[Talk-ca] openstreetmap.org tile rendering unreliable

[Paul Norman]

There's been a fair amount of information given in this thread. 
Unfortunately, most of it is wrong, leading to much confusion.

In no particular order, I want to address a few points

The degraded service on July 18th is unrelated, as it only impacted the 
standard tile layer for 2.5 hours. It's best not to bring up unrelated 
events, as they only lead to confusion.

Several people have been talking about the north american render server 
(pyrene) without even checking that Martin is being served by that 
server! It's bad to speculate without information, as that only leads to 
confusion. I would guess it's about a 50% chance they are being served 
from pyrene, and 50% from the European servers.

Even if they're on pyrene, it doesn't establish that it's capacity has 
anything to do with the problems experienced. Although it's true that 
pyrene no longer has the capacity to serve all of the traffic from the 
Americas, it isn't being sent all of that traffic. It is at peak 
capacity between 8 AM and 6 PM EDT, and during that time the main impact 
is not that users see empty tiles, it's that the map is not up to date.

It is better to ask more questions and not have an answer then to give a 
possibly wrong answer and misinform someone, which is what has happened 
here.

There's two questions that could be asked as follow-up, which can then 
guide towards a correct diagnosis.

The first is for information from the 
https://tile.openstreetmap.org/cgi-bin/debug page. This will reliably 
get you a user's IP and what cache POP they are being served from. This 
is very limited information, and in most cases I would go for the second 
option.

The second is to get the user to use the browser developer tools to see 
a) if the tiles are timing out or error codes being returned, b) what 
tiles in particular are showing problems, c) what the POP and render 
server for the problem tiles is, using the headers. It's important to 
remember that this render server can be a different one for each tile, 
and will not be the same as the debug page.

I asked the first because this is enough information for me to confirm 
or disprove a couple of theories. I'm not going into details on those 
theories, because there's no point confusing a user with possibly 
incorrect theories.

On 2022-07-25 5:06 p.m., Pierre Béland via Talk-ca wrote:
> Daniel,
>
> Les tuiles z13+ sont mise-à-jour assez rapidement. Par contre, 
> compte-tenu de la surcharge des serveurs de tuile, la décision prise 
> est de mettre-à-jour moins fréquemment les tuiles z12-.
>
> Et pour montrer encore la complexité à résoudre de tels problèmes, il 
> faut lire le billet 
> https://github.com/openstreetmap/operations/issues/625#issuecomment-1146706362
> *tomhughes <https://github.com/tomhughes> * commented on 27 May 
> <https://github.com/openstreetmap/operations/issues/625#issuecomment-1139501453> 
> où on parle entre autres des règles de rendu
>
>    The north american render server is quite busy (at least relative 
> to the others) and
>    it will prioritise rendering tiles which don't exist yet over 
> rerendering out of date ones when it is busy.
>
>
> Pierre
>
>
> Le lundi 25 juillet 2022 à 16 h 39 min 19 s UTC−4, Daniel @jfd553 
> <jfd553 at hotmail.com> a écrit :
>
>
> Merci Pierre,
>
> Malgré avoir vidé la cache de mon navigateur, certaines régions ne 
> sont toujours pas rendues après une semaine. Un nouveau serveur sera 
> le bienvenu.
>
> Daniel
>
>
> _______________________________________________
> Talk-ca mailing list
> Talk-ca at openstreetmap.org
> https://lists.openstreetmap.org/listinfo/talk-ca
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/talk-ca/attachments/20220726/80c183a7/attachment-0001.htm>