Page MenuHomePhabricator

Echo: don't delete every event when reaching the 2,000 limit
Open, Needs TriagePublic

Description

At the moment, only the last 2,000 notifications are kept. The otherones are deleted (Gerrit).

Based on this conversation, some notifications should be kept:

  • The number of thanks could help understanding social issues.
  • A list of mentions is useful if you can't find a discussion you were pinged in.

Event Timeline

Restricted Application added a subscriber: Aklapper. · View Herald Transcript
MMiller_WMF subscribed.

Thanks for filing this, @Trizek-WMF. We'll need to wait to see if this becomes a widespread issue for many users before spending any time with it, so I'll put this in "Revisit".

Notifications are messages for specific individual editors. The notifications which should be kept are the ones which the specific individual editor has not requested to be deleted. No reason has been given for this action whatsoever. What is the motivation here? I want to see a practical reason for why you want to delete other peoples messages. No, the fact that they will be kept indefinitely is not a practical reason. Everything on Wikipedia is kept indefinitely unless it is against site policy or outside the scope. Allowing people to manage their own messages is definitely not outside the scope. We shouldn't have to explain that. You wouldn't be very happy if Outlook or Google mail started to delete your messages. Come on now, don't be deleting peoples messages. You are already auto deleting them. Never delete a message until the person that message is for presses a button which says, "Press this to delete these messages". That is the function you should be considering your next step here.

I'm sorry if I seem condescending but this is a condescending suggestion. Rather than think little of you, this is totally like something I might decide was a good idea myself. You are busy people on this stuff. You don't sit around wondering if you should wonder what to do. You just try to select stuff and do it. However, don't delete the notifications/messages. I've posted this a couple of times and nobody has affirmed it makes sense, so maybe you just aren't quick to respond over here, and I think I detect a tone in my response. Honestly, I've been wondering if recent events haven't been similar to when the whole world went to war, that it affects us in an anonymous internationally collaborative scheme, tensions fray without even thinking about them. I don't mean to be rude. Also, I've never done a single thing to help the development of this software. I am certainly not the boss around here or anything totally stupid, because I do feel like I am demanding on this suggestion... a sense of urgency which was probably not necessary.... but all the same, please don't delete peoples notifications. Instead, please add buttons and features that allow them to much easier manage the deletions themselves... Slightly more difficult yes, but ultimately more functional...

"Waiting to see if this becomes a widespread issue." Meanwhile data is being irretrievably lost,.

So notifications not marked as read can get lost if one has more than 2000 notifications?! That's a huge problem. Is there a way to at least see if one is affected? I'd like to prevent this data loss because I need to get back to many of my unread notifications and not lose them.

So notifications not marked as read can get lost if one has more than 2000 notifications?! That's a huge problem. Is there a way to at least see if one is affected? I'd like to prevent this data loss because I need to get back to many of my unread notifications and not lose them.

FWIW, We bumped that number to 10,000 but we delete all (read and unread) notifications that are older than five years. See T383948: Echo notification expiration should be time-based, not hardcoded 2000

That's great news regarding the count limit, thanks. (The 5 years limit is still an issue for people who stay off or largely off the sites for some years but then come back.) Is there any tool that can check whether one has any unread notifications that are approaching 5 years of age or the 10,000 notifications ago limits (or to see how far from these one is)? If not, it would be useful but maybe would go into a separate issue as a feature request.

(The 5 years limit is still an issue for people who stay off or largely off the sites for some years but then come back.)

I honestly don't think any mention or link or message on talk page that is five year old is of any value. The notifications have ephemeral nature and old notifications are currently a gigantic burden on the infrastructure (more than half of a database cluster is just old notifications)

@Ladsgroup This depends on how one uses notifications and varies per case or types of cases. I can guarantee you 5-years old notifications can have a large value. I'm using unread notifications as basically a form of todos so for example if I made a post about some technical need on Commons VP/Technical and got a notification about it being archived, I'll get back to it for example by creating a Community Wishlist wish for the functionality if I think it would be very useful to many or ask about it at a more suitable place like say Wikidata request a SPARQL query or build the tool myself or whatever. That's just one example. It's important to not mistake how one thinks people are using a feature or how one is using a feature oneself (eg just to track Wikipedia article talk page discussions) with the actual complete usefulness/use of sth for all contributors. Notifications don't just have ephemeral value.

Better solutions would be deleting just particular kinds of notifications that hit such limits such as the Thanks notifications or to restructure how they are stored. For example, one could make old ones not display in the dropdown but still be accessible from some /Notifications page. One could also just make it possible to download all 'archived' notifications that hit some limits and the notifications being inaccessible otherwise. All better than having people risk data loss and creating working on flawed assumptions such as that no community contributors take some years off the site and then return or that old notifications can't be valuable.