I need advice! I might of messed my server up.

MagniPhiCat

@jsg said:
Do NOT reboot! Just keep the box alive and running for the moment.

Thank you @jsg ! I was not planning to, it is wayyyy too risky, plus I don't yet have a backup solution right now in case the website goes down after the reboot.

Can it be repaired? Chances are 98% that yes, it can - but not by you. You'd need someone with considerable experience and know-how.

You're most certainly right! I admit it without blushing...

"In a matter of minutes"? I don't think so, at least not always, but still they have a very decent response time.

That's what I did read, word for word (almost)

Unfortunately it seems that @seriesn is going through a troublesome period - the person, not the company from what I know (and I know a bit).

I figured as much. I reckoned it was weird for a reactive provider to not answer a cry of help in 2 days. Hope everything will work out fine.

Our needs:
- several small servers around the globe
- each server with its own task/responsibility (mail, database, files, web...)
- for a website with around 5k users / day worldwide

No problem.

• budget of 60-80$ per month (for around 8 servers)

No problem if VPS are good enough.

Namely...? ??

MagniPhiCat

@jsg said:
If however you want VDS ("dedicated resources") you'll probably have a hard time finding those boxes.

It would be better for stability, right? Do you think it's mandatory to have dedicated resources, or do you rather think that 5k visitors can be handled easily by a VPS without dedicated resources (as long as they are not oversold ofc)?

Out of the 23 providers that I already analyzed a bit (yes, I did a bit of research beforehand, although it will never be exhaustive), only 7 (unless I am mistaken) seem to offer dedicated resources (I like the acronym VDS, but it does not seem widespread):

• netcup: not well rated at all (on HostAdvice and TrustPilot), but excellent prices (dedicated resources only for 2nd tier servers) (only EU) (although I did read some high praise here I think)
• kernelhost: well rated, prices above netcup but still affordable (maybe too much ram though... would be good for greedy mailcow however) (only EU)
• inmotion: extremely expensive, but rather well rated (only US)
• hostinger: more expensive than netcup but still affordable, very well reviewed (EU + US + Asia)
• bluevps: too expensive for my client's budget, averagely rated (EU + US + Asia)
• ovhcloud: only their 2nd tier has dedicated resources, same prices as bluevps, very badly reviewed (worldwide too)
• ionos (1&1): prices above netcup but still affordable, poorly rated (EU + US) (but I don't like 1&1, sorry, mainly the UI and "feel"... I hope nobody reading this is working there ??)

Which leaves (if I'm being picky):

• EU: kernelhost
• World: hostinger

Any other ideas? (even NexusBytes if "fair share" for CPU)
Do you think I'm too picky?

• of course: root ssh access, dedicated IP, port 25 for the mail server (on which I will install MailCow)

No problem, although some providers won't like port 25 (outgoing). How about using @jar's MXroute service? Not expensive and I hear only good things about them.

I don't know this. NexusBytes mail server is not enough for my client (I have already adressed it with them).
Do you have a link for this service? All I know is that Google for Business (5$ / user / month) is too expensive for my client (roughly 10 users, it would eat most of the whole budget...)

• if possible: guaranteed/dedicated CPU

Highly likely not.

Yes, only 7 providers seem to meet this criteria (again, if I'm not mistaken).
Do you think I could do well without dedicated resources? If so, which provider? (I did read that Contabo was supposedly overselling a lot, which caused latencies and things like that)

I can link to my Excel sheet of providers if need be.

Followup question: we need the different servers (worldwide) to share the same storage space/disk/server (I don't know which term applies). I am having a (very!!) hard time finding information on this kind of setup. How should we proceed so that all webservers share a "network-drive" (or something, instead of just a local path on the local drive) to save user-uploaded-files on it? (without needing each webserver to have 500GB of storage and to rsync everything both ways between all webservers)

No problem but might get a little bit more tricky if it's sensitive data.

Could you please elaborate? Please do as if I didn't know anything about sys admin ??
Because it would require a private network to be set up? If so, I have indeed no idea how to do it, and even if it will work with different providers.

MagniPhiCat

@jsg said:
Now some thoughts/ideas/suggestions off the top of my head

Sure, please shoot!!

• There are basically 2 routes, (a) all eggs in one basket, i.e. a provider with presence all over the world, or (b) spreading the eggs over multiple baskets. (a) has the main advantages of, well, only one provider to deal with plus a better starting point wrt your ""private global network" idea, (b) has the advantage of highly likely being cheaper but the disadvantage(?) of having to deal with multiple providers and potentially different panels. I myself would pick (b) but I have quite some experience.

I would pick B too if need be, but I don't know if the NFS + private network thing would work with different providers.

Although there are some providers that are worldwide and meet the criteria, e.g. hostinger! what do you think of it?

(why so much quotation marks around "private global network"? ?? Please use the proper terms, so that I could learn them and adapt my vocabulary! )

• "good providers" - forget it. I know of no provider who is good everywhere. Every provider has some weak points and locations (one of the reasons why I would pick (b)). And yes, that includes NexusBytes (of whom I'm known to be a fanboy). Therefore I do not provide a "good provider" list without knowing your needs and geo-locations better; it just makes no sense.

Thank you for this measured/realistic answer @jsg !

Step 1 (asap): USA only (70% of global traffic is in eastern USA)
- one mail server (maybe with 6-8GB ram because of greedy mailcow)
- one database master server (mariadb ; database is 2GB without log/event tables) (I already have a slave, that I will have to reconfigure obviously)
- one web server (at first) (only nginx+php+redis)
- one NFS (if not too complicated to do it in step 1) (200GB of files as of today)

Step 2 (in a few weeks): EU + Asia
- one web server in EU (mainly western europe)
- one web server in Asia (to cover middle east and asia)
- (CloudFlare?) as a DNS with geo-routing
- (CloudFlare?) as a CDN for accessing (i.e. retrieving using HTTP) media from the NFS

Step 3 (tbd): USA++ (could happen before step 2 if necessary)
- a second web server in USA to handle more load
- a load-balancer in front of both web-servers

Step 4:
- maybe adding a database slave in each geo region for read queries

Do you need more info than that? If so, which one?

• Your budget is good enough for 8 servers/locations.

Thanks! It's the first time someone is telling me this... I am resilient, but I was still beginning to despair...

But: it sounds like you want to spread your (customers) stuff kind of indifferently all over the place equally. Not smart!

Uploaded files will be centralized (just like they are today, actually... but then a CDN will allow for better latency).

As for website visitors, the objective is to leverage geo-routing at the DNS level to route them to the closest server.

I suggest to think about the following: It seems very likely that only a part of your stuff really needs powerful VDS while the bigger part can run perfectly well on smaller (== cheaper) VPS.

Which one, for example?

I reckon that I don't need the same kind of server for a mail server and for a web server, but I was specifically hoping for advice about this ^^ (initially I was hoping to get it from NexusBytes, actually).

If I had to choose, I would do a wild guess, because I don't know better.

I only know that MailCow theoretically requires 6-8GB of RAM (but that it can be lowered if not too many users, and/or if deactivating spamassassin and whatnot, and/or if using some swap instead of the RAM, which is way cheaper).

I reckon a webserver (nginx/php/redis) could do well with 2 cores and 2 GB, but I am not a specialist and could completely be mistaken. Also, it depends on the CPU speed, whether it's guaranteed/dedicated or not, the load, etc. Difficult for me to evaluate. But I reckon that having multiple small web servers is more agile than having a single one that is either too small or too big.

So, it might be smart to get say 2 VDS, say at $20/mo each plus 5 or 6 small VPS in the $5 - $8 range. But that depends on what you need to run/what services to provide. Without providing more information it'll be hard/nearly impossible to provide sound advice.

I think I replied to all your interrogations but please do tell me if I did not!

I also reckoned that the web server in the US would need more horsepower than the one in the EU, because the US have 70% of traffic, hence more load. EU + Asia share 30%.

• Warning: do not fall for "you need Ryzen (or whatever) boxes". Some of your jobs will but most won't. In other words: pick the right tool(s) for the job(s).

Couldn't agree more with the saying! ????

I hope this helps. If you want to continue: your turn.

I asked you a few questions myself. I would be happy to read your response, if you have enough time!

Thank you again for all this @jsg !

MagniPhiCat

(ok, it seems it's because my answer was somehow too long... I had to split it... does everyone experiences this when writing a "too long" answer?)

DataWagon

If you're using an unmanaged server, it is not the providers responsibility to fix issues like what you've described. As long as your VM is up and there are no network / virtualization issues, any software problems are yours to solve.

Almost every VPS provider gives you VNC / ISO mount access to your VM. If there's something wrong with your OS, you'll have to boot to a live OS via ISO and repair it on your own.

MagniPhiCat

@chocolateshirt said:
I believe they only provide unmanaged services. And now you asked them to manage your server for free. If that is the case, you should buy a managed services, dude!

Who said I demanded that? Merely advice for server size/horsepower, or infra setup help/tips, but not (at all) a managed service... dude

deltatux

@MagniPhiCat said:

I don't know this. NexusBytes mail server is not enough for my client (I have already adressed it with them).
Do you have a link for this service? All I know is that Google for Business (5$ / user / month) is too expensive for my client (roughly 10 users, it would eat most of the whole budget...)

If all they need is mail then of course Google Workspace would be overkill, paying for services they don't need. MXRoute can be found here: https://mxroute.com/, not sure if they have any specials right now though.

Other mail solutions from Hostinger & Zoho Mail offer email boxes for as low as $1/month/user with minimum of 5GB per box.

Hxxx

@MagniPhiCat read my response...

MagniPhiCat

@KermEd said:

Isn't an "NFS" a good fit here? Really this must sound like a dumb question, for professionals like you all. I am sorry... I really need help figuring this all out.

It's all good, I'm not good at explaining things today. It's not a dumb question and most professionals don't seem to know how to cluster things properly anyway imho.

It doesn't tell me if it's an "NFS" situation or not 😅

All in one services like Azure make this stupid easy within their service. Outside of this, you'd want to google something akin to 'VPS Private Networks for Businesses' or maybe 'VLAN for Cloud Servers'. Most cloud heavy companies though can afford to do this kind of stuff in house. And so good documentation is hard to find at times.

Thank you! I'll look into it (although you make it feel like it'll be very hard? ^^)

If so, where should I configure this network?

Usually a single provider would be ideal for all your single instance servers and the Private Network

There is only one provider that can meet my requirements (if I am picky and demand "guaranteed/dedicated" resources), and that can be worldwide, it's hostinger...

In this case you'd google something along the lines of Apache configuration for an external Private Network. It's not done often I imagine, and documentation may be sparse.

Thank you for these tips!

Usually you have full redundancy (servers, storage, everything) not too spaghetti style or it becomes very difficult to maintain and scale as it grows.

We don't need too much. Otherwise it would be a "rich person problem" and we would be using some services like you and other people advised (e.g. AWS or Azure).

We only need:

• one master database server
• one NFS storage
• one mail server
• one webserver (nginx+php+redis) for each geo area (first only the US, but then -- in a matter of weeks -- we will add EU and Asia, and use a DNS with geo-routing on top of all this)

I don't think it justifies going the AWS way, especially since my client does not have the budget. But, again, it is not my core job at all. I'm doing the best I can with the requirements I am given.

This helps a lot, what I'd probably do is:

Main Provider:

• Private Network
• DB
• Storage
• Mail Server
  ** These can all be different VPSes, but will be easier to configure being on one provider ad the private network is likely much easier to configure

Then individual providers:

• Web Instance

You then would setup a CDN, and a configuration for the web instances to connect to the private network. It's likely going to involve a lot of whitelisting of IPs and some VPN style configuration setup. I'd also document this really well so in a month you can add a node without needing to think about it much.

I hope this helps a bit!

This helps a lot! This is a lot of sufficiently detailed information, thank you @KermEd !! 🤩👍

This is not the most urgent part for me, but I will still look into it now that I have some leads!

The reason I recommended recreating it as-is and breaking it apart later, isn't because you can't do it all in two days. But so you don't have to if you get a big bandaid in place. But it's just down to choice really

Like skipping the NFS / private network part?

cadddr

If you want to fuck with glibc please atleast do it inside a container first. It seems like you definitely need a managed service.

MagniPhiCat

@deltatux said:
Being a sys admin is a full time job, if your client can't afford a sysadmin, they should be offloading them into the cloud. For example, email hosting can be had for very cheap via services like MXRoute, Hostinger or Zoho Mail. They should also look into PaaS solutions so that you can focus on being a developer while letting others handle the infrastructure.

They don't have that kind of money. I have been managing and setting up everything for them for like 5-6 years now. But yeah, this is not my cup of tea, really.

Docker lets you spin up containerized applications modularly and allows you to connect one container with another. Due to its portability, I'm a big fan of Docker containers, you should really look into it. If you're deploying and maintaining a lot of containers & servers, look into Kubernetes.

Up until now I really didn't have to use such tools (setting up a classic web server was a breeze), but I guess it would come in handy when managing a couple more servers...!

Thank you for this @deltatux

Which ones would you recommend? Reliable and affordable (I could ask my client for a budget increase, maybe I could get it)

You can't go wrong with big players like AWS, Azure & Google Cloud Platform. However, if the big players are too costly, I've heard great things about Linode & Digital Ocean for their cloud instances.

Too costly as well, I'm afraid... (Linode costs a bit less than DO, but is still a bit over the budget)

Oracle Cloud provides an Always Free Tier and their ARM-based instances are great value, you can spin up to 4 ARM-based VMs with up to 24 GB of RAM for free, for many web apps, they are architecture-agnostic, so running on ARM CPUs shouldn't really matter. I personally pair my very cheap VPS with Oracle Cloud's Always Free tier to cover all my project needs so far.

I discovered it a few days ago, actually. It is (extremely) impressive! But some people are saying they wouldn't host anything serious/critical on it (don't know why though). Also, not sure about bandwidth...

Maybe it would help with backup servers or something...

I am using it (personally) for my side projects. CPU is limited, but RAM is awesome!

chocolateshirt

@MagniPhiCat said:

@chocolateshirt said:
I believe they only provide unmanaged services. And now you asked them to manage your server for free. If that is the case, you should buy a managed services, dude!

Who said I demanded that? Merely advice for server size/horsepower, or infra setup help/tips, but not (at all) a managed service... dude

Yes, from this discussion you clearly need a managed services..

Levi

If you feel that provider is not as you expected - immediatelly dump. His problems is not your problems. As for suggestion: you need managed service and that is very clear. You have big mouth, but small hands as they say.

MagniPhiCat

@Hxxx said:
@MagniPhiCat
Keep things simple.
You have only 1 server currently. It is not working properly. You want a managed VPS provider, ok.

No I don't.

I want (because I need) to manage it myself.
I only said that maybe NexusBytes would give me some hints as to how much horsepower to use or how I should think about setting up the infrastructure, which I reckon they can do if they are nice people... Nothing more.

-First find that provider to replace that single server.

I could grab another server on the same provider if need be. The server being (almost) dead is my fault only. My client wants me to spend up to 4 days finding a good reliable / affordable VPS hosting provider.

-Then after that take your time to analyze your needs and your proposed solution. For example you said you want 8 servers, each for one task but then you want all the servers to share a storage and the servers are in different locations. Basically you went from something simple to something really complex with basically no decent budget.

I can't choose my budget...
Also, I don't think it's too complex? Just 3 web servers sharing access to 1 DB, 1 NFS, 1 mail server.

In general all you need is an OpenVPN server (your own), and assuming I understood you correctly and servers are in different location, then your servers connect as clients to that VPN server, and then configure the VPN so that client can talk to each other.

Mmmh interesting! Is it performant enough for real time communication? They just need to access the NFS, really, for write only (or, rather, "mainly").

Thank you for this nice suggestion @Hxxx !! 🤩👍

But I get this feeling you want this big infrastructure just because you want to throw servers to the problem.

No I don't... The problem (with the current server being half-dead) was me making a mistake during an upgrade. How is it relevant?

In reality probably all you need is one server, and make sure you have good backups.

No I don't. I need several servers in several geo locations. I can't do that with one server.

And I need this because my client asked for this.

Having good backups is mostly already done, as I have a real time DB backup and an rsync backup of files (that stopped working 2 days ago because of my mistake).

Granted, I don't have a secondary/backup web server, but I have already addressed the issue, and this is precisely what I am trying to accomplish by having several small servers: in case one goes pouf, another one can replace it immediately.

Assuming your stack is LAMP or LEMP, my recommendation is to contact KnownHost, pinging @jonathanspw, they have the balance between budget and good managed VPS server. Their replies are very fast, so this will help with your anxiety. You can also contact BigScoots, they might not be within your budget but they also take care of customers that want managed VPS.

I don't want a managed VPS, but I will still look into this, thank you!!

(yes, I have nginx+php+mariadb = LEMP)

But essentially if your stack really needs that many servers with such custom setups

Not so custom, actually. 3 web servers with LEMP (trivial ; even 1 would suffice at first). 1 mail server with MailCow (trivial too I guess). 1 DB server with MariaDB (trivial).

Ok, the next part is less trivial... At least for me. Configuring an NFS and a private network for all these servers to talk privately together... But it may be optional at first

For nostalgia...
If this was 2015 I would say contact WiredTree they were probably one of the best managed VPS server company at the time. But sadly it got purchased by LiquidWeb and dissolved . I liked the brand.

Sorry for your loss 😅

And thank you for the kind help! 🙏

Archie

choose serious providers for production:
1)aws google azure alibaba
2)do vultr linode hetzner

MagniPhiCat

@paijrut said:
Agreed with @Hxxx simple is best..
You could just go dedicated server, and throw everything there, setup proper backup server for failover, use cdn for that geo something.. it will work just fine for 5k user/day

What does a CDN have to do with geo-routing users to newly created web servers that are close to them? Did you mean DNS? In any case I would need several servers.

Plus, my client does no longer want "one server that does everything"...

Thank you anyway for the input @paijrut !

MagniPhiCat

@grep said:
This guy is complaining about @seriesn inactivity, when he is being inactive regarding his client and cries when @Falzo basically said "get off your ass and do something". Crazy.

"This guy" does not like being talked down to like that.

What do you know about my client, what I do for them, and what they tell me regarding the situation and what they are asking me?

Basically nothing. So please.... You know the rest of the sentence.

There is a saying in French that goes roughly like this:
"Sometimes it's better to just shut up and risk passing for a fool, than opening one's mouth and not leaving any doubt about that"

paijrut

With this low level of planning and lack of comprehension, imagine what would happen when the storage server fail or there is a network failure on the vpn server.
With that budget you should just keep it simple for now

Hxxx

@MagniPhiCat
One server, LEMP or LAMP, with a CDN provider in front, like CloudFlare free or paid, ... done... I really don't get why you want to have different servers because of regions, that's why CDN providers exist.

Mailing server, fuck that, pay an email service. Want to send transactional emails to your customers, pay Amazon SES , Mailgun or Mandrill (from mailchimp). Want to send email campaigns? Use mailchimp (free or paid). For regular inbox, MXroute or any of the usual.

MagniPhiCat

@DataWagon said:
If you're using an unmanaged server, it is not the providers responsibility to fix issues like what you've described.

Who said it was?

As long as your VM is up and there are no network / virtualization issues, any software problems are yours to solve.

Who said it wasn't?

If there's something wrong with your OS, you'll have to boot to a live OS via ISO and repair it on your own.

Who said I wanted to?

...

MagniPhiCat

@deltatux said:
If all they need is mail then of course Google Workspace would be overkill, paying for services they don't need. MXRoute can be found here: https://mxroute.com/, not sure if they have any specials right now though.

Thank you for the link!

Unfortunately, I already discussed it with my client, and 300 messages / hour is not enough for their needs.

Other mail solutions from Hostinger & Zoho Mail offer email boxes for as low as $1/month/user with minimum of 5GB per box.

Please correct me if I'm wrong, but I fear they don't have a good enough deliverability (or so I heard).

MagniPhiCat

@Hxxx said:
@MagniPhiCat read my response...

Sorry @Hxxx , there is one that I didn't see and didn't reply to? I thought I did 🤔

paijrut

@MagniPhiCat said:

@paijrut said:
Agreed with @Hxxx simple is best..
You could just go dedicated server, and throw everything there, setup proper backup server for failover, use cdn for that geo something.. it will work just fine for 5k user/day

What does a CDN have to do with geo-routing users to newly created web servers that are close to them? Did you mean DNS? In any case I would need several servers.

Plus, my client does no longer want "one server that does everything"...

Thank you anyway for the input @paijrut !

Put the cdn in front of the web server and that geo routing something will be done by the cdn providers..
Your top priority right now is to move your files from that broken by you server as soon as possible, because who knows? There might be problem with electricity and there bam your server have to reboot anytime now. Get the logic? Its your job to convince your client on what needs to be done

jsg

OK, let's get this out of the way first ...

@MagniPhiCat said:
... not well rated at all (on HostAdvice and TrustPilot)

Wrong approach. Those sites are, with rather few exceptions, utterly worthless and even misleading. Let me explain:

It's basically a psycho thing. Very few happy customers go - at all - to spread positive feedback unless expressly asked and even then most won't make the effort. Unhappy customers (as well as refused ones) however love to vote providers down, to criticize them and even to "hunt" them.
Plus 50+% (to put it diplomatically) do not even begin to have the needed knowledge and understanding to hand out grades, and, sorry to all democracy lovers, 85+% of all people wouldn't have an easy time trying to win an IQ contest against a plank of wood.

MagniPhiCat

@cadddr said:
If you want to fuck with glibc please atleast do it inside a container first. It seems like you definitely need a managed service.

I didn't plan to. When I saw the bug, I stopped everything.

I would love not to have anything to do, but a managed infrastructure is out of my client's budget.

chocolateshirt

@MagniPhiCat said:

@deltatux said:
If all they need is mail then of course Google Workspace would be overkill, paying for services they don't need. MXRoute can be found here: https://mxroute.com/, not sure if they have any specials right now though.

Thank you for the link!

Unfortunately, I already discussed it with my client, and 300 messages / hour is not enough for their needs.

Other mail solutions from Hostinger & Zoho Mail offer email boxes for as low as $1/month/user with minimum of 5GB per box.

Please correct me if I'm wrong, but I fear they don't have a good enough deliverability (or so I heard).

You are funny, do you think your email setup have better deliverability than zoho & hostinger?

You say it is trivial things, yes it is very trivial, most member here suggest you to get simpler way, but you choose to make it complicated... Lol

@deank your phrase please..

MagniPhiCat

@chocolateshirt said:
Yes, from this discussion you clearly need a managed services..

You may think I do. But I cannot.

Also, I have been managing, all by myself, several websites for years. Ok, I don't have (far from it) the skills of a sys admin and I often have to google things. But I can be useful anyway.

MagniPhiCat

@LTniger said:
If you feel that provider is not as you expected - immediatelly dump. His problems is not your problems.

It's my fault the server broke. Not the provider's.

As for suggestion: you need managed service and that is very clear. You have big mouth, but small hands as they say.

Aside from the insult, this is not helping. I said multiple times that my client does not have the budget for a managed hosting!

I may need a smaller mouth and bigger hands, but some seem to need reading lessons/glasses.

MagniPhiCat

@Archie said:
choose serious providers for production:
1)aws google azure alibaba
2)do vultr linode hetzner

Sorry, as I said, they are out of the budget (except for Hetzner, but it does not cover Asia, only half of their servers cover the USA, and they have a bad rep -- deserved or not -- on some reviewing plateforms)

bruh21

@MagniPhiCat said:

@jsg said:
If however you want VDS ("dedicated resources") you'll probably have a hard time finding those boxes.

It would be better for stability, right? Do you think it's mandatory to have dedicated resources, or do you rather think that 5k visitors can be handled easily by a VPS without dedicated resources (as long as they are not oversold ofc)?

Out of the 23 providers that I already analyzed a bit (yes, I did a bit of research beforehand, although it will never be exhaustive), only 7 (unless I am mistaken) seem to offer dedicated resources (I like the acronym VDS, but it does not seem widespread):

• netcup: not well rated at all (on HostAdvice and TrustPilot), but excellent prices (dedicated resources only for 2nd tier servers) (only EU) (although I did read some high praise here I think)
• kernelhost: well rated, prices above netcup but still affordable (maybe too much ram though... would be good for greedy mailcow however) (only EU)
• inmotion: extremely expensive, but rather well rated (only US)
• hostinger: more expensive than netcup but still affordable, very well reviewed (EU + US + Asia)
• bluevps: too expensive for my client's budget, averagely rated (EU + US + Asia)
• ovhcloud: only their 2nd tier has dedicated resources, same prices as bluevps, very badly reviewed (worldwide too)
• ionos (1&1): prices above netcup but still affordable, poorly rated (EU + US) (but I don't like 1&1, sorry, mainly the UI and "feel"... I hope nobody reading this is working there ??)

Which leaves (if I'm being picky):

• EU: kernelhost
• World: hostinger

Any other ideas? (even NexusBytes if "fair share" for CPU)
Do you think I'm too picky?

• of course: root ssh access, dedicated IP, port 25 for the mail server (on which I will install MailCow)

No problem, although some providers won't like port 25 (outgoing). How about using @jar's MXroute service? Not expensive and I hear only good things about them.

I don't know this. NexusBytes mail server is not enough for my client (I have already adressed it with them).
Do you have a link for this service? All I know is that Google for Business (5$ / user / month) is too expensive for my client (roughly 10 users, it would eat most of the whole budget...)

• if possible: guaranteed/dedicated CPU

Highly likely not.

Yes, only 7 providers seem to meet this criteria (again, if I'm not mistaken).
Do you think I could do well without dedicated resources? If so, which provider? (I did read that Contabo was supposedly overselling a lot, which caused latencies and things like that)

I can link to my Excel sheet of providers if need be.

Followup question: we need the different servers (worldwide) to share the same storage space/disk/server (I don't know which term applies). I am having a (very!!) hard time finding information on this kind of setup. How should we proceed so that all webservers share a "network-drive" (or something, instead of just a local path on the local drive) to save user-uploaded-files on it? (without needing each webserver to have 500GB of storage and to rsync everything both ways between all webservers)

No problem but might get a little bit more tricky if it's sensitive data.

Could you please elaborate? Please do as if I didn't know anything about sys admin ??
Because it would require a private network to be set up? If so, I have indeed no idea how to do it, and even if it will work with different providers.

You don’t need dedicated resources for a webserver. 5k visitors a day is not that much and if the workload is spread out through 8 machines then you should be good with some pretty light VMs. If you want a “real” production provider, try Hetzner cloud or something along those lines. I would avoid hostinger since a lot of their positive reviews are paid and you have to pay 3 years up front to get anywhere near the advertised prices. I have also used services from @combahton_it with little issue, and the CPU allotment is generous. I have averaged 50+% usage for 24 hours before with no issues, but keep in mind that the cpu is slower than other providers.

Overall you shouldn’t need that much power for these servers. I’ve run a website with likely similar visitor numbers on an oversold 2 core 4gb OpenVZ from @VPSSLIM with no problem for months on end. I only ever had to contact support like once or twice due to degraded network, but no noticeable downtime ever occurred