yacy_search_server/htroot/QuickCrawlLink_p.java
Michael Peter Christen cf47d94888 performance hack to parse numbers inside of substrings without actually
generating a substring. This avoids the allocation of a String object
ech time a substring is parsed. Should affect CPU load during RWI
transmission.
2012-05-21 13:40:46 +02:00

202 lines
8.3 KiB
Java

//QuickCrawlLink_p.java
//-----------------------
//part of the AnomicHTTPD caching proxy
//(C) by Michael Peter Christen; mc@yacy.net
//first published on http://www.anomic.de
//Frankfurt, Germany, 2004
//
//This file was contributed by Martin Thelian
//$LastChangedDate$
//$LastChangedBy$
//$LastChangedRevision$
//
//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
//You must compile this file with
//javac -classpath .:../classes IndexCreate_p.java
//if the shell's current path is HTROOT
import java.net.MalformedURLException;
import java.util.Date;
import net.yacy.cora.document.UTF8;
import net.yacy.cora.protocol.HeaderFramework;
import net.yacy.cora.protocol.RequestHeader;
import net.yacy.cora.services.federated.yacy.CacheStrategy;
import net.yacy.kelondro.data.meta.DigestURI;
import net.yacy.kelondro.util.NumberTools;
import net.yacy.search.Switchboard;
import net.yacy.search.index.Segment;
import net.yacy.search.index.Segments;
import de.anomic.crawler.CrawlProfile;
import de.anomic.crawler.retrieval.Request;
import de.anomic.server.serverObjects;
import de.anomic.server.serverSwitch;
public class QuickCrawlLink_p {
/**
* Example Javascript to call this servlet:
* <code>javascript:w = window.open('http://user:pwd@localhost:8090/QuickCrawlLink_p.html?indexText=on&indexMedia=on&crawlingQ=on&xdstopw=on&title=' + escape(document.title) + '&url=' + location.href,'_blank','height=150,width=500,resizable=yes,scrollbar=no,directory=no,menubar=no,location=no'); w.focus();</code>
* @param header the complete HTTP header of the request
* @param post any arguments for this servlet, the request carried with (GET as well as POST)
* @param env the serverSwitch object holding all runtime-data
* @return the rewrite-properties for the template
*/
public static serverObjects respond(final RequestHeader header, final serverObjects post, final serverSwitch env) {
final serverObjects prop = new serverObjects();
final Switchboard sb = (Switchboard) env;
// get segment
Segment indexSegment = null;
if (post != null && post.containsKey("segment")) {
final String segmentName = post.get("segment");
if (sb.indexSegments.segmentExist(segmentName)) {
indexSegment = sb.indexSegments.segment(segmentName);
}
} else {
// take default segment
indexSegment = sb.indexSegments.segment(Segments.Process.PUBLIC);
}
if (post == null) {
// send back usage example
prop.put("mode", "0");
// get the http host header
final String hostSocket = header.get(HeaderFramework.CONNECTION_PROP_HOST);
//String host = hostSocket;
int port = 80;
final int pos = hostSocket.indexOf(':',0);
if (pos != -1) {
port = NumberTools.parseIntDecSubstring(hostSocket, pos + 1);
//host = hostSocket.substring(0, pos);
}
prop.put("mode_host", "localhost");
prop.put("mode_port", port);
return prop;
}
prop.put("mode", "1");
// get the URL
String crawlingStart = post.get("url",null);
crawlingStart = UTF8.decodeURL(crawlingStart);
// get the browser title
final String title = post.get("title",null);
// get other parameters if set
final String crawlingMustMatch = post.get("mustmatch", CrawlProfile.MATCH_ALL_STRING);
final String crawlingMustNotMatch = post.get("mustnotmatch", CrawlProfile.MATCH_NEVER_STRING);
final int CrawlingDepth = post.getInt("crawlingDepth", 0);
final boolean crawlDynamic = post.get("crawlingQ", "").equals("on");
final boolean indexText = post.get("indexText", "on").equals("on");
final boolean indexMedia = post.get("indexMedia", "on").equals("on");
final boolean storeHTCache = post.get("storeHTCache", "").equals("on");
final boolean remoteIndexing = post.get("crawlOrder", "").equals("on");
final boolean xsstopw = post.get("xsstopw", "").equals("on");
final boolean xdstopw = post.get("xdstopw", "").equals("on");
final boolean xpstopw = post.get("xpstopw", "").equals("on");
prop.put("mode_url", (crawlingStart == null) ? "unknown" : crawlingStart);
prop.putHTML("mode_title", (title == null) ? "unknown" : title);
if (crawlingStart != null) {
crawlingStart = crawlingStart.trim();
try {crawlingStart = new DigestURI(crawlingStart).toNormalform(true, true);} catch (final MalformedURLException e1) {}
// check if url is proper
DigestURI crawlingStartURL = null;
try {
crawlingStartURL = new DigestURI(crawlingStart);
} catch (final MalformedURLException e) {
prop.put("mode_status", "1");
prop.put("mode_code", "1");
return prop;
}
final byte[] urlhash = crawlingStartURL.hash();
indexSegment.urlMetadata().remove(urlhash);
sb.crawlQueues.noticeURL.removeByURLHash(urlhash);
sb.crawlQueues.errorURL.remove(urlhash);
// create crawling profile
CrawlProfile pe = null;
try {
pe = new CrawlProfile(
crawlingStartURL.getHost(),
crawlingStartURL,
crawlingMustMatch,
CrawlProfile.MATCH_ALL_STRING,
CrawlProfile.MATCH_NEVER_STRING,
"",
crawlingMustNotMatch,
CrawlingDepth,
true,
60 * 24 * 30, // recrawlIfOlder (minutes); here: one month
-1, // domMaxPages, if negative: no count restriction
crawlDynamic,
indexText,
indexMedia,
storeHTCache,
remoteIndexing,
xsstopw,
xdstopw,
xpstopw,
CacheStrategy.IFFRESH);
sb.crawler.putActive(pe.handle().getBytes(), pe);
} catch (final Exception e) {
// mist
prop.put("mode_status", "2");//Error with url
prop.put("mode_code", "2");
prop.putHTML("mode_status_error", e.getMessage());
return prop;
}
// stack URL
String reasonString = null;
reasonString = sb.crawlStacker.stackCrawl(new Request(
sb.peers.mySeed().hash.getBytes(),
crawlingStartURL,
null,
(title==null)?"CRAWLING-ROOT":title,
new Date(),
pe.handle(),
0,
0,
0,
0
));
// validate rejection reason
if (reasonString == null) {
prop.put("mode_status", "0");//start msg
prop.put("mode_code", "0");
} else {
prop.put("mode_status", "3");//start msg
prop.put("mode_code","3");
prop.putHTML("mode_status_error", reasonString);
}
}
return prop;
}
}