malobeo job evaluation is getting to CPU intensive on the current host. It slows down other services which external users rely on. For this reason a migration to our own machine is needed.
I propose using one of the thinkcentres for this, which we got donated.
malobeo job evaluation is getting to CPU intensive on the current host. It slows down other services which external users rely on. For this reason a migration to our own machine is needed.
I propose using one of the `thinkcentres` for this, which we got donated.
I disabled hydra on the server. Till we have our own buildhost setup the hydra callback ci pipeline should be ignored on prs.
Starting the hydra-evaluator.service leads to consumption of roughly 26GB of RAM. The host itself only has 32GB in total.
I suspect the pull request evaluation to be the cause since this high amount of RAM usage is new. We should investigate if such high RAM usage is also happening on albert once its running.
I disabled hydra on the server. Till we have our own buildhost setup the hydra callback ci pipeline should be ignored on prs.
Starting the `hydra-evaluator.service` leads to consumption of roughly 26GB of RAM. The host itself only has 32GB in total.
I suspect the pull request evaluation to be the cause since this high amount of RAM usage is new. We should investigate if such high RAM usage is also happening on `albert` once its running.
I now just disabled the malobeo2 jobset, which evaluated PRs, leading to a instant drop of RAM consumption.
So hydra is running again, just PR evaluation is disabled at the moment - means autoupdate continues to work as expected.
I now just disabled the `malobeo2` jobset, which evaluated PRs, leading to a instant drop of RAM consumption.
So hydra is running again, just PR evaluation is disabled at the moment - means autoupdate continues to work as expected.
Did you setup the pr evaluation on this host already?
And if yes, how many PRs is it evaluating currently?
> It seems fine on albert for now <img width="553" alt="image.png" src="attachments/1ae0ebaa-8869-4ef3-8cc0-53492b68f502">
Did you setup the pr evaluation on this host already?
And if yes, how many PRs is it evaluating currently?
Yes,
Currently not evaling anything but you can see the peak was 13 G which is a lot but I also changed it to 6 parallels
Wait, isn't albert in the vpn? then we can already switch over by just using that ip
Yes, 
Currently not evaling anything but you can see the peak was 13 G which is a lot but I also changed it to 6 parallels
Wait, isn't albert in the vpn? then we can already switch over by just using that ip
Also, comparing to the cloud server, evals seem to be like 10x faster
Also, comparing to the cloud server, evals seem to be like 10x faster <img width="361" alt="image.png" src="attachments/48efa9d5-0df6-48e9-8766-ce23176b5c5a">
<img width="427" alt="image.png" src="attachments/dedc6610-bf07-4c10-934e-353c3839768e">
Wait, isn't albert in the vpn? then we can already switch over by just using that ip
#170 needs to be merged before vpn will pickup albert.
If you want to start using albert through vpn proxy forward on server (host_config.nix) and vpn would need to be setup and dns configured to the servers public ip.
But this would need to be changed again as soon as albert is setup within malobeo. Therefore i suggest to wait with it till we have it running locally. But if you still want to go ahead with it thats fine for me.
> Wait, isn't albert in the vpn? then we can already switch over by just using that ip
https://git.dynamicdiscord.de/malobeo/infrastructure/pulls/170 needs to be merged before vpn will pickup albert.
If you want to start using albert through vpn proxy forward on server (host_config.nix) and vpn would need to be setup and dns configured to the servers public ip.
But this would need to be changed again as soon as albert is setup within malobeo. Therefore i suggest to wait with it till we have it running locally. But if you still want to go ahead with it thats fine for me.
Uh apparently it worked at some point, theres a new problem now which is probably caused by albert not being reachable for some reason...
Uh apparently it worked at some point, theres a new problem now <img width="1029" alt="image.png" src="attachments/8d7fe8fc-67cd-4691-8878-62acf88cb7cc"> which is probably caused by albert not being reachable for some reason...
This should have been done before the migration. Once this is updated fanny/albert need to be deployed manually and afterwards the autoupdate hopefully starts working.
The caches pubkey was never added to the trusted hosts: https://git.dynamicdiscord.de/malobeo/infrastructure/src/commit/60c12ce7afa8f8e5f7b78a74b5eb035732e8eb36/machines/modules/host_builder.nix#L33
This should have been done before the migration. Once this is updated fanny/albert need to be deployed manually and afterwards the autoupdate hopefully starts working.
The fact its missing makes external requests fail since they redirect to alberts local IP:
I also think the nginx directive on albert it self should not have been removed as it was done here: 103c7730be
It should stay, only forceSSL and enableACME should be disabled. This way albert webinterface can still be used without depending on a secondary host like fanny.
Also you forgot to set `proxy_set_header` :D It needs to be set as it is done for all other nginx virtualHosts on fanny: https://git.dynamicdiscord.de/malobeo/infrastructure/src/commit/60c12ce7afa8f8e5f7b78a74b5eb035732e8eb36/machines/fanny/configuration.nix#L220
The fact its missing makes external requests fail since they redirect to alberts local IP:

I also think the nginx directive on albert it self should not have been removed as it was done here:
https://git.dynamicdiscord.de/malobeo/infrastructure/commit/103c7730beb85da278cf086284cd807f77c5bb76
It should stay, only forceSSL and enableACME should be disabled. This way albert webinterface can still be used without depending on a secondary host like fanny.
But the inbound port 80/443 forwards to fanny anyway so theres no way to reach that nginx directive on albert. Or do you mean setting up services.nginx.virtualHosts."hydra.hq.malobeo.org"?
But the inbound port 80/443 forwards to fanny anyway so theres no way to reach that nginx directive on albert. Or do you mean setting up `services.nginx.virtualHosts."hydra.hq.malobeo.org"`?
No i was talking about the hydra.malobeo.org - but it is not really important for now. When migrating to HAProxy this will be needed again, for now we can leave it as it is.
No i was talking about the `hydra.malobeo.org` - but it is not really important for now. When migrating to HAProxy this will be needed again, for now we can leave it as it is.
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
malobeo job evaluation is getting to CPU intensive on the current host. It slows down other services which external users rely on. For this reason a migration to our own machine is needed.
I propose using one of the
thinkcentresfor this, which we got donated.Hydra Migratointo Hydra MigrationI disabled hydra on the server. Till we have our own buildhost setup the hydra callback ci pipeline should be ignored on prs.
Starting the
hydra-evaluator.serviceleads to consumption of roughly 26GB of RAM. The host itself only has 32GB in total.I suspect the pull request evaluation to be the cause since this high amount of RAM usage is new. We should investigate if such high RAM usage is also happening on
albertonce its running.It seems fine on albert for now
I'll probably go by malo tomorrow and plug it in so we can continue the switch
I now just disabled the
malobeo2jobset, which evaluated PRs, leading to a instant drop of RAM consumption.So hydra is running again, just PR evaluation is disabled at the moment - means autoupdate continues to work as expected.
Did you setup the pr evaluation on this host already?
And if yes, how many PRs is it evaluating currently?
Yes,
Currently not evaling anything but you can see the peak was 13 G which is a lot but I also changed it to 6 parallels
Wait, isn't albert in the vpn? then we can already switch over by just using that ip
Also, comparing to the cloud server, evals seem to be like 10x faster

#170 needs to be merged before vpn will pickup albert.
If you want to start using albert through vpn proxy forward on server (host_config.nix) and vpn would need to be setup and dns configured to the servers public ip.
But this would need to be changed again as soon as albert is setup within malobeo. Therefore i suggest to wait with it till we have it running locally. But if you still want to go ahead with it thats fine for me.
meh not worth it for the day
#170 is merged but fanny will not auto update
Uh apparently it worked at some point, theres a new problem now
which is probably caused by albert not being reachable for some reason...
The caches pubkey was never added to the trusted hosts:
This should have been done before the migration. Once this is updated fanny/albert need to be deployed manually and afterwards the autoupdate hopefully starts working.
Also you forgot to set
proxy_set_header:D It needs to be set as it is done for all other nginx virtualHosts on fanny:The fact its missing makes external requests fail since they redirect to alberts local IP:

I also think the nginx directive on albert it self should not have been removed as it was done here:
103c7730beIt should stay, only forceSSL and enableACME should be disabled. This way albert webinterface can still be used without depending on a secondary host like fanny.
But the inbound port 80/443 forwards to fanny anyway so theres no way to reach that nginx directive on albert. Or do you mean setting up
services.nginx.virtualHosts."hydra.hq.malobeo.org"?No i was talking about the
hydra.malobeo.org- but it is not really important for now. When migrating to HAProxy this will be needed again, for now we can leave it as it is.Done! yay!