Too many values for UnInvertedField faceting on field object_closure

[kshefchek]

Solr-dev is currently returning an exception for any biolink call that facets over object_closures:
java.lang.IllegalStateException: Too many values for UnInvertedField faceting on field object_closure

We've appeared to have hit some limit, but it's not obvious from looking at the data:
Solr Production
Total Docs: 37502996
Unique values in object_closure: 4005165

Solr Dev:
Total Docs: 38759008
Unique values in object_closure: 4350271

It's also possible the limit is based on the number of values per a single document, but this is harder to gather without iterating over each document.

Related:
https://issues.apache.org/jira/browse/SOLR-11240

Possible solutions:

• Use facet.method=enum, but this is about 10x slower than method=fc
• Use docValues - https://lucene.apache.org/solr/guide/6_6/docvalues.html, requires small update to golr schema, https://github.com/berkeleybop/golr-schema
• Update solr to 7+
• Reindex and hope for the best, possible that small data changes could affect these counts

cc @kltm @DoctorBud @deepakunni3

[cmungall]

4m is more than I would have expected, iI guess this is because genes, variants, etc make their way into the closure?

[kshefchek]

it does seem high, I could put the values in a file so we can debug.

[kshefchek]

Another solution is to only close over equivalentClass for entities such as genes, variants, etc.

[kltm]

Kent, It may also be that the resources applied to the Solr instance are insufficient to build the necessary caches. Would you have taken a look already at the admin/stats page (e.g. fieldValueCache)? https://wiki.apache.org/solr/SimpleFacetParameters#facet.method https://wiki.apache.org/solr/SolrCaching#fieldValueCache Did this error crop up after a data load or schema change, or has it just not been noticed until now? Cheers,

…

-Seth

On 3/13/19 12:14 PM, Kent Shefchek wrote: Another solution is to only close over equivalentClass for entities such as genes, variants, etc. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#44 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ACkDnA2mkBA_bstaW_o9jjZbnmn96O6xks5vWU4AgaJpZM4btneC>.

[kshefchek]

This occurred after a new data load, and we've tried increasing the memory.

some more stats on the object_closure field:
sum: 819945513
max: 648
mean: 21.15496642741734

All values:
object_closure.txt.gz

Nothing I can see in the logs when the server is starting up.

[kshefchek]

Here is some more info after reindexing:

1. Indexed on monarch5, moved files to solr-dev

• facets on object_closure work

2. Optimized index

• Exception when faceting on object_closure

So oddly this is related to optimizing on the solr-dev VM, versus optimizing on monarch5, I changed this as we were running out of disk space on monarch5.

[kshefchek]

I'm uncertain of the cause of this, but at least understand how to avoid it. Propose we close since it's no longer an issue.

[kshefchek]

this is rearing its head again, although not for every load, and it seems re-indexing fixes it. Will play around with memory config.

[cmungall]

does this help? https://lucene.472066.n3.nabble.com/java-lang-IllegalStateException-Too-many-values-for-UnInvertedField-faceting-on-field-content-td4218197.html

…

On Wed, Dec 30, 2020 at 10:27 AM Kent Shefchek ***@***.***> wrote: this is rearing its head again, although not for every load, and it seems re-indexing fixes it. Will play around with memory config. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#44 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAAMMONDL6L6AEUHRTUB4QLSXNWIHANCNFSM4G5WO6BA> .

[kshefchek]

I have to look back through my notes to see if I tried docValues. This has also been fixed in Solr 7 and above so another reason to push for an update: https://issues.apache.org/jira/browse/SOLR-11240