AccountDeletionWorker locking up sidekiq

[thatcosmonaut]

Expected behaviour

Sidekiq should not have all workers locked up by AccountDeletionWorker running forever

Actual behaviour

Sidekiq has all workers locked up by AccountDeletionWorker running forever. There are hundreds of jobs queued all with the same parameters: 43875, {"reserve_username"=>false, "skip_activitypub"=>true}

When a worker starts this job, it gets stuck forever.

Steps to reproduce the problem

Just running the server normally.

Specifications

Note: we are running glitch-soc
Commit: ec9ec81

[thatcosmonaut]

I dug into the rails console and got this info, maybe it will help?

irb(main):029:0> Sidekiq::Queue.new('pull').find { |j| j.item['class'] == 'AccountDeletionWorker' && j.item['args'][0] == 43875 }
=> #<Sidekiq::Job:0x000055762f1f17d8 @args=nil, @value="{\"retry\":true,\"queue\":\"pull\",\"class\":\"AccountDeletionWorker\",\"args\":[43875,{\"reserve_username\":false,\"skip_activitypub\":true}],\"jid\":\"503fe0a39624d84b475e6d5a\",\"created_at\":1608438017.8601284,\"enqueued_at\":1608443329.3542883,\"error_message\":\"Attempted to destroy a stale object: Poll.\",\"error_class\":\"ActiveRecord::StaleObjectError\",\"failed_at\":1608443282.46411,\"retry_count\":0}", @item={"retry"=>true, "queue"=>"pull", "class"=>"AccountDeletionWorker", "args"=>[43875, {"reserve_username"=>false, "skip_activitypub"=>true}], "jid"=>"503fe0a39624d84b475e6d5a", "created_at"=>1608438017.8601284, "enqueued_at"=>1608443329.3542883, "error_message"=>"Attempted to destroy a stale object: Poll.", "error_class"=>"ActiveRecord::StaleObjectError", "failed_at"=>1608443282.46411, "retry_count"=>0}, @queue="pull">

[ClearlyClaire]

I cannot find commit ec9ec81, but this issue doesn't ring a bell, so I don't think that's something we solved recently anyway.

The DeleteAccountService seems to fail because of ActiveRecord::StaleObjectError: “Attempted to destroy a stale object: Poll.”. I will investigate how it's possible and how it could be fixed.

[ClearlyClaire]

I couldn't find any way to reproduce it, and I can't see how anything but a very specific race condition could cause this, but does the following help?

diff --git a/app/models/poll.rb b/app/models/poll.rb
index b5deafcc2..e1ca55252 100644
--- a/app/models/poll.rb
+++ b/app/models/poll.rb
@@ -25,7 +25,7 @@ class Poll < ApplicationRecord
   belongs_to :account
   belongs_to :status
 
-  has_many :votes, class_name: 'PollVote', inverse_of: :poll, dependent: :destroy
+  has_many :votes, class_name: 'PollVote', inverse_of: :poll, dependent: :delete_all
 
   has_many :notifications, as: :activity, dependent: :destroy
 

More generally, I think there's a lot of stuff we can delete instead of destroying in the DeleteAccountService, this may also make account deletion faster. On the other hand, while this may be true right now, it might not be in the future…