It appears that when newer versions activegraph encounter a deadlock, they hang and no longer responds to signals, meaning the process needs to be kill -9'd. I encountered this in the wild on centos 7 but was able to replicate the hang on osx. This issue did not occur prior to the swap to neo4j-ruby-driver that occurred in 10.x. I do not believe this is related to the thread-saftey issue reported in #1620 as it can occur when a single thread deadlocks with a transaction from another process.

Code example (inline, gist, or repo)

The 'hang' method in the following code run on a fresh db with UUID constraints set for Foo and Bar will replicate the issue.

Note that despite this being a threaded method, this will still occur if you remove the thread block and run 2 instances of the hang method in different processes.

require 'neo4j_ruby_driver'
require 'active_graph/core'
require 'active_graph'
class Foo
  include ActiveGraph::Node
  property :name, type: String
  has_many :out, :bars, type: :foo_bar, model_class: 'Bar', unique: true
  end
class Bar
  include ActiveGraph::Node
  property :name, type: String
  has_one :in, :foo, type: :foo_bar, model_class: 'Foo', unique: true
end
module Hang
  def setup
    ActiveGraph::Base.driver = Neo4j::Driver::GraphDatabase.driver('bolt://localhost:7687', Neo4j::Driver::AuthTokens.basic('neo4j', ''), encryption: false)
  end

  def hang
    setup
    threads = (1..5).map do |n|
      Thread.new do
        (1..100).each do |m|
          ActiveGraph::Base.write_transaction do
            foo = Foo.find_or_create_by(name: 'test-hang')
            Bar.create(foo: foo, name: "#{n}-#{m}-#{Time.now.to_i}")
          end
          puts "thread: #{n} loop: #{m}"
        end
      end
    end
    threads.map(&:join)
  end
end

Runtime information:

OS: Found initially on Centos7 replicated on OSX 11.6
Ruby: 2.7.2
Neo4j database version: 3.5.19-1
acrivegraph gem version: 10.1.0
neo4j-ruby-driver gem version: 1.7.5 (ffi)