Remote Executors

[proffalken]

OK, so this is quite possibly taking the project way past what it was ever intended to be, but stay with me on this one...

It would be amazing if I could launch "remote executors" and install them on various devices/virtual instances but have them report back to a central Uptime-Kuma instance.

This is a feature that some of the more advanced site-monitoring tools have, and would allow me to spin up instances across the globe to see what the response time was from each region etc.

As a rough outline, I'd probably be looking for the agents to either post their results back to the "primary" setup via HTTPS, or just load the results onto an MQTT Queue and have the "primary" read those results and log them.

Having it as a lightweight app would also mean that I could deploy onto a Raspberry Pi or similar if I didn't want to use virtual machines in a cloud-provider.

[proffalken]

@gaby This isn't about push notifications, it's about being able to deploy multiple instances of Uptime-Kuma that all feed back to a central instance.

The idea is that I could deploy copies of Uptime Kuma into AWS EU-West-1, US-East-2, and APAC-West-1, but have them all report the latency back into an instance in EU-West-2 that shows the actual data, rather than just being able to push notifications from anywhere on the planet back to my browser/mobile device?

I'd rather keep this open if people are happy to do so?

[louislam]

Yes, should keep this open. Don't forget to give a 👍 to your post.

[deefdragon]

Very much bike-sheding but I couldn't resist. Given Kuma=Bear, I propose that we call remote executors (at least in the case of many-to-one) "Cub" instances, with the primary being the "Mother" instance.

[codeagencybe]

I like this idea a lot, as coming from updown.io service.
They also have this option to have checks from places all over the globe and report what kind of latency and apdex score it returns.

But this kind of setup does come with some complex caveats to expect which I also face randomly from time to time from updown.io and others:

• it can result in false negatives warnings and triggering alerts for no reason.
  Then the question is: what do you with that information and how should Uptime Kuma handle/output the metrics?
  Let's say you have 7 locations and 1 or 2 report down while others report up.
  Are you going to show down in statuspage or up? Are you going to show the metrics publicly for all locations so users can interpret themselves?
  Or we need a raft consensus that makes a decision that a certain percentage of checks is required before it change status down/up.

I do like the idea alot as I like it from updown.io, but I think it's a challenge to get something like this developed properly and leave alone deployed properly.
Really curious where Uptime Kuma wants to go with features like this.

[trogper]

I agree it would be nice to have, but I think kuma is supposed to be light-weight, not enterprise-ready. Not many people would use it.

[proffalken]

I agree it would be nice to have, but I think kuma is supposed to be light-weight, not enterprise-ready. Not many people would use it.

I'm talking about something that's optional here, not required - Uptime Kuma would continue to run as a single instance by default, but adding some kind of "scaling" ability would be a configuration/settings flag.

At the moment, I'm just deploying multiple Uptime Kuma installations, but until #898 is in a better state it's very difficult to determine which datacentre/cloud region the metrics and alerts are coming from in Prometheus, and it requires custom configuration of each installation.

[Peppershade]

It would be nice to have a possibility to have a second docker container on a remote location to handle the same checks to prevent false positives, which could be scalable and report back with a voting system. I'd love to replace Uptime Robot at our office, but just one server to handle the checks is not enough for our infrastructure monitoring.

I do use Uptime Kuma for personal servers and for a large Dutch Minecraft server with a complex server infrastructure. So far I am very happy with this platform, and it has a lot of potential.

[proffalken]

I agree it would be nice to have, but I think kuma is supposed to be light-weight, not enterprise-ready. Not many people would use it.

I'm coming back to this comment because there's soon going to be incident management and reporting added to UptimeKuma, which to me moves it firmly into the Enterprise space as far as functionality is concerned - I know very few people who run incident management on their home labs or similar!

With the above in mind, to me the concept of "remote nodes" becomes even more important - the last thing you want to do is declare an incident that your entire infrastructure is down just because one instance of Uptime Kuma lost it's connectivity.

I realise that most will be using additional tooling, but this is just a small thing that would appear to make it a lot easier to monitor disparate systems and consolidate the results in one place.

[temamagic]

it can result in false negatives warnings and triggering alerts for no reason.
Then the question is: what do you with that information and how should Uptime Kuma handle/output the metrics?
Let's say you have 7 locations and 1 or 2 report down while others report up.

It should use something like "paxos distributed consensus algorithm"

It would be cool to have this option.

For example, I check the operation of the service from different servers, but one of them has a network problem. In this case, he will say that "there was a problem." If the check came from 2 or more places, and only 1 server said about the problem, then I would conclude that the problem is on the checking server, but not on the service being checked

[davidak]

@hanisirfan @jmiaje @officiallymarky add your 👍 to the initial issue if you want this feature and please don't spam everyone that is subscribed to this issue with unhelpful comments. That makes it take longer. This applies to GitHub and other places in general!

[MikeBishop]

It seems like there are two different "versions" of this that can be envisioned, one much more complex than the other.

The more complex one is monitoring one endpoint from many perspectives, and then consolidating the perspectives into an up/degraded/down output. Great to have, but there's a simpler version.

If you just want to monitor certain services from certain viewpoints, that's already possible, just clunky. From each peripheral instance, expose a status page. The central instance adds a monitor which queries /api/status-page/heartbeat/ on the remote and checks that everything / a particular service is status=1.

I think a step in this direction would be the ability to add a monitor type which simply mirrors the state of a monitor from another UptimeKuma instance.

[modem7]

One of the major benefits of something like this (which is how I came across this issue), could be multiple hosts with different setups.

E.g.

Primary/master uptime-kuma instance on a VPS somewhere outside the main infrastructure (especially for self hosting).
Secondary uptime-kuma instance with Docker containers and docker endpoint exposed so that it can monitor said containers.
Tertiary uptime-kuma instance with databases which aren't exposed to the public.

Primary is then able to consolidate the information without exposing the services publicly, including the consolidation of maintenance windows utilising a single node.

This would go hand in hand with @MikeBishop's suggestion of mirroring a monitor from a slave instance.

[maxistviews]

I could do this now with something like Home Assistant, but I would prefer to have this option!

[Sid-Sun]

I have added another method to replicator-kuma, and documented the different methods on https://github.com/Sid-Sun/replicator-kuma. I haven't yet tried upgrades but I've been using it with snapshot mode for a few months without issues so far.
P.S. If you plan to use S3, you should go with Cloudflare R2 and I quickly reached free limits on AWS S3. R2 has worked great (and without additional cost).
Edit: Just upgraded and it went as expected.

[NHendriks01]

Any news on this? I'm not into enterprise or anything but would love to run multiple instances. My view on this might be a bit different but also a bit simpler. Multiple nodes that all show the same data but also have a monitor to each other. More or less like a CDN. Now there is no redundancy except building it by hand, which is possible but a bit clunky.

[officiallymarky]

I don't think it is going to happen.

[codeagencybe]

been asking and waiting since 2021, I also don't think it's ever going to happen.
As much as I live uptime kuma and been sponsoring this project, I gave up on this feature and currently looking into OpenStatus which supports multi-zone/regio monitoring out of the box.
Just not entirely sure if this is equally open source like Uptime Kuma, but it does tick several boxes for features that UK is missing for many years and doesn't seem to show any movement or changes coming anytime soon which is a pity.

[CommanderStorm]

I think the basic usecase (monitoring one thing from mutliple sides) behind this feature can also be resolved via ripe atlas.
I know that this is somewhat limited for (very good) ethical+security reasons (f.ex. you cannot http a CP-site from somebody elses' device). I think that the limitations are likely fine for most users.

• Allowing ping/dns/http calls via RIPE ATLAS for distributed network monitoring #4031

I plan on implementing a monitor for this when I have time again.

About the redundancy aspect:
We are not a distributed system. A distributed system requires an entirely different kind of effort to maintian and extend.
This is especially true in terms of

• the consistency/leader election/heartbeat scheduling side (yes, RAFT is possible, but defeats the entire point of being a simple)
• in terms of security/privacy

Given the engineering resources we currently have, I would currently classify this as out of scope.
Using a second instance (or a commercial status page like uptimerobot) to monitor if Uptime Kuma is up is simple enough and does not require this level of complexity.

[ramphex]

Also interested in this. I monitor several networks and it would be nice to have a VM with a kuma instance watching all of the local devices there and reporting back to my main mothership. That would help avoid manually visiting several different kuma instances to check up on every network.