Different scores depending on cloud node

[Robert Brown <ro...@intelcompute.com>]

Hi,

I have 2 shards, each with 1 replica.

When sending the same request to the cluster, I'm seeing the same 
results, but ordered differently, and with different scores.

Does this highlight an issue with my index, or is this an accepted anomaly?

Example of 8 results:

1st call:

160.2047
160.2047
157.86732
157.86732
157.86732
157.86732
152.6514
152.6514

2nd call:

157.86732
157.86732
157.86732
157.86732
157.64246
157.64246
150.39238
150.39238



Thanks,
Rob

Re: Different scores depending on cloud node

[Shawn Heisey <ap...@elyograg.org>]

On 3/8/2016 6:56 AM, Robert Brown wrote:
> I have 2 shards, each with 1 replica.
>
> When sending the same request to the cluster, I'm seeing the same
> results, but ordered differently, and with different scores.
>
> Does this highlight an issue with my index, or is this an accepted
> anomaly?

SolrCloud's method of operation can result in a different number of
deleted documents on different replicas.  Deleted documents that still
exist within the index can affect scores.  Because SolrCloud picks an
available replica at random to satisfy queries, different requests will
use different replicas.

Distributed IDF, available starting in version 5.0 and described on the
following documentation page, can help even these differences out, but
will not completely eliminate them:

https://cwiki.apache.org/confluence/display/solr/Distributed+Requests

The only way that I know of to completely wipe out these anomalies is to
optimize your collection.  This will completely rewrite the index,
getting rid of deleted documents as it runs, which tends to be very slow
and can be very disruptive to Solr's performance.  It will also block
deleteByQuery requests until it is finished.

Thanks,
Shawn