Ever growing amount of Rolled back Read Transactions

Neo4j Graph Platform Operations
, , ,
1

Hi there,

We are experiencing, since ever, and evergrowing amount of rolled back read transactions. I can't find anything neither in the query.log, debug.log nor Haling Tool (Queries tab) that happens so often or can generate that many reads or operations running.

Here's a plot of the transactions rollback metrics (from Graphite):

image
image907×316 14.7 KB

As you can see, there are no rolled back writes, only read transactions, and grows at a ~3 millon+ per hour.

I'm using the following APOC functions:

  • apoc.util.md5
  • apoc.convert.toJson
  • apoc.convert.fromJsonMap
  • apoc.util.sleep

I have configured Neo4j Streams both as a Sink and as a Source.

streams.source.topic.nodes.conversation-nc-admin-assignment=Assign
streams.source.topic.relationships.conversation-nc-tag=PART_TAG{*}
streams.source.topic.relationships.nubot-tag-message=SENT_BY{*}

I am currently running:

  • Neo4j 3.5.14
  • APOC 3.5.0.7-all
  • Neo4j Streams 3.5.6

I was wondering if the APOC or Neo4j streams could be causing it?

Any help would be appreciated :smiley: