We are seeing inconsistent results for a get on one of our documents. When
I tried a get using the doc id, sometimes the document would come back and
sometimes it would not. We are using routing for this index but this
happens even when specifying the right routing key. To try and fix the
problem, I reindexed the doc to try and fix the problem but I am still
seeing inconsistent results. A document is now consistently returned but
when the original doc is returned, it has version set to 2. When the new
doc that was indexed is returned, version is set to 1.
Do you index with replica level > 0?
How many nodes?
Do you execute GET against same node or against different nodes?
Does searching the doc _id work?
Is it only this document, and others are ok?
Are are messages in a server logfile?
Does preference "primary" make a difference?
We are seeing inconsistent results for a get on one of our documents.
When I tried a get using the doc id, sometimes the document would come
back and sometimes it would not. We are using routing for this index but
this happens even when specifying the right routing key. To try and fix
the problem, I reindexed the doc to try and fix the problem but I am still
seeing inconsistent results. A document is now consistently returned but
when the original doc is returned, it has version set to 2. When the new
doc that was indexed is returned, version is set to 1.
Index is set to one replica. 4 data nodes, 3 masters, 1 client node.
Executing the get against a client node on the cluster but masters yield
the same result. Data nodes have http turned off currently. Other docs
seem to work fine but we suspect the problem is more widespread. Is there
a way to search by document version numbers? Searching without routing
returns both docs, with routing only returns 1 (but I can't tell which
one, guessing the version 1 doc). Using preference _primary always causes
the version 1 doc to be retrieved. There do not seem to be any useful
messages in the cluster logfiles. We are planning on upgrading to 1.2 soon.
Is there a possibility this problem will be fixed if we upgrade?
-drew
On Thursday, August 7, 2014 3:33:56 PM UTC-6, Jörg Prante wrote:
Do you index with replica level > 0?
How many nodes?
Do you execute GET against same node or against different nodes?
Does searching the doc _id work?
Is it only this document, and others are ok?
Are are messages in a server logfile?
Does preference "primary" make a difference?
We are seeing inconsistent results for a get on one of our documents.
When I tried a get using the doc id, sometimes the document would come
back and sometimes it would not. We are using routing for this index but
this happens even when specifying the right routing key. To try and fix
the problem, I reindexed the doc to try and fix the problem but I am still
seeing inconsistent results. A document is now consistently returned but
when the original doc is returned, it has version set to 2. When the new
doc that was indexed is returned, version is set to 1.
Searching without routing returns both docs, with routing only returns
1 (but I can't tell which one, guessing the version 1 doc). Using
preference _primary always causes the version 1 doc to be retrieved. There
do not seem to be any useful messages in the cluster logfiles. We are
planning on upgrading to 1.2 soon. Is there a possibility this problem will
be fixed if we upgrade?
Searching without routing returns both docs, with routing only returns
1 (but I can't tell which one, guessing the version 1 doc). Using
preference _primary always causes the version 1 doc to be retrieved. There
do not seem to be any useful messages in the cluster logfiles. We are
planning on upgrading to 1.2 soon. Is there a possibility this problem will
be fixed if we upgrade?
Apache, Apache Lucene, Apache Hadoop, Hadoop, HDFS and the yellow elephant
logo are trademarks of the
Apache Software Foundation
in the United States and/or other countries.