iwAccess is the reader lock to iwCommit's writer lock - so the scenario
you bring up should be protected - the reader lock is used in only one
place in the class (addDoc), while every other call to openWriter is
protected by the writer lock.
I'd worry more about the case where two add documents hit at the same
time when the indexWriter is null - then, because both calls to
openWriter are protected with the reader lock, you could race - but
that's what the synchronized(this) protects against ;)
It all looks good to me.
- Mark
On 9/9/10 4:17 AM, Bharat Jain wrote:
> Hi,
>
> We are using SOLR 1.3 and getting this error often. There is only one
> instance that index the data. I did some analysis which I have put below and
> scenario when this error can happen. Can you guys please validate the issue?
> thanks a lot in advance.
>
> SEVERE: org.apache.lucene.store.LockObtainFailedException: Lock obtain timed
> out
> : SingleInstanceLock: write.lock
> at org.apache.lucene.store.Lock.obtain(Lock.java:85)
> at org.apache.lucene.index.IndexWriter.init(IndexWriter.java:1140)
> at org.apache.lucene.index.IndexWriter.<init>(IndexWriter.java:938)
> at
> org.apache.solr.update.SolrIndexWriter.<init>(SolrIndexWriter.java:116)
> at
> org.apache.solr.update.UpdateHandler.createMainIndexWriter(UpdateHand
> ler.java:122)
> at
> org.apache.solr.update.DirectUpdateHandler2.openWriter(DirectUpdateHa
> ndler2.java:167)
> at
> org.apache.solr.update.DirectUpdateHandler2.addDoc(DirectUpdateHandle
> r2.java:221)
>
> I think this error can happen in the following scneario
>
> 1. Thread T1 enters the commit method and its actually an optimize request.
> a. T1 gets the iwCommit lock
> b. T1 enters the openWriter() method
> c. T1 does the (writer == null check) -> time x1
>
> 2. Thread T2 enters the addDoc method
> a. T2 gets iwAccess lock
> b. T2 gets the mutex "this"
> c. T2 enters operWriter method
> d. T2 does the (writer == null check) -> time x1
>
> Now after the 1.c is done thread yields and 2.d gets execution and it also
> sees writer as null and so now both threads will try to create the
> indexwriter and it will fail. I have pasted the relevant portion of code
> here.
>
>
> // iwCommit protects internal data and open/close of the IndexWriter and
> // is a mutex. Any use of the index writer should be protected by
> iwAccess,
> // which admits multiple simultaneous acquisitions. iwAccess is
> // mutually-exclusive with the iwCommit lock.
> protected final Lock iwAccess, iwCommit;
>
>
>
> // must only be called when iwCommit lock held
> protected void openWriter() throws IOException {
> if (writer==null) {
> writer = createMainIndexWriter("DirectUpdateHandler2", false);
> }
> }
>
> addDoc(...) {
> ...
>
> iwAccess.lock();
> try {
>
> // We can't use iwCommit to protect internal data here, since it would
> // block other addDoc calls. Hence, we synchronize to protect
> internal
> // state. This is safe as all other state-changing operations are
> // protected with iwCommit (which iwAccess excludes from this block).
> synchronized (this) {
> // adding document -- prep writer
> closeSearcher();
> openWriter();
> tracker.addedDocument();
> } // end synchronized block
>
> // this is the only unsynchronized code in the iwAccess block, which
> // should account for most of the time
>
> }
>
> commit() {
> ...
>
> iwCommit.lock();
> try {
> log.info("start "+cmd);
>
> if (cmd.optimize) {
> closeSearcher();
> openWriter();
> writer.optimize(cmd.maxOptimizeSegments);
> }
>
> closeSearcher();
> closeWriter();
>
> callPostCommitCallbacks();
> }
>
> Thanks
> Bharat Jain
>
