You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@jackrabbit.apache.org by Fabrizio Giustina <fg...@gmail.com> on 2005/06/14 15:39:42 UTC
Lucene and deleted nodes
While accessing the results of a jcr search I often experience
problems probably due to the presence, in the lucene index, of deleted
nodes.
This is actually a big problem since a missing reference can block
from accessing all the search results (the error is thrown in
NodeIterator.hasNext, while iterating on the search result).
The following is the stacktrace for the error:
ERROR org.apache.jackrabbit.core.query.lucene.DocOrderNodeIteratorImpl
DocOrderNodeIteratorImpl.java(compare:186) 10.06.2005 17:28:54
Exception while sorting nodes in document order:
javax.jcr.ItemNotFoundException: 2537b990-5387-4b5b-a15a-8db53d4353e1
javax.jcr.ItemNotFoundException: 2537b990-5387-4b5b-a15a-8db53d4353e1
at org.apache.jackrabbit.core.ItemManager.createItemInstance(ItemManager.java:518)
at org.apache.jackrabbit.core.ItemManager.getItem(ItemManager.java:372)
at org.apache.jackrabbit.core.query.lucene.DocOrderNodeIteratorImpl$1.compare(DocOrderNodeIteratorImpl.java:142)
at java.util.Arrays.mergeSort(Arrays.java:1307)
at java.util.Arrays.mergeSort(Arrays.java:1296)
at java.util.Arrays.sort(Arrays.java:1223)
at org.apache.jackrabbit.core.query.lucene.DocOrderNodeIteratorImpl.initOrderedIterator(DocOrderNodeIteratorImpl.java:136)
at org.apache.jackrabbit.core.query.lucene.DocOrderNodeIteratorImpl.hasNext(DocOrderNodeIteratorImpl.java:95)
while doing:
NodeIterator nodeIterator = (javax.jcr.query.QueryResult)result.getNodes();
while (nodeIterator.hasNext()) {
...
I am trying to understand if the problem is due to a someway corrupted
lucene index, or if this situation should be handled by jackrabbit (or
by the user? but it sounds strange since the error is thrown from the
iterator).
Did anyone see a similar problem? How deleted/missing nodes are
supposed to be handled in the search index?
fabrizio
Re: Lucene and deleted nodes
Posted by Fabrizio Giustina <fg...@gmail.com>.
On 6/16/05, Marcel Reutegger <ma...@gmx.net> wrote:
> Are you able to reproduce this error in a single session environment?
> That is, no other session is modifying the workspace at the same time.
> If you do, then this is clearly a bug as you described it. Can you then
> please open a jira issue? Thanks
Hi Marcel, I'll try to set up a Junit test which can reproduce the
bug, I will surely open a detailed issue as soon as I can reproduce
it.
> It is also possible that nodes get deleted after the search has
> completed calculating the uuids of the result nodes.
Nope, those nodes have been deleted days ago and the query still return then :/
This happens on a pretty big repository (the zipped repository -using
the standard jackrabbit filesystem- is > 40 Mb), and no other
inconsistency/corruption has been reported till now.
> Do you know if you get same error when you simply execute the query
> again? This would indicate that the index is really out of sync with the
> workspace.
Yes, the missing UUIDs always show up in the results... search is
definitively broken if any of the removed nodes is picked up
>
> regards
> marcel
>
thanks for your answer
fabrizio
Re: Lucene and deleted nodes
Posted by Marcel Reutegger <ma...@gmx.net>.
Hi Fabrizio,
Fabrizio Giustina wrote:
> While accessing the results of a jcr search I often experience
> problems probably due to the presence, in the lucene index, of deleted
> nodes.
Are you able to reproduce this error in a single session environment?
That is, no other session is modifying the workspace at the same time.
If you do, then this is clearly a bug as you described it. Can you then
please open a jira issue? Thanks
It is also possible that nodes get deleted after the search has
completed calculating the uuids of the result nodes.
The current implementation of the query result calculates the actual
order of the result nodes when hasNext() is called for the first time.
That means one of the result nodes might get deleted in the meantime.
A possible solution to this is probably to fetch all the nodes in in the
query result right from the beginning when it is created by the query.
But I have to think about this more thoroughly if this really solves the
issue.
> This is actually a big problem since a missing reference can block
> from accessing all the search results (the error is thrown in
> NodeIterator.hasNext, while iterating on the search result).
> The following is the stacktrace for the error:
Do you know if you get same error when you simply execute the query
again? This would indicate that the index is really out of sync with the
workspace.
> ERROR org.apache.jackrabbit.core.query.lucene.DocOrderNodeIteratorImpl
> DocOrderNodeIteratorImpl.java(compare:186) 10.06.2005 17:28:54
> Exception while sorting nodes in document order:
> javax.jcr.ItemNotFoundException: 2537b990-5387-4b5b-a15a-8db53d4353e1
> javax.jcr.ItemNotFoundException: 2537b990-5387-4b5b-a15a-8db53d4353e1
> at org.apache.jackrabbit.core.ItemManager.createItemInstance(ItemManager.java:518)
> at org.apache.jackrabbit.core.ItemManager.getItem(ItemManager.java:372)
> at org.apache.jackrabbit.core.query.lucene.DocOrderNodeIteratorImpl$1.compare(DocOrderNodeIteratorImpl.java:142)
> at java.util.Arrays.mergeSort(Arrays.java:1307)
> at java.util.Arrays.mergeSort(Arrays.java:1296)
> at java.util.Arrays.sort(Arrays.java:1223)
> at org.apache.jackrabbit.core.query.lucene.DocOrderNodeIteratorImpl.initOrderedIterator(DocOrderNodeIteratorImpl.java:136)
> at org.apache.jackrabbit.core.query.lucene.DocOrderNodeIteratorImpl.hasNext(DocOrderNodeIteratorImpl.java:95)
>
>
> while doing:
> NodeIterator nodeIterator = (javax.jcr.query.QueryResult)result.getNodes();
> while (nodeIterator.hasNext()) {
> ...
>
> I am trying to understand if the problem is due to a someway corrupted
> lucene index, or if this situation should be handled by jackrabbit (or
> by the user? but it sounds strange since the error is thrown from the
> iterator).
This is because the nodes are resolved lazily from uuids returned by the
query result.
> Did anyone see a similar problem? How deleted/missing nodes are
> supposed to be handled in the search index?
well, the index and the workspace should actually be in sync at any
time. so, this shouldn't happen. if it does, then its a bug.
but as mentioned above, it is possible that result nodes get deleted in
the meantime and because of the lazy loading you might run into this error.
regards
marcel