You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
yacy_search_server/source/de/anomic/crawler/IndexingStack.java

638 lines
26 KiB

// IndexingStack.java
// (C) 2005 by Michael Peter Christen; mc@yacy.net
// first published on http://yacy.net
// Frankfurt, Germany, 2005
//
// $LastChangedDate$
// $LastChangedRevision$
// $LastChangedBy$
//
// LICENSE
//
// This program is free software; you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation; either version 2 of the License, or
// (at your option) any later version.
//
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU General Public License
// along with this program; if not, write to the Free Software
// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
package de.anomic.crawler;
import java.io.File;
import java.io.IOException;
import java.net.MalformedURLException;
import java.util.Collection;
import java.util.Date;
import java.util.Iterator;
import java.util.concurrent.ConcurrentHashMap;
import de.anomic.http.httpResponseHeader;
import de.anomic.index.indexURLReference;
import de.anomic.kelondro.kelondroBase64Order;
import de.anomic.kelondro.kelondroNaturalOrder;
import de.anomic.kelondro.kelondroRow;
import de.anomic.kelondro.kelondroStack;
import de.anomic.plasma.plasmaHTCache;
import de.anomic.plasma.plasmaSwitchboardConstants;
import de.anomic.plasma.plasmaWordIndex;
import de.anomic.server.serverDate;
import de.anomic.server.logging.serverLog;
import de.anomic.yacy.yacySeed;
import de.anomic.yacy.yacySeedDB;
import de.anomic.yacy.yacyURL;
public class IndexingStack {
kelondroStack sbQueueStack;
CrawlProfile profiles;
major step forward to network switching (target is easy switch to intranet or other networks .. and back) This change is inspired by the need to see a network connected to the index it creates in a indexing team. It is not possible to divide the network and the index. Therefore all control files for the network was moved to the network within the INDEX/<network-name> subfolder. The remaining YACYDB is superfluous and can be deleted. The yacyDB and yacyNews data structures are now part of plasmaWordIndex. Therefore all methods, using static access to yacySeedDB had to be rewritten. A special problem had been all the port forwarding methods which had been tightly mixed with seed construction. It was not possible to move the port forwarding functions to the place, meaning and usage of plasmaWordIndex. Therefore the port forwarding had been deleted (I guess nobody used it and it can be simulated by methods outside of YaCy). The mySeed.txt is automatically moved to the current network position. A new effect causes that every network will create a different local seed file, which is ok, since the seed identifies the peer only against the network (it is the purpose of the seed hash to give a peer a location within the DHT). No other functional change has been made. The next steps to enable network switcing are: - shift of crawler tables from PLASMADB into the network (crawls are also network-specific) - possibly shift of plasmaWordIndex code into yacy package (index management is network-specific) - servlet to switch networks git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@4765 6c8d7289-2bf4-0310-a012-ef5d649a1542
17 years ago
plasmaWordIndex wordIndex;
private final File sbQueueStackPath;
ConcurrentHashMap<String, QueueEntry> queueInProcess;
public IndexingStack(final plasmaWordIndex wordIndex, final File sbQueueStackPath, final CrawlProfile profiles) {
this.sbQueueStackPath = sbQueueStackPath;
this.profiles = profiles;
major step forward to network switching (target is easy switch to intranet or other networks .. and back) This change is inspired by the need to see a network connected to the index it creates in a indexing team. It is not possible to divide the network and the index. Therefore all control files for the network was moved to the network within the INDEX/<network-name> subfolder. The remaining YACYDB is superfluous and can be deleted. The yacyDB and yacyNews data structures are now part of plasmaWordIndex. Therefore all methods, using static access to yacySeedDB had to be rewritten. A special problem had been all the port forwarding methods which had been tightly mixed with seed construction. It was not possible to move the port forwarding functions to the place, meaning and usage of plasmaWordIndex. Therefore the port forwarding had been deleted (I guess nobody used it and it can be simulated by methods outside of YaCy). The mySeed.txt is automatically moved to the current network position. A new effect causes that every network will create a different local seed file, which is ok, since the seed identifies the peer only against the network (it is the purpose of the seed hash to give a peer a location within the DHT). No other functional change has been made. The next steps to enable network switcing are: - shift of crawler tables from PLASMADB into the network (crawls are also network-specific) - possibly shift of plasmaWordIndex code into yacy package (index management is network-specific) - servlet to switch networks git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@4765 6c8d7289-2bf4-0310-a012-ef5d649a1542
17 years ago
this.wordIndex = wordIndex;
this.queueInProcess = new ConcurrentHashMap<String, QueueEntry>();
initQueueStack();
}
public static final kelondroRow rowdef = new kelondroRow(
"String url-256, " + // the url
"String refhash-" + yacySeedDB.commonHashLength + ", " + // the url's referrer hash
"Cardinal modifiedsince-11 {b64e}, " + // from ifModifiedSince
"byte[] flags-1, " + // flags
"String initiator-" + yacySeedDB.commonHashLength + ", " + // the crawling initiator
"Cardinal depth-2 {b64e}, " + // the prefetch depth so far, starts at 0
"String profile-" + yacySeedDB.commonHashLength + ", " + // the name of the prefetch profile handle
"String urldescr-80",
kelondroNaturalOrder.naturalOrder,
0);
private void initQueueStack() {
sbQueueStack = kelondroStack.open(sbQueueStackPath, rowdef);
}
/*
private void resetQueueStack() {
try {sbQueueStack.close();} catch (Exception e) {}
if (sbQueueStackPath.exists()) sbQueueStackPath.delete();
initQueueStack();
}
*/
public int size() {
return (sbQueueStack == null) ? 0 : sbQueueStack.size();
}
public synchronized void push(final QueueEntry entry) throws IOException {
if (entry == null) return;
if (sbQueueStack == null) return; // may occur during shutdown
sbQueueStack.push(sbQueueStack.row().newEntry(new byte[][]{
entry.url.toString().getBytes(),
major step forward to network switching (target is easy switch to intranet or other networks .. and back) This change is inspired by the need to see a network connected to the index it creates in a indexing team. It is not possible to divide the network and the index. Therefore all control files for the network was moved to the network within the INDEX/<network-name> subfolder. The remaining YACYDB is superfluous and can be deleted. The yacyDB and yacyNews data structures are now part of plasmaWordIndex. Therefore all methods, using static access to yacySeedDB had to be rewritten. A special problem had been all the port forwarding methods which had been tightly mixed with seed construction. It was not possible to move the port forwarding functions to the place, meaning and usage of plasmaWordIndex. Therefore the port forwarding had been deleted (I guess nobody used it and it can be simulated by methods outside of YaCy). The mySeed.txt is automatically moved to the current network position. A new effect causes that every network will create a different local seed file, which is ok, since the seed identifies the peer only against the network (it is the purpose of the seed hash to give a peer a location within the DHT). No other functional change has been made. The next steps to enable network switcing are: - shift of crawler tables from PLASMADB into the network (crawls are also network-specific) - possibly shift of plasmaWordIndex code into yacy package (index management is network-specific) - servlet to switch networks git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@4765 6c8d7289-2bf4-0310-a012-ef5d649a1542
17 years ago
(entry.referrerHash == null) ? "".getBytes() : entry.referrerHash.getBytes(),
kelondroBase64Order.enhancedCoder.encodeLong((entry.ifModifiedSince == null) ? 0 : entry.ifModifiedSince.getTime(), 11).getBytes(),
new byte[]{entry.flags},
major step forward to network switching (target is easy switch to intranet or other networks .. and back) This change is inspired by the need to see a network connected to the index it creates in a indexing team. It is not possible to divide the network and the index. Therefore all control files for the network was moved to the network within the INDEX/<network-name> subfolder. The remaining YACYDB is superfluous and can be deleted. The yacyDB and yacyNews data structures are now part of plasmaWordIndex. Therefore all methods, using static access to yacySeedDB had to be rewritten. A special problem had been all the port forwarding methods which had been tightly mixed with seed construction. It was not possible to move the port forwarding functions to the place, meaning and usage of plasmaWordIndex. Therefore the port forwarding had been deleted (I guess nobody used it and it can be simulated by methods outside of YaCy). The mySeed.txt is automatically moved to the current network position. A new effect causes that every network will create a different local seed file, which is ok, since the seed identifies the peer only against the network (it is the purpose of the seed hash to give a peer a location within the DHT). No other functional change has been made. The next steps to enable network switcing are: - shift of crawler tables from PLASMADB into the network (crawls are also network-specific) - possibly shift of plasmaWordIndex code into yacy package (index management is network-specific) - servlet to switch networks git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@4765 6c8d7289-2bf4-0310-a012-ef5d649a1542
17 years ago
(entry.initiator == null) ? "".getBytes() : entry.initiator.getBytes(),
kelondroBase64Order.enhancedCoder.encodeLong(entry.depth, rowdef.width(5)).getBytes(),
major step forward to network switching (target is easy switch to intranet or other networks .. and back) This change is inspired by the need to see a network connected to the index it creates in a indexing team. It is not possible to divide the network and the index. Therefore all control files for the network was moved to the network within the INDEX/<network-name> subfolder. The remaining YACYDB is superfluous and can be deleted. The yacyDB and yacyNews data structures are now part of plasmaWordIndex. Therefore all methods, using static access to yacySeedDB had to be rewritten. A special problem had been all the port forwarding methods which had been tightly mixed with seed construction. It was not possible to move the port forwarding functions to the place, meaning and usage of plasmaWordIndex. Therefore the port forwarding had been deleted (I guess nobody used it and it can be simulated by methods outside of YaCy). The mySeed.txt is automatically moved to the current network position. A new effect causes that every network will create a different local seed file, which is ok, since the seed identifies the peer only against the network (it is the purpose of the seed hash to give a peer a location within the DHT). No other functional change has been made. The next steps to enable network switcing are: - shift of crawler tables from PLASMADB into the network (crawls are also network-specific) - possibly shift of plasmaWordIndex code into yacy package (index management is network-specific) - servlet to switch networks git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@4765 6c8d7289-2bf4-0310-a012-ef5d649a1542
17 years ago
(entry.profileHandle == null) ? "".getBytes() : entry.profileHandle.getBytes(),
(entry.anchorName == null) ? "-".getBytes("UTF-8") : entry.anchorName.getBytes("UTF-8")
}));
}
public synchronized QueueEntry pop() throws IOException {
if (sbQueueStack.size() == 0) return null;
final kelondroRow.Entry b = sbQueueStack.pot();
if (b == null) return null;
return new QueueEntry(b);
}
public synchronized QueueEntry remove(final String urlHash) {
final Iterator<kelondroRow.Entry> i = sbQueueStack.stackIterator(true);
kelondroRow.Entry rowentry;
QueueEntry entry;
while (i.hasNext()) {
rowentry = i.next();
entry = new QueueEntry(rowentry);
if (entry.urlHash().equals(urlHash)) {
i.remove();
return entry;
}
}
return null;
}
public void clear() {
sbQueueStack = kelondroStack.reset(sbQueueStack);
}
public void close() {
if (sbQueueStack != null) {
sbQueueStack.close();
}
sbQueueStack = null;
}
protected void finalize() throws Throwable {
try {
close();
} catch (final Exception e) {
throw new IOException("plasmaSwitchboardQueue.finalize()" + e.getMessage());
}
super.finalize();
}
public Iterator<QueueEntry> entryIterator(final boolean up) {
// iterates the elements in an ordered way.
// returns plasmaSwitchboardQueue.Entry - type Objects
return new entryIterator(up);
}
public class entryIterator implements Iterator<QueueEntry> {
Iterator<kelondroRow.Entry> rows;
public entryIterator(final boolean up) {
rows = sbQueueStack.stackIterator(up);
}
public boolean hasNext() {
return rows.hasNext();
}
public QueueEntry next() {
return new QueueEntry(rows.next());
}
public void remove() {
rows.remove();
}
}
public QueueEntry newEntry(final yacyURL url, final String referrer, final Date ifModifiedSince, final boolean requestWithCookie,
final String initiator, final int depth, final String profilehandle, final String anchorName) {
return new QueueEntry(url, referrer, ifModifiedSince, requestWithCookie, initiator, depth, profilehandle, anchorName);
}
public void enQueueToActive(final QueueEntry entry) {
queueInProcess.put(entry.urlHash(), entry);
}
public QueueEntry getActiveEntry(final String urlhash) {
// show one entry from the queue
return this.queueInProcess.get(urlhash);
}
public int getActiveQueueSize() {
return this.queueInProcess.size();
}
public Collection<QueueEntry> getActiveQueueEntries() {
// todo: check dead entries?
return this.queueInProcess.values();
}
public static final int QUEUE_STATE_FRESH = 0;
public static final int QUEUE_STATE_PARSING = 1;
public static final int QUEUE_STATE_CONDENSING = 2;
public static final int QUEUE_STATE_STRUCTUREANALYSIS = 3;
public static final int QUEUE_STATE_INDEXSTORAGE = 4;
public static final int QUEUE_STATE_FINISHED = 5;
public class QueueEntry {
yacyURL url; // plasmaURL.urlStringLength
String referrerHash; // plasmaURL.urlHashLength
Date ifModifiedSince; // 6
byte flags; // 1
String initiator; // yacySeedDB.commonHashLength
int depth; // plasmaURL.urlCrawlDepthLength
String profileHandle; // plasmaURL.urlCrawlProfileHandleLength
String anchorName; // plasmaURL.urlDescrLength
int status;
// computed values
private CrawlProfile.entry profileEntry;
private httpResponseHeader responseHeader;
private yacyURL referrerURL;
public QueueEntry(final yacyURL url, final String referrer, final Date ifModifiedSince, final boolean requestWithCookie,
final String initiator, final int depth, final String profileHandle, final String anchorName) {
this.url = url;
this.referrerHash = referrer;
this.ifModifiedSince = ifModifiedSince;
this.flags = (requestWithCookie) ? (byte) 1 : (byte) 0;
this.initiator = initiator;
this.depth = depth;
this.profileHandle = profileHandle;
this.anchorName = (anchorName==null)?"":anchorName.trim();
this.profileEntry = null;
this.responseHeader = null;
this.referrerURL = null;
this.status = QUEUE_STATE_FRESH;
}
public QueueEntry(final kelondroRow.Entry row) {
final long ims = row.getColLong(2);
final byte flags = row.getColByte(3);
try {
this.url = new yacyURL(row.getColString(0, "UTF-8"), null);
} catch (final MalformedURLException e) {
this.url = null;
}
this.referrerHash = row.getColString(1, "UTF-8");
this.ifModifiedSince = (ims == 0) ? null : new Date(ims);
this.flags = ((flags & 1) == 1) ? (byte) 1 : (byte) 0;
this.initiator = row.getColString(4, "UTF-8");
this.depth = (int) row.getColLong(5);
this.profileHandle = row.getColString(6, "UTF-8");
this.anchorName = row.getColString(7, "UTF-8");
this.profileEntry = null;
this.responseHeader = null;
this.referrerURL = null;
this.status = QUEUE_STATE_FRESH;
}
public QueueEntry(final byte[][] row) throws IOException {
final long ims = (row[2] == null) ? 0 : kelondroBase64Order.enhancedCoder.decodeLong(new String(row[2], "UTF-8"));
final byte flags = (row[3] == null) ? 0 : row[3][0];
try {
this.url = new yacyURL(new String(row[0], "UTF-8"), null);
} catch (final MalformedURLException e) {
this.url = null;
}
this.referrerHash = (row[1] == null) ? null : new String(row[1], "UTF-8");
this.ifModifiedSince = (ims == 0) ? null : new Date(ims);
this.flags = ((flags & 1) == 1) ? (byte) 1 : (byte) 0;
this.initiator = (row[4] == null) ? null : new String(row[4], "UTF-8");
this.depth = (int) kelondroBase64Order.enhancedCoder.decodeLong(new String(row[5], "UTF-8"));
this.profileHandle = new String(row[6], "UTF-8");
this.anchorName = (row[7] == null) ? null : (new String(row[7], "UTF-8")).trim();
this.profileEntry = null;
this.responseHeader = null;
this.referrerURL = null;
this.status = QUEUE_STATE_FRESH;
}
public void updateStatus(final int newStatus) {
this.status = newStatus;
}
public void close() {
queueInProcess.remove(this.url.hash());
}
protected void finalize() {
this.close();
}
redesigned NURL-handling: - the general NURL-index for all crawl stack types was splitted into separate indexes for these stacks - the new NURL-index is managed by the crawl balancer - the crawl balancer does not need an internal index any more, it is replaced by the NURL-index - the NURL.Entry was generalized and is now a new class plasmaCrawlEntry - the new class plasmaCrawlEntry replaces also the preNURL.Entry class, and will also replace the switchboardEntry class in the future - the new class plasmaCrawlEntry is more accurate for date entries (holds milliseconds) and can contain larger 'name' entries (anchor tag names) - the EURL object was replaced by a new ZURL object, which is a container for the plasmaCrawlEntry and some tracking information - the EURL index is now filled with ZURL objects - a new index delegatedURL holds ZURL objects about plasmaCrawlEntry obects to track which url is handed over to other peers - redesigned handling of plasmaCrawlEntry - handover, because there is no need any more to convert one entry object into another - found and fixed numerous bugs in the context of crawl state handling - fixed a serious bug in kelondroCache which caused that entries could not be removed - fixed some bugs in online interface and adopted monitor output to new entry objects - adopted yacy protocol to handle new delegatedURL entries all old crawl queues will disappear after this update! git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@3483 6c8d7289-2bf4-0310-a012-ef5d649a1542
18 years ago
public yacyURL url() {
return url;
}
public String urlHash() {
return url.hash();
}
public boolean requestedWithCookie() {
return (flags & 1) == 1;
}
public File cacheFile() {
return plasmaHTCache.getCachePath(url);
}
public boolean proxy() {
// FIXME the equals seems to be incorrect: String.equals(boolean)
major step forward to network switching (target is easy switch to intranet or other networks .. and back) This change is inspired by the need to see a network connected to the index it creates in a indexing team. It is not possible to divide the network and the index. Therefore all control files for the network was moved to the network within the INDEX/<network-name> subfolder. The remaining YACYDB is superfluous and can be deleted. The yacyDB and yacyNews data structures are now part of plasmaWordIndex. Therefore all methods, using static access to yacySeedDB had to be rewritten. A special problem had been all the port forwarding methods which had been tightly mixed with seed construction. It was not possible to move the port forwarding functions to the place, meaning and usage of plasmaWordIndex. Therefore the port forwarding had been deleted (I guess nobody used it and it can be simulated by methods outside of YaCy). The mySeed.txt is automatically moved to the current network position. A new effect causes that every network will create a different local seed file, which is ok, since the seed identifies the peer only against the network (it is the purpose of the seed hash to give a peer a location within the DHT). No other functional change has been made. The next steps to enable network switcing are: - shift of crawler tables from PLASMADB into the network (crawls are also network-specific) - possibly shift of plasmaWordIndex code into yacy package (index management is network-specific) - servlet to switch networks git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@4765 6c8d7289-2bf4-0310-a012-ef5d649a1542
17 years ago
return (initiator == null) || (initiator.equals(initiator.length() == 0));
}
public String initiator() {
return (initiator == null) ? "------------" : initiator; // TODO: this may cause problems for methods that check if the initiator is the proxy
}
public yacySeed initiatorPeer() {
major step forward to network switching (target is easy switch to intranet or other networks .. and back) This change is inspired by the need to see a network connected to the index it creates in a indexing team. It is not possible to divide the network and the index. Therefore all control files for the network was moved to the network within the INDEX/<network-name> subfolder. The remaining YACYDB is superfluous and can be deleted. The yacyDB and yacyNews data structures are now part of plasmaWordIndex. Therefore all methods, using static access to yacySeedDB had to be rewritten. A special problem had been all the port forwarding methods which had been tightly mixed with seed construction. It was not possible to move the port forwarding functions to the place, meaning and usage of plasmaWordIndex. Therefore the port forwarding had been deleted (I guess nobody used it and it can be simulated by methods outside of YaCy). The mySeed.txt is automatically moved to the current network position. A new effect causes that every network will create a different local seed file, which is ok, since the seed identifies the peer only against the network (it is the purpose of the seed hash to give a peer a location within the DHT). No other functional change has been made. The next steps to enable network switcing are: - shift of crawler tables from PLASMADB into the network (crawls are also network-specific) - possibly shift of plasmaWordIndex code into yacy package (index management is network-specific) - servlet to switch networks git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@4765 6c8d7289-2bf4-0310-a012-ef5d649a1542
17 years ago
if ((initiator == null) || (initiator.length() == 0)) return null;
if (initiator.equals(wordIndex.seedDB.mySeed().hash)) {
// normal crawling
return null;
}
// this was done for remote peer (a global crawl)
return wordIndex.seedDB.getConnected(initiator);
}
public int depth() {
return depth;
}
public long size() {
if (cacheFile().exists()) return cacheFile().length();
return 0;
}
public CrawlProfile.entry profile() {
if (profileEntry == null) profileEntry = profiles.getEntry(profileHandle);
return profileEntry;
}
private void getResponseHeader() {
if (this.responseHeader == null) try {
this.responseHeader = plasmaHTCache.loadResponseHeader(this.url);
} catch (final Exception e) {
serverLog.logSevere("PLASMA", "responseHeader: failed to get header", e);
}
}
public String getMimeType() {
this.getResponseHeader();
return (responseHeader == null) ? null : responseHeader.mime();
}
public String getCharacterEncoding() {
getResponseHeader();
return (responseHeader == null) ? null : responseHeader.getCharacterEncoding();
}
public Date getModificationDate() {
getResponseHeader();
return (responseHeader == null) ? new Date() : responseHeader.lastModified();
}
public yacyURL referrerURL() {
if (referrerURL == null) {
// FIXME the equals seems to be incorrect: String.equals(boolean)
if ((referrerHash == null) || ((initiator != null) && (referrerHash.equals(initiator.length() == 0)))) return null;
final indexURLReference entry = wordIndex.getURL(referrerHash, null, 0);
if (entry == null) referrerURL = null; else referrerURL = entry.comp().url();
}
return referrerURL;
}
public String referrerHash() {
return (referrerHash == null) ? "" : referrerHash;
}
public String anchorName() {
return anchorName;
}
public int processCase() {
// we must distinguish the following cases: resource-load was initiated by
// 1) global crawling: the index is extern, not here (not possible here)
// 2) result of search queries, some indexes are here (not possible here)
// 3) result of index transfer, some of them are here (not possible here)
// 4) proxy-load (initiator is "------------")
// 5) local prefetch/crawling (initiator is own seedHash)
// 6) local fetching for global crawling (other known or unknwon initiator)
int processCase = plasmaSwitchboardConstants.PROCESSCASE_0_UNKNOWN;
// FIXME the equals seems to be incorrect: String.equals(boolean)
if ((initiator == null) || initiator.length() == 0 || initiator.equals("------------")) {
// proxy-load
processCase = plasmaSwitchboardConstants.PROCESSCASE_4_PROXY_LOAD;
} else if (initiator.equals(wordIndex.seedDB.mySeed().hash)) {
// normal crawling
processCase = plasmaSwitchboardConstants.PROCESSCASE_5_LOCAL_CRAWLING;
} else {
// this was done for remote peer (a global crawl)
processCase = plasmaSwitchboardConstants.PROCESSCASE_6_GLOBAL_CRAWLING;
}
return processCase;
}
/**
* decide upon header information if a specific file should be indexed
* this method returns null if the answer is 'YES'!
* if the answer is 'NO' (do not index), it returns a string with the reason
* to reject the crawling demand in clear text
*
* This function is used by plasmaSwitchboard#processResourceStack
*/
public final String shallIndexCacheForProxy() {
if (profile() == null) {
return "shallIndexCacheForProxy: profile() is null !";
}
// check profile
if (!profile().indexText() && !profile().indexMedia()) {
return "Indexing_Not_Allowed";
}
// -CGI access in request
// CGI access makes the page very individual, and therefore not usable in caches
if (!profile().crawlingQ()) {
if (url.isPOST()) {
return "Dynamic_(POST)";
}
if (url.isCGI()) {
return "Dynamic_(CGI)";
}
}
// -authorization cases in request
// we checked that in shallStoreCache
// -ranges in request
// we checked that in shallStoreCache
// a picture cannot be indexed
if (plasmaHTCache.noIndexingURL(url)) {
return "Media_Content_(forbidden)";
}
// -cookies in request
// unfortunately, we cannot index pages which have been requested with a cookie
// because the returned content may be special for the client
if (requestedWithCookie()) {
// System.out.println("***not indexed because cookie");
return "Dynamic_(Requested_With_Cookie)";
}
getResponseHeader();
if (responseHeader != null) {
// -set-cookie in response
// the set-cookie from the server does not indicate that the content is special
// thus we do not care about it here for indexing
// a picture cannot be indexed
final String mimeType = responseHeader.mime();
if (plasmaHTCache.isPicture(mimeType)) {
return "Media_Content_(Picture)";
}
if (!plasmaHTCache.isText(mimeType)) {
return "Media_Content_(not_text)";
}
// -if-modified-since in request
// if the page is fresh at the very moment we can index it
final Date ifModifiedSince = getModificationDate();
if ((ifModifiedSince != null) && (responseHeader.containsKey(httpResponseHeader.LAST_MODIFIED))) {
// parse date
Date d = responseHeader.lastModified();
if (d == null) {
d = new Date(serverDate.correctedUTCTime());
}
// finally, we shall treat the cache as stale if the modification time is after the if-.. time
if (d.after(ifModifiedSince)) {
//System.out.println("***not indexed because if-modified-since");
return "Stale_(Last-Modified>Modified-Since)";
}
}
// -pragma in cached response
if (responseHeader.containsKey(httpResponseHeader.PRAGMA) &&
(responseHeader.get(httpResponseHeader.PRAGMA)).toUpperCase().equals("NO-CACHE")) {
return "Denied_(pragma_no_cache)";
}
// see for documentation also:
// http://www.web-caching.com/cacheability.html
// look for freshnes information
// -expires in cached response
// the expires value gives us a very easy hint when the cache is stale
// sometimes, the expires date is set to the past to prevent that a page is cached
// we use that information to see if we should index it
final Date expires = responseHeader.expires();
if (expires != null && expires.before(new Date(serverDate.correctedUTCTime()))) {
return "Stale_(Expired)";
}
// -lastModified in cached response
// this information is too weak to use it to prevent indexing
// even if we can apply a TTL heuristic for cache usage
// -cache-control in cached response
// the cache-control has many value options.
String cacheControl = responseHeader.get(httpResponseHeader.CACHE_CONTROL);
if (cacheControl != null) {
cacheControl = cacheControl.trim().toUpperCase();
/* we have the following cases for cache-control:
"public" -- can be indexed
"private", "no-cache", "no-store" -- cannot be indexed
"max-age=<delta-seconds>" -- stale/fresh dependent on date
*/
if (cacheControl.startsWith("PRIVATE") ||
cacheControl.startsWith("NO-CACHE") ||
cacheControl.startsWith("NO-STORE")) {
// easy case
return "Stale_(denied_by_cache-control=" + cacheControl + ")";
// } else if (cacheControl.startsWith("PUBLIC")) {
// // ok, do nothing
} else if (cacheControl.startsWith("MAX-AGE=")) {
// we need also the load date
final Date date = responseHeader.date();
if (date == null) {
return "Stale_(no_date_given_in_response)";
}
try {
final long ttl = 1000 * Long.parseLong(cacheControl.substring(8)); // milliseconds to live
if (serverDate.correctedUTCTime() - date.getTime() > ttl) {
//System.out.println("***not indexed because cache-control");
return "Stale_(expired_by_cache-control)";
}
} catch (final Exception e) {
return "Error_(" + e.getMessage() + ")";
}
}
}
}
return null;
}
/**
* decide upon header information if a specific file should be indexed
* this method returns null if the answer is 'YES'!
* if the answer is 'NO' (do not index), it returns a string with the reason
* to reject the crawling demand in clear text
*
* This function is used by plasmaSwitchboard#processResourceStack
*/
public final String shallIndexCacheForCrawler() {
if (profile() == null) {
return "shallIndexCacheForCrawler: profile() is null !";
}
// check profile
if (!profile().indexText() && !profile().indexMedia()) {
return "Indexing_Not_Allowed";
}
// -CGI access in request
// CGI access makes the page very individual, and therefore not usable in caches
if (!profile().crawlingQ()) {
if (url().isPOST()) { return "Dynamic_(POST)"; }
if (url().isCGI()) { return "Dynamic_(CGI)"; }
}
// -authorization cases in request
// we checked that in shallStoreCache
// -ranges in request
// we checked that in shallStoreCache
// a picture cannot be indexed
getResponseHeader();
if (responseHeader != null) {
final String mimeType = responseHeader.mime();
if (plasmaHTCache.isPicture(mimeType)) { return "Media_Content_(Picture)"; }
if (!plasmaHTCache.isText(mimeType)) { return "Media_Content_(not_text)"; }
}
if (plasmaHTCache.noIndexingURL(url())) { return "Media_Content_(forbidden)"; }
// -if-modified-since in request
// if the page is fresh at the very moment we can index it
// -> this does not apply for the crawler
// -cookies in request
// unfortunately, we cannot index pages which have been requested with a cookie
// because the returned content may be special for the client
// -> this does not apply for a crawler
// -set-cookie in response
// the set-cookie from the server does not indicate that the content is special
// thus we do not care about it here for indexing
// -> this does not apply for a crawler
// -pragma in cached response
// -> in the crawler we ignore this
// look for freshnes information
// -expires in cached response
// the expires value gives us a very easy hint when the cache is stale
// sometimes, the expires date is set to the past to prevent that a page is cached
// we use that information to see if we should index it
// -> this does not apply for a crawler
// -lastModified in cached response
// this information is too weak to use it to prevent indexing
// even if we can apply a TTL heuristic for cache usage
// -cache-control in cached response
// the cache-control has many value options.
// -> in the crawler we ignore this
return null;
}
/*
* (non-Javadoc)
*
* @see java.lang.Object#toString()
*/
public String toString() {
return "QueueEntry of "+ url.toString() + ", ref="+referrerHash +", initiator="+initiator +", flags="+ flags +", anchor="+ anchorName;
}
} // class Entry
}