[Tarantool-patches] [PATCH v2 0/5] send feedback on start and on key events

[Kirill Yukhin]

Hello,

On 11 апр 16:15, Vladislav Shpilevoy via Tarantool-patches wrote:
> Thanks for the patchset, Sergey! Technically the patchset looks bearable.
> Below text is for the people who forced us to do this ticket.
> 
> 
> Disclaimer: this is not toxic. I mostly just ask questions and express
> my opinion.
> 
> I don't like the patch. The hack to workaround some CI issues with a sleep,
> and the behaviour we introduce on the whole. I don't think creation of a
> first space/index makes any sense to track.
> 
> The reason about "going through a tutorial" is a very weak argument. Spaces
> are created not only as a part of a tutorial obviously. And I see no other
> reason for sending these "events".
> 
> This rant is for Mons, Artur, and whoever else pushed sending of that useless
> information and introduction of a **known bug** which we hack with 10s sleep. I
> bet they have no idea how to really use it **exactly**, not in some abstract
> way they will try to figure out later, and they just wanted to send "something"
> to close some story-points.
> 
> We on purpose **push buggy code** only to send these feedback events we don't
> know what to do with.
> 
> Here is what I see in the issue:
> 
> 	In order for us to assess the real situation for the product, it is
> 	necessary to send feedback from the product immediately after the
> 	launch and after the key actions are completed.
> 
> How the hell is it related to a "real situation for the product"? How is it
> "necessary"? What was wrong with the feedback we have now? How are the
> instances working less than 1 hour are so important? And why do you need to
> know exactly when a first space/index was created? Well, even if we need it,
> we could remember the timestamps and send them with the next regular report.
> 
> Another cite:
> 
> 	If Tarantool was installed using a script from the Download page,
> 	then you need to send feedback immediately and further after the
> 	following events are completed:
> 
> 	...
> 
> 	The rest of Tarantool that were downloaded outside of the Download
> 	page does not contain a special meta-file and should be considered as
> 	internal usage and no feedback sent from them. These are usually CI
> 	instances.
> 
> What is that "metafile"? We send the feedback anyway regardless of some metafile
> in this patch, and from where the executable file was downloaded. Which again
> renders this "feature" absolutely useless. Besides, creation of a first index
> and space won't cover instances which don't create any indexes and spaces. For
> example, vshard routers.
> 
> Additionally, implementation of this code stole time from the scaling team on
> writing the code, doing the reviews, doing the review fixes. How is it even
> related to the scaling team at all? Why was it so urgent and more important than
> finishing a fix for a serious bug in the synchronous replication? I would propose
> Artur to send a PR for that next time. Tickets like that should go to the
> wishlist right away, instead of the known crashes and optimizations.
> 
> I hope we will drop these feedback events in the future when will realize it
> does not help anyhow with a problem which can't even be properly formulated,
> and because it introduces a bug we don't know for sure why is happening, and
> how can be fixed except the 10s sleep hack.

This was our commitement, so it should be done. I'll escalate this to our PMs.

--
Regards, Kirill Yukhin