Changed retries and rescued in ActivityPub::ProcessingWorker

[noellabo]

• Reduce the number of retries, as repeated retries are not effective.
• Give up RecordNotUnique immediately.

[Gargron]

I think this is fine but would like opinion from @ThibG as well

[ClearlyClaire]

Hmm… where would ActiveRecord::RecordNotUnique be raised? It seems fine, but it may also hide bugs, I guess…
I think reducing the number of retries should be fine: while temporary errors are possible, they seem unlikely.

[noellabo]

The cause of ActiveRecord::RecordNotUnique has not been identified. May be due to configuration. I'll investigate more closely.

On the other hand, this error report is useful to the developer, but I'm not sure if it helps the server administrator, because the server administrator can't handle it.

P.S.
My server is accessing two replicated PostgreSQL servers through makara from puma and sidekiq located in different VPS operators. I am getting the above error from time to time.

[ClearlyClaire]

Well, having the error visible to administrators would enable them reporting it to us. I'd be more confident with this patch if we knew what caused that error, even though it's unlikely to be a big deal.

[Gargron]

Sidekiq may process jobs before data has replicated in postgresql, leading to some jobs being discarded (because we often catch and discard record not found errors). One of the reasons why my Sidekiq is connected directly to master... i wonder if you experience the same issue?

[noellabo]

I am monitoring for dead jobs where RecordNotUnique occurs.

1. Duplication of index_statuses_on_uri by Announce
2. Duplication of index_favourites_on_account_id_and_status_id by Like

My server has 3 relays connected and the same Object is often reached 4 times in Create or Announce.

• Create did Redis Lock, but Announce did not. Fix to add RedisLock to handle Announce activity #14365
• Like Activity comes from Misskey's reaction. I look forward to the progress of Pleroma-style emoji reactions #13275.
• Most of the other dead jobs in ActivityPub::ProcessingWorker were WebFinger timeouts. Webfinger timeout seems too short : increasing sidekiq error rate caused by PR #13836 #14091

The cause of RecordNotUnique has been resolved, so I will leave it so that I notice the abnormality when it occurs again.

[ClearlyClaire]

Hmm, not sure what to do about the Misskey reactions… maybe just… drop them, since they don't really mean the same thing as a Like?