alfresco-community-repo/source/java/org/alfresco/repo/node/index/FullIndexRecoveryComponent.java
Dave Ward a7b885a1c6 Merged V3.4-BUG-FIX to HEAD
29057: ALF-9491: Bitrock 7.2.2
   29063: ALF-8766 Concatenated strings in EN webclient file
   29066: Merge DEV/DEV/BELARUS/V3.4-BUG-FIX-2011_07_13 to DEV/V3.4-BUG-FIX
      29010: ALF-7396: Japanese- Untranslated
   29072: HomeFolderProvider work - Changes as a result of Dave Ward's comments
     (HomeFolderManager not fully done as there is a spring issue with using NodeService, FileFolderService, fileFolderService, SearchService or searchService) 
   29074: ALF-7637 - Share displays incorrect folder contents after copy-on-outbound rule against working copy
   29075: ALF-8406 - Configuring the datalist display for sub-types does not work
   29082: ALF-6847 translation: "Collega" should be reverted to English: "Link" as per term list.
   29087: ALF-5717 property names for wcm quickstart website-model had an invalid format or did not end in .description or .title
   29093: Merge V3.3 to DEV/V3.4-BUG-FIX (28596)
      28596: Remove dependency between subsystems and all the object factories in the parent context!
         - Do not allow eager initialization when looking up parent post processors
         - Removes circular dependencies from sysAdmin subsystem
   29094: Merge HEAD to DEV/V3.4-BUG-FIX ()
      28892: Broke circular references between NodeService beans, NodeIndexer, Lucene and back to NodeService.
         - NodeIndexer is now bootstrapped to pull out reference to the Lucene beans
   29100: Revert Merge V3.3 to DEV/V3.4-BUG-FIX (28596) Caused RepositoryStartupTest to fail 
      28596: Remove dependency between subsystems and all the object factories in the parent context!
         - Do not allow eager initialization when looking up parent post processors
         - Removes circular dependencies from sysAdmin subsystem
   29102: ALF-9048: Make apply_amps.bat work from its installed location
   29103: ALF-8746: Restored Japanese choice format translations
   29104: Merged V3.3 to V3.4-BUG-FIX (Reinstated this revision as it is required)
      28596: Remove dependency between subsystems and all the object factories in the parent context!
         - Do not allow eager initialization when looking up parent post processors
         - Removes circular dependencies from sysAdmin subsystem
   29105: Use org.springframework.aop.target.LazyInitTargetSource in the NodeService public proxy to break a circular dependency
   29106: Make PersonService interact with HomeFolderManager via a lazy proxy to prevent another circular dependency
   - Simple HomeFolderManager interface created
   - Implementation class renamed to PortableHomeFolderManager
   - Removed TODOs from authentication-services-context.xml
   29107: Forgot to remove the serviceRegistry dependency from homeFolderManager
   29108: ALF-9529: Installer memory consumption and startup time improvements
   - Bitrock discover the for loop!
   29109: ALF-9530: Postgres installed as Windows service should run as a postgres user, not System
   - Fix from Bitrock
   29118: Fix for ALF-6737 - It's impossible to view any version of the wiki page if it was renamed with native characters
   29119: Fix for ALF-5787 - strings extracted for L10N in Web form creation help text
   29124: ALF-9530: Follow up fix from Bitrock
   29126: Fix for ALF-8344 - Incorrect message is displayed while recover deleted file
   29127: Fix for ALF-9445 - French - Share, translation on Transfer Target configuration
   29129: ALF-9476: Make FTPS work on IBM JDK
   29133: Fix failing DictionaryRestApiTest
   29136: Fix build issues from 29104:
   - run as system when creating home folders (PortableHomeFolderManager)
   - re-factored onCreateNode out of PortableHomeFolderManager into PersonServiceImpl
   - re-factored property PortableHomeFolderManager.enableHomeFolderCreationAsPeopleAreCreated to PersonServiceImpl.homeFolderCreationEager 
   29137: Fix for ALF-8831 - Internal error occurs in My Tasks Webscripts component
   29138: Fix for ALF-8765 - Layout is displaced if translated string occupies more than 1 line
   29140: Fix for ALF-8668 - Deleting author account causes Failed to load the forum posts
   29142: - PortableHomeFolderManager: Moved code to run as System into PersonServiceImpl so that one must have a valid authority to call the publick makeHomeFolder method. The authority should already be valid if called via PersonServiceImpl.
   - Removed unused policyBehaviourFilter property from PersonServiceImpl
   29146: ALF-8701: partially translated string in html-upload.get_fr
   29147: ALF-8727: DE - changes to Root Category
   29149: ALF-8731: DE - Wiki changes (space before full stop)
   29152: ALF-9503: Add space after colon in strings in file wdr-messages.properties
   29153: Fixed ALF-7899: association.ftl does not render when showTargetLink=true in workflow
   29165: ALF-8749: on submit action properties in wcn-workflow-messages.properties
   29166: Fix for ALF-6220 - Language pack - .ftl localization
   29167: ALF-9550 - Typos in new section of webclient.properties
   29169: Fix for ALF-7844 - W3C: Impossible to activate 'Choose from popular tags in this site' link by Enter/Space keys
   29170: Merge V3.4-TEAM to V3.4-BUG-FIX (3.4.4)
      27471: Fix for ALF-8150 - check for visibility before applying focus to element for IE.
   29171: Fixes: ALF-8922, removes date formatting from API (now returns ISO8601) and instead formats it on the client, using L10N strings.
   29172: Fix for ALF-2023 - Repository Action - Copy item to specific space doesn not include content. The option to 'deep copy' is now exposed in the UI for Run Action and Rules in Explorer.
   29173: Fix for ALF-1446 - Sorting of inline descendants is not observed
   29175: ALF-241 - The item is not coppied via 'Paste All' in Shelf when 'Details' page is opened
   29177: Fix for ALF-9520 - confusing sample config. Reordered sample config file as suggested.
   29178: Fixed ALF-6400: GERMAN: Explorer mouse over hints for TinyMCE are not localized
   Fixed ALF-5766: ALL translations errors in Explorer - Calendars are not localizable for content based on webforms
   29202: Merge DEV/BELARUS/V3.4-BUG-FIX-2011_04_12 to V3.4-BUG-FIX (3.4.4)
      27836: ALF-8524: CLONE - Sharepoint doesn't work with HTTPS
         Changes in url links required for HTTPS support.
   29203: Restored removal of postgresCreateSymLinksLinuxBuildingFromWindows tag (32 bit Linux) from revision 26582
   29211: Fix for ALF-1051 - It is impossible to find link by tag from link details page
   29212: Fix for ALF-5301 - TinyMCE is replacing carriage return with white spaces
   29250: Latest L10N update for all languages (DE, ES, FR, IT, JA) from Gloria (based on r29172)
   29253: L10N Update from Gloria
   29270: Fixed ALF-516: Unable to add content/delete tables in webform content when using FireFox
   29271: Update from Gloria
   29272: Merged BRANCHES/DEV/BELARUS/V3.4-BUG-FIX-2011_07_13 to BRANCHES/DEV/V3.4-BUG-FIX: (with minor modification)
      29223: ALF-7619: When document A has an association with a document B editing A's properties fails if user has no permission to edit B
   29274: ALF-9517 Incorrect behaviour of versions on Copy action. Version is 0.1 rather than 1.0
   29283: Resolve ALF-8273: Valid datetime value cannot be parsed by CMIS AtomPub interface
   29284: Update from Gloria
   29286: ALF-9596: Merged PATCHES/V3.4.1 to V3.4-BUG-FIX
      28150: ALF-8607: Detailed debug logging when out of sync transaction detected by index checker / tracker
      28177: ALF-8607: Corrections to debug logging in AbstractReindexComponent
      28213: ALF-8607: Further corrections to debug logging in AbstractReindexComponent
      - Log attributes from indexes, rather than nodeService properties
      28341: ALF-8607: Stop index checker from 'lying'
      - isTxnPresentInIndex() call must be made in a new transaction in order to get a database view in sync with the current indexes
      28352: ALF-8607: Revisit transaction delineation. Nested transaction only required in checkTxnsImpl()
      28403: ALF-8607: Merged PATCHES/V3.3.4 to PATCHES/V3.4.1
         27823: ALF-7237: Index tracker needs to perform a cascade reindex on updated nodes in order to cope with node moves
      28406: ALF-8607: Improvement to FTS fix. Prevent FTS from restoring documents that have been deleted!
      28412: ALF-8607: Invalidate properties and aspects as well as parent assocs when stale cache entry dected during transaction tracking
      28427: ALF-8607: Prevent NPE with bad NodeRef in ADMLuceneIndexerImpl.createDocumentsImpl()
      28705: ALF-8607: Validate transaction IDs when fetching parent associations
      - Compare the cached child node transaction ID against one fetched from the DB
      - Stops us from pairing up the cached node for an older or newer transaction with the wrong parent associations
      28707: ALF-8607: Merged PATCHES/V3.3.4 to PATCHES/V3.4.1
         28588: ALF-7237: Prevent FTS from ever wiping out a document that still exists and ignore duplicates
      28708: ALF-8607: Make FTS capable of recovering from cache concurrency issues by using a RetryingTransactionHelper and better exception handling.
      - Also avoids skipping the entire batch when the reindexing of a particular document fails.
      28710: ALF-8607: Corrected transaction delineation
      28753: ALF-8607: Prevent errors caused by AbstractReindexComponent diagnostics trying to parse FTSREF document IDs as NodeRefs (which they aren't!)
      28755: ALF-8607: When 'failing over' during FTS indexing, don't bother adding a FTS status document so we don't get stuck in a loop with a problematic document
      28815: ALF-8607: Do two way validation of cached / fetched nodes and their parent associations to avoid skew
      - Should resolve problem of tracking moves to the archive store and moves in general
      28862: ALF-8607: Lucene indexers now support 'read through' behaviour for FTS and Index tracking batches
      - Small discrete read only transactions used to read each reindexed node from the database / cache
      - Avoids cache 'drift' and 'skew' after long running indexing transactions
      28863: ALF-8607: Missing file
      28869: ALF-8607: isTxnPresentInIndex() needs to 'read through' so index tracker and checker don't pollute the cache
      28872: ALF-8607: Optimization to prevent constant writing to AVM indexes whilst 'ticking over'.
      28950: ALF-8607: Improved logic in AbstractReindexComponent.isTxnPresentInIndex() so that we can reliably cope with multi-store transactions (e.g. archive store + spaces store)
      - Due to FTS, the txn ID may have 'drifted in' to one store but not the other so we must validate all stores in the txn
      29098: ALF-8607: Use getNodeRefStatus as a cache validation point for reindexing 'read through' transactions
      - Guarantees that FTS reindexed node will see correct state (well if we had consistent read behaviour it would!)
      - Removes stale nodeRef -> ID mappings (e.g. when original node moved to archive store and substituted with deleted node)
      - Inexplicably seems to produce a ~30x speedup in performance tests on MySQL! Appears to remove a contention point. More investigation required to find out what!
   29287: ALF-9598: Merged PATCHES/V3.4.1 to V3.4-BUG-FIX
      28653: ALF-9189: More efficient usage of IndexReaders to avoid huge memory churn with large indexes
      - A single reading thread could block out all other reading threads because a write lock is obtained whilst constructing a set of FilterIndexReaderByStringId readers and all deletions across all indexes have to be evaluated. We now cache a FilterIndexReaderByStringId for each 'layer' of the index so that we get some reuse. We also defer evaluation of deletions to AFTER the write lock is returned and in some cases never have to evaluate the deletions at all.
      - When merging deletions we now make use of a cached index reader for locating the documents, and only resort to a new reader if deletions have to be performed. Hopefully this will mean that the reader for the largest indexes, containing the least recently used stuff, will get left alone most of the time. 
      28690: ALF-9189: Corrections to previous fix
      - Forgot to remove non-lazy reader initialization
      - Fixed NPE
      - Reinstated correct looping behaviour - each processed delta must be considered as one of the indexes to search for the next processed delta
      29099: ALF-9189: Avoid having to allocate a byte array full of number ones for all occurrences of a term to 'fake' norms.
      - Severe Lucene memory hog during FTS
      29262: ALF-9189: Fixed memory leak during index tracking / reindexing and further memory leak regression
      - Fixed up Lucene refcounting again - remember to propagate through decrefs on ReferenceCounting readers
      - Refined ALF-9189 fix to guarantee mainreader clean up
      - Remember to flush the delta during reindexing / tracking
      - Some extra trace diagnostics to help
   29288: ALF-9600: Merged PATCHES/V3.4.1 to V3.4-BUG_FIX
      28876: ALF-9041: Merged HEAD to PATCHES/V3.4.1
         28850: Latest SpringSurf libs
            - Fix to SSO connector passing empty username
   29289: ALF-8241: assemble-tomcat populates endorsed directory with xalan.jar and serializer.jar and Bitrock installer installs these too
   29291: Merged DEV/SWIFT to V3.4-BUG-FIX (3.4.4) - already merged to HEAD as part of a larger merge
      26104: RM: Remove incomplete and unnecessary unit test     
   29302: Fix for ALF-8885 - Unable to paste item due to system error:null


git-svn-id: https://svn.alfresco.com/repos/alfresco-enterprise/alfresco/HEAD/root@29325 c4b6b30b-aa2e-2d43-bbcb-ca4b014f7261
2011-07-25 19:32:17 +00:00

533 lines
20 KiB
Java

/*
* Copyright (C) 2005-2010 Alfresco Software Limited.
*
* This file is part of Alfresco
*
* Alfresco is free software: you can redistribute it and/or modify
* it under the terms of the GNU Lesser General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* Alfresco 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 Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public License
* along with Alfresco. If not, see <http://www.gnu.org/licenses/>.
*/
package org.alfresco.repo.node.index;
import java.util.ArrayList;
import java.util.Date;
import java.util.Iterator;
import java.util.List;
import org.alfresco.model.ContentModel;
import org.alfresco.repo.domain.node.Transaction;
import org.alfresco.repo.node.index.IndexTransactionTracker.IndexTransactionTrackerListener;
import org.alfresco.repo.transaction.RetryingTransactionHelper.RetryingTransactionCallback;
import org.alfresco.service.cmr.repository.ChildAssociationRef;
import org.alfresco.service.cmr.repository.NodeRef;
import org.alfresco.service.cmr.repository.StoreRef;
import org.alfresco.service.namespace.NamespaceService;
import org.alfresco.service.namespace.QName;
import org.apache.commons.logging.Log;
import org.apache.commons.logging.LogFactory;
import org.springframework.extensions.surf.util.I18NUtil;
/**
* Component to check and recover the indexes. By default, the server is
* put into read-only mode during the reindex process in order to prevent metadata changes.
* This is not critical and can be {@link #setLockServer(boolean) switched off} if the
* server is required immediately.
* <p>
*
* @see RecoveryMode
*
* @author Derek Hulley
*/
public class FullIndexRecoveryComponent extends AbstractReindexComponent
{
private static final String ERR_INDEX_OUT_OF_DATE = "index.recovery.out_of_date";
private static final String MSG_TRACKING_STARTING = "index.tracking.starting";
private static final String MSG_TRACKING_COMPLETE = "index.tracking.complete";
private static final String MSG_TRACKING_PROGRESS = "index.tracking.progress";
private static final String MSG_RECOVERY_STARTING = "index.recovery.starting";
private static final String MSG_RECOVERY_COMPLETE = "index.recovery.complete";
private static final String MSG_RECOVERY_PROGRESS = "index.recovery.progress";
private static final String MSG_RECOVERY_TERMINATED = "index.recovery.terminated";
private static final String MSG_RECOVERY_ERROR = "index.recovery.error";
private static Log logger = LogFactory.getLog(FullIndexRecoveryComponent.class);
public static enum RecoveryMode
{
/** Do nothing - not even a check. */
NONE,
/**
* Perform a quick check on the state of the indexes. This only checks that the
* first N and last M transactions are present in the index and doesn't guarantee that
* the indexes are wholely consistent. Normally, the indexes are consistent up to a certain time.
*/
VALIDATE,
/**
* Performs a validation and starts a recovery if necessary. In this mode, if start
* transactions are missing then FULL mode is enabled. If end transactions are missing
* then the indexes will be "topped up" to bring them up to date.
*/
AUTO,
/**
* Performs a full pass-through of all recorded transactions to ensure that the indexes
* are up to date.
*/
FULL;
}
private RecoveryMode recoveryMode;
private boolean lockServer;
private IndexTransactionTracker indexTracker;
private boolean stopOnError;
private int maxTransactionsPerLuceneCommit;
private final QName vetoName = QName.createQName(NamespaceService.APP_MODEL_1_0_URI, "FullIndexRecoveryComponent");
/**
* <ul>
* <li><b>recoveryMode: </b>VALIDATE</li>
* <li><b>stopOnError:</b> true</li>
* </ul>
*
*/
public FullIndexRecoveryComponent()
{
recoveryMode = RecoveryMode.VALIDATE;
maxTransactionsPerLuceneCommit = 100;
}
/**
* Set the type of recovery to perform. Default is {@link RecoveryMode#VALIDATE to validate}
* the indexes only.
*
* @param recoveryMode one of the {@link RecoveryMode } values
*/
public void setRecoveryMode(String recoveryMode)
{
this.recoveryMode = RecoveryMode.valueOf(recoveryMode);
}
/**
* Set the number of transactions to process per Lucene write.
* Larger values generate less contention on the Lucene IndexInfo files.
*/
public void setMaxTransactionsPerLuceneCommit(int maxTransactionsPerLuceneCommit)
{
this.maxTransactionsPerLuceneCommit = maxTransactionsPerLuceneCommit;
}
/**
* Set this on to put the server into READ-ONLY mode for the duration of the index recovery.
* The default is <tt>true</tt>, i.e. the server will be locked against further updates.
*
* @param lockServer true to force the server to be read-only
*/
public void setLockServer(boolean lockServer)
{
this.lockServer = lockServer;
}
/**
* Set the tracker that will be used for AUTO mode.
*
* @param indexTracker an index tracker component
*/
public void setIndexTracker(IndexTransactionTracker indexTracker)
{
this.indexTracker = indexTracker;
}
/**
* Set whether a full rebuild should stop in the event of encoutering an error. The default is
* to stop reindexing, and this will lead to the server startup failing when index recovery mode
* is <b>FULL</b>. Sometimes, it is necessary to start the server up regardless of any errors
* with particular nodes.
*
* @param stopOnError <tt>true</tt> to stop reindexing when an error is encountered.
*/
public void setStopOnError(boolean stopOnError)
{
this.stopOnError = stopOnError;
}
@Override
protected void reindexImpl()
{
if (logger.isDebugEnabled())
{
logger.debug("Performing index recovery for type: " + recoveryMode);
}
// Ignore when NONE
if (recoveryMode == RecoveryMode.NONE)
{
return;
}
// put the server into read-only mode for the duration
boolean allowWrite = !transactionService.isReadOnly();
try
{
if (lockServer)
{
// set the server into read-only mode
transactionService.setAllowWrite(false, vetoName);
}
List<Transaction> startTxns = nodeDAO.getTxnsByCommitTimeAscending(
Long.MIN_VALUE, Long.MAX_VALUE, 1000, null, false);
InIndex startAllPresent = areTxnsInStartSample(startTxns);
List<Transaction> endTxns = nodeDAO.getTxnsByCommitTimeDescending(
Long.MIN_VALUE, Long.MAX_VALUE, 1000, null, false);
InIndex endAllPresent = areAllTxnsInEndSample(endTxns);
// check the level of cover required
switch (recoveryMode)
{
case AUTO:
if (startAllPresent == InIndex.NO)
{
// Initial transactions are missing - rebuild
performFullRecovery();
}
else if (endAllPresent == InIndex.NO)
{
performPartialRecovery();
}
break;
case VALIDATE:
// Check
if ((startAllPresent == InIndex.NO) || (endAllPresent == InIndex.NO))
{
// Index is out of date
logger.warn(I18NUtil.getMessage(ERR_INDEX_OUT_OF_DATE));
}
break;
case FULL:
performFullRecovery();
break;
}
}
finally
{
// restore read-only state
transactionService.setAllowWrite(true, vetoName);
}
}
/**
* @return Returns <tt>false</tt> if any one of the transactions aren't in the index.
*/
protected InIndex areAllTxnsInEndSample(List<Transaction> txns)
{
int count = 0;
int yesCount = 0;
for (Transaction txn : txns)
{
count++;
InIndex inIndex = isTxnPresentInIndex(txn, true);
if (inIndex == InIndex.NO)
{
// Missing txn
return InIndex.NO;
}
if (inIndex == InIndex.YES)
{
yesCount++;
if((yesCount > 1) && (count >= 10))
{
return InIndex.YES;
}
}
}
return InIndex.INDETERMINATE;
}
protected InIndex areTxnsInStartSample(List<Transaction> txns)
{
int count = 0;
InIndex current = InIndex.INDETERMINATE;
for (Transaction txn : txns)
{
count++;
current = isTxnPresentInIndex(txn, true);
if (current == InIndex.NO)
{
// Missing txn
return InIndex.NO;
}
if((current == InIndex.YES) && (count >= 10))
{
return InIndex.YES;
}
}
return current;
}
private void performPartialRecovery()
{
// Log the AUTO recovery
IndexTransactionTrackerListener trackerListener = new IndexTransactionTrackerListener()
{
long lastLogged = 0L;
public void indexedTransactions(long fromTimeInclusive, long toTimeExclusive)
{
long now = System.currentTimeMillis();
if (now - lastLogged < 10000L)
{
// Don't log more than once a minute
return;
}
lastLogged = now;
// Log it
Date toTimeDate = new Date(toTimeExclusive);
String msgAutoProgress = I18NUtil.getMessage(MSG_TRACKING_PROGRESS, toTimeDate.toString());
logger.info(msgAutoProgress);
}
};
try
{
// Register the listener
indexTracker.setListener(trackerListener);
// Trigger the tracker, which will top up the indexes
logger.info(I18NUtil.getMessage(MSG_TRACKING_STARTING));
indexTracker.reindex();
logger.info(I18NUtil.getMessage(MSG_TRACKING_COMPLETE));
}
finally
{
// Remove the listener
indexTracker.setListener(null);
}
}
private static final int MAX_TRANSACTIONS_PER_ITERATION = 1000;
private static final long MIN_SAMPLE_TIME = 10000L;
private void performFullRecovery()
{
RetryingTransactionCallback<Void> deleteWork = new RetryingTransactionCallback<Void>()
{
public Void execute() throws Exception
{
// delete stores
for(StoreRef storeRef : nodeService.getStores())
{
if(!storeRef.getProtocol().equals(StoreRef.PROTOCOL_AVM))
{
indexer.deleteIndex(storeRef);
}
}
return null;
}
};
transactionService.getRetryingTransactionHelper().doInTransaction(deleteWork, true, true);
int txnCount = nodeDAO.getTransactionCount();
// starting
String msgStart = I18NUtil.getMessage(MSG_RECOVERY_STARTING, txnCount);
logger.info(msgStart);
// count the transactions
int processedCount = 0;
long fromTimeInclusive = nodeDAO.getMinTxnCommitTime();
long maxToTimeExclusive = nodeDAO.getMaxTxnCommitTime() + 1;
// Our first sample will be 10 seconds long (as we often hit 'fake' transactions with time zero). We'll rebalance intervals from there...
long toTimeExclusive = fromTimeInclusive + MIN_SAMPLE_TIME;
long sampleStartTimeInclusive = fromTimeInclusive;
long sampleEndTimeExclusive = -1;
long txnsPerSample = 0;
List<Long> lastTxnIds = new ArrayList<Long>(MAX_TRANSACTIONS_PER_ITERATION);
while(true)
{
boolean startedSampleForQuery = false;
List<Transaction> nextTxns = nodeDAO.getTxnsByCommitTimeAscending(
fromTimeInclusive,
toTimeExclusive,
MAX_TRANSACTIONS_PER_ITERATION,
lastTxnIds,
false);
// have we finished?
if (nextTxns.size() == 0)
{
if (toTimeExclusive >= maxToTimeExclusive)
{
// there are no more
break;
}
}
// reindex each transaction
List<Long> txnIdBuffer = new ArrayList<Long>(maxTransactionsPerLuceneCommit);
Iterator<Transaction> txnIterator = nextTxns.iterator();
while (txnIterator.hasNext())
{
Transaction txn = txnIterator.next();
Long txnId = txn.getId();
// Remember the IDs of the last simultaneous transactions so they can be excluded from the next query
long txnCommitTime = txn.getCommitTimeMs();
if (lastTxnIds.isEmpty() || txnCommitTime != fromTimeInclusive)
{
if (!startedSampleForQuery)
{
sampleStartTimeInclusive = txnCommitTime;
sampleEndTimeExclusive = -1;
txnsPerSample = 0;
startedSampleForQuery = true;
}
else
{
txnsPerSample += lastTxnIds.size();
sampleEndTimeExclusive = txnCommitTime;
}
lastTxnIds.clear();
fromTimeInclusive = txnCommitTime;
}
lastTxnIds.add(txnId);
// check if we have to terminate
if (isShuttingDown())
{
String msgTerminated = I18NUtil.getMessage(MSG_RECOVERY_TERMINATED);
logger.warn(msgTerminated);
return;
}
// Allow exception to bubble out or not
if (stopOnError)
{
reindexTransaction(txnId);
}
else
{
// Add the transaction ID to the buffer
txnIdBuffer.add(txnId);
// Reindex if the buffer is full or if there are no more transactions
if (!txnIterator.hasNext() || txnIdBuffer.size() >= maxTransactionsPerLuceneCommit)
{
try
{
reindexTransactionAsynchronously(txnIdBuffer, true);
}
catch (Throwable e)
{
String msgError = I18NUtil.getMessage(MSG_RECOVERY_ERROR, txnId, e.getMessage());
logger.info(msgError, e);
}
// Clear the buffer
txnIdBuffer = new ArrayList<Long>(maxTransactionsPerLuceneCommit);
}
}
// dump a progress report every 10% of the way
double before = (double) processedCount / (double) txnCount * 10.0; // 0 - 10
processedCount++;
double after = (double) processedCount / (double) txnCount * 10.0; // 0 - 10
if (Math.floor(before) < Math.floor(after)) // crossed a 0 - 10 integer boundary
{
int complete = ((int)Math.floor(after))*10;
String msgProgress = I18NUtil.getMessage(MSG_RECOVERY_PROGRESS, complete);
logger.info(msgProgress);
}
}
// Wait for the asynchronous process to catch up
waitForAsynchronousReindexing();
// Move the start marker on and extend the sample time if we have completed results
if (nextTxns.size() < MAX_TRANSACTIONS_PER_ITERATION)
{
// Move past the query end
if (!lastTxnIds.isEmpty())
{
txnsPerSample += lastTxnIds.size();
lastTxnIds.clear();
}
fromTimeInclusive = toTimeExclusive;
sampleEndTimeExclusive = toTimeExclusive;
}
// Move the end marker on based on the current transaction rate
long sampleTime;
if (txnsPerSample == 0)
{
sampleTime = MIN_SAMPLE_TIME;
}
else
{
sampleTime = Math.max(MIN_SAMPLE_TIME, MAX_TRANSACTIONS_PER_ITERATION
* (sampleEndTimeExclusive - sampleStartTimeInclusive) / txnsPerSample);
}
toTimeExclusive = fromTimeInclusive + sampleTime;
}
// done
String msgDone = I18NUtil.getMessage(MSG_RECOVERY_COMPLETE);
logger.info(msgDone);
}
/**
* Perform full reindexing of the given transaction. A read-only transaction is created
* <b>if one doesn't already exist</b>.
*
* @param txnId the transaction identifier
*/
public void reindexTransaction(final long txnId)
{
if (logger.isDebugEnabled())
{
logger.debug("Reindexing transaction: " + txnId);
}
RetryingTransactionCallback<Object> reindexWork = new RetryingTransactionCallback<Object>()
{
public Object execute() throws Exception
{
// get the node references pertinent to the transaction
List<NodeRef.Status> nodeStatuses = nodeDAO.getTxnChanges(txnId);
// reindex each node
for (NodeRef.Status nodeStatus : nodeStatuses)
{
NodeRef nodeRef = nodeStatus.getNodeRef();
if (nodeStatus.isDeleted()) // node deleted
{
// only the child node ref is relevant
ChildAssociationRef assocRef = new ChildAssociationRef(
ContentModel.ASSOC_CHILDREN,
null,
null,
nodeRef);
indexer.deleteNode(assocRef);
}
else // node created
{
// reindex
indexer.updateNode(nodeRef);
}
}
// done
return null;
}
};
transactionService.getRetryingTransactionHelper().doInTransaction(reindexWork, true, false);
// done
}
}