/table/EcoTable.java @@ -96,8 +96,10 @@ public class EcoTable implements ObjectIndex { this.taildef = new Row(cols, NaturalOrder.naturalOrder, -1); // initialize table file + boolean freshFile = false; if (!tablefile.exists()) { // make new file + freshFile = true; FileOutputStream fos = null; try { fos = new FileOutputStream(tablefile); @@ -167,43 +169,40 @@ public class EcoTable implements ObjectIndex { } } - // check consistency - //System.out.print(" -ordering- .."); - //System.out.flush(); + // open the file this.file = new BufferedEcoFS(new EcoFS(tablefile, rowdef.objectsize), this.buffersize); - final ArrayList doubles = index.removeDoubles(); - //assert index.size() + doubles.size() + fail == i; - //System.out.println(" -removed " + doubles.size() + " doubles- done."); - if (doubles.size() > 0) { - Log.logInfo("ECOTABLE", tablefile + ": WARNING - EcoTable " + tablefile + " has " + doubles.size() + " doubles"); - // from all the doubles take one, put it back to the index and remove the others from the file - // first put back one element each - final byte[] record = new byte[rowdef.objectsize]; - key = new byte[rowdef.primaryKeyLength]; - for (final Integer[] ds: doubles) { - file.get(ds[0].intValue(), record, 0); - System.arraycopy(record, 0, key, 0, rowdef.primaryKeyLength); - index.addi(key, ds[0].intValue()); - } - // then remove the other doubles by removing them from the table, but do a re-indexing while doing that - // first aggregate all the delete positions because the elements from the top positions must be removed first - final TreeSet delpos = new TreeSet(); - for (final Integer[] ds: doubles) { - for (int j = 1; j < ds.length; j++) delpos.add(ds[j]); - } - // now remove the entries in a sorted way (top-down) - Integer top; - while (delpos.size() > 0) { - top = delpos.last(); - delpos.remove(top); - removeInFile(top.intValue()); + + // remove doubles + if (!freshFile) { + final ArrayList doubles = index.removeDoubles(); + //assert index.size() + doubles.size() + fail == i; + //System.out.println(" -removed " + doubles.size() + " doubles- done."); + if (doubles.size() > 0) { + Log.logInfo("ECOTABLE", tablefile + ": WARNING - EcoTable " + tablefile + " has " + doubles.size() + " doubles"); + // from all the doubles take one, put it back to the index and remove the others from the file + // first put back one element each + final byte[] record = new byte[rowdef.objectsize]; + key = new byte[rowdef.primaryKeyLength]; + for (final Integer[] ds: doubles) { + file.get(ds[0].intValue(), record, 0); + System.arraycopy(record, 0, key, 0, rowdef.primaryKeyLength); + index.addi(key, ds[0].intValue()); + } + // then remove the other doubles by removing them from the table, but do a re-indexing while doing that + // first aggregate all the delete positions because the elements from the top positions must be removed first + final TreeSet delpos = new TreeSet(); + for (final Integer[] ds: doubles) { + for (int j = 1; j < ds.length; j++) delpos.add(ds[j]); + } + // now remove the entries in a sorted way (top-down) + Integer top; + while (delpos.size() > 0) { + top = delpos.last(); + delpos.remove(top); + removeInFile(top.intValue()); + } } } - /* try { - assert file.size() == index.size() + doubles.size() + fail : "file.size() = " + file.size() + ", index.size() = " + index.size() + ", doubles.size() = " + doubles.size() + ", fail = " + fail + ", i = " + i; - } catch (IOException e) { - e.printStackTrace(); - }*/ } catch (final FileNotFoundException e) { // should never happen e.printStackTrace(); @@ -297,6 +296,7 @@ public class EcoTable implements ObjectIndex { Integer L; Row.Entry inconsistentEntry; // iterate over all entries that have inconsistent index references + long lastlog = System.currentTimeMillis(); for (final Integer[] is: index.removeDoubles()) { // 'is' is the set of all indexes, that have the same reference // we collect that entries now here @@ -318,6 +318,9 @@ public class EcoTable implements ObjectIndex { s = d.last(); d.remove(s); this.removeInFile(s.intValue()); + if (System.currentTimeMillis() - lastlog > 30000) { + Log.logInfo("EcoTable", "removing " + d.size() + " entries in " + this.filename()); + } } assert file.size() == index.size() + fail : "file.size() = " + file.size() + ", index.size() = " + index.size(); return report;