Hi Tim,

You definitely have something seriously wrong with your setup--some people have reported thousands of documents indexed per second, and I've personal indexed millions of documents sans <commit/>. We've never had a report of documents not being in the index despite being sent to Solr, nor missing index files which weren't caused by corruption (usually related to sudden power failure).

How many simultaneous threads are you using to post to Solr? What version of Solr are you using? Is the method you are using to post to Solr timing out and resending?

I'm afraid without a _lot_ more detail on how you're using Solr, it is difficult to us to help you.

-Mike

On 7-May-08, at 2:15 AM, Tim Mahy wrote:

Hi all,

it seems that we just post to much to fast to Solr.

When we post 100 documents (seperate calls) and perform a commit everything goes well, but as soon as we start sending thousands of documents and than use autocommit or send the commit message we have the situation that there are a lot of documents not in the index although they were sended to Solr ...

has anyone experience with how much documents you can import and at which speed so that Solr stays stable ?

We use Tomcat 5.5 and our java memory limit is 2gb.

Greetings,
Tim
________________________________________
Van: Mike Klaas [EMAIL PROTECTED]
Verzonden: dinsdag 6 mei 2008 20:17
Aan: solr-user@lucene.apache.org
Onderwerp: Re: Delete's increase while adding new documents

On 6-May-08, at 4:56 AM, Tim Mahy wrote:

Hi all,

it seems that we get errors during the auto-commit :


java.io.FileNotFoundException: /opt/solr/upload/nl/archive/data/
index/_4x.fnm (No such file or directory)
      at java.io.RandomAccessFile.open(Native Method)
      at java.io.RandomAccessFile.&lt;init&gt;
(RandomAccessFile.java:212)
      at org.apache.lucene.store.FSDirectory$FSIndexInput
$Descriptor.&lt;init&gt;(FSDirectory.java:501)
      at org.apache.lucene.store.FSDirectory
$FSIndexInput.&lt;init&gt;(FSDirectory.java:526)

the _4x.fnm file is not on the file system. When we switch from
autocommit to manual commits throughout xml messages we get the same
kind of errors.
Any idea what could be wrong in our configuration to cause these
exceptions ?

I have only heard of that error appearing in two cases.  Either the
index is corrupt, or something else deleted the file.  Are you sure
that there is only one Solr instance that accesses the directory, and
that nothing else ever touches it?

Can you reproduce the deletion issue with a small number of documents
(something that could be tested by one of us)?

-Mike




Please see our disclaimer, http://www.infosupport.be/Pages/Disclaimer.aspx

Reply via email to