Ladsgroup added a comment.

  I do understand your point @Trizek-WMF but I think it's a bit different:
  
  - Read-only limits write (editing, etc.). Deleting the main page (while being 
quite an adventure, I admit) heavily impacts reads and Wikipedia has a massive 
disparity between read and write. Specially the fact that around 3% of all 
reads to wikipedia go to the main page. For example, French Wikipedia yesterday 
had 24K edits yesterday but had 1B page views last month. A napkin calculation 
gives the ratio of 1408 page view to edit for French Wikipedia.
  - We intentionally chose the time that human edits is at the lowest to 
minimize the impact. For example, in the whole minute of 6:00AM UTC yesterday, 
there were only eight edits on French Wikipedia. Six edits the day before and 
so on. I admit it's a bit inconvenient for me to wake up in such hour 
(specially as a night owl) but I personally suggested that time knowing the 
implications.
  - It doesn't really break anything. Bots are expected to wait and retry in 
such cases, pywikibot does it automatically. And both editors (VE and 
traditional) wouldn't lose edits, it just gives you a message like a "hit the 
submit button again in a minute". Very similar workflow to when your edit token 
expires ("session lost" error) which I get way more often than getting 
read-only error (the session lost error can be reproduced when you keep the 
edit tab open for too long). And in thirty seconds, if you try again, it's 
already saved.
  - In SRE practices, we have concept of error budget 
<https://sre.google/sre-book/embracing-risk/#xref_risk-management_unreliability-budgets>.
 We have way more edits failing due to intermittent network issues, race 
conditions, dead locks, etc. Even we just take edits and assume 8 fail every 
day (e.g. read-only taking a full minute and we do it every day which we 
don't), the ratio of failed edits is 0.03% and well within in our error budget. 
A more realistic calculation would be something along the lines 0.0014% or 
14ppm. That's extremely low.
  - The last but not least, we have to increase our maintenance in our 
databases. We are addressing really long-standing high priority issues (e.g. 
links normalization, revision migration, etc.) on top of addressing tech debts 
(e.g. standardizing timestamp fields) on top of regular maintenance (e.g. 
kernel reboots for security updates, mariadb upgrades, OS upgrades, random 
power or network issues, etc.). This year, we did four times as schema changes 
than we did last year. Part of this is inevitable as if we don't address the 
first category (high priority schema improvements), no one would be able to 
edit Wikipedia in two or three years. Commons is already in an extremely 
dangerous state reaching 2TB.
  - We really have to streamline the process of master switchovers, I already 
made it that the task and patches are done automatically (example: T314369 
<https://phabricator.wikimedia.org/T314369>)
  
  P.S. Regarding deleting the main page, A similar story from English Wikipedia 
<https://en.wikipedia.org/wiki/Wikipedia:Don%27t_delete_the_main_page>:
  
  > One admin was discussing the deletion of the main page in IRC and asked if 
the technical ability to delete pages with over 5,000 revisions, like the main 
page, had ever been re-enabled. Another admin (jokingly) commented that he had 
tested it and found that the main page still couldn't be deleted. The first 
admin thought he would test it for himself. The main page got deleted.[1][a]

TASK DETAIL
  https://phabricator.wikimedia.org/T313398

EMAIL PREFERENCES
  https://phabricator.wikimedia.org/settings/panel/emailpreferences/

To: Marostegui, Ladsgroup
Cc: Ladsgroup, Quiddity, Trizek-WMF, RhinosF1, Aklapper, Marostegui, Wangombe, 
Astuthiodit_1, karapayneWMDE, Invadibot, Lectrician1, UOzurumba, Devnull, 
PallaviPatke, caldera, LSobanski, maantietaja, Jgiannelos, Zblace, NavinRizwi, 
Rileych, ItamarWMDE, Nintendofan885, Akuckartz, 50019062, Iflorez, 94rain, 
DannyS712, Nandana, kostajh, Lahi, Gq86, GoranSMilovanovic, chapulina, 
Chicocvenancio, Nattes, QZanden, EnricoCNC, Alfa80, LawExplorer, Minhnv-2809, 
_jensen, rosalieper, Soum213, Taiwania_Justo, Nizil, Scott_WUaS, Ixocactus, 
Thibaut120094, Wikidata-bugs, aude, geraki, Amire80, Gryllida, jeblad, Catrope, 
Lydia_Pintscher, Jsahleen, Darkdadaah, 01tonythomas, ssastry, Nikerabbit, 
Arrbee, santhosh, KartikMistry, Alchimista, Mbch331, Jay8g, Ltrlg, Krenair, 
Legoktm, Tgr
_______________________________________________
Wikidata-bugs mailing list -- wikidata-bugs@lists.wikimedia.org
To unsubscribe send an email to wikidata-bugs-le...@lists.wikimedia.org

Reply via email to