How can you have a certificate error if you're using S3? I'm sure their certificate is OK...

Hi UnsightlySeagull42 , I think this is a follow-up to an existing discussion, right?

🙂

Hi MagnificentWorm7 , you're right, I'll ask the guys to check...

Hi NastyFox63 ,
You can set the ClearML Server's services.events.events_retrieval.max_metrics_count and services.events.events_retrieval.max_variants_count settings.
Currently, these values are 100 and 100 respectively - this is why you only get 100 plots since you're probably using more than 100 variants (or series as its called when reporting a plot).
The only limitation is that the server cannot fetch more than 10,000 results when querying the ES backend for plots, so th...

SmallDeer34 https://clear.ml/docs is the official documentation website (previous version have moved there as well). For now, we're keeping the old links alive to make sure any old references won't be broken.

The output from the agent-services is normal and does not affect the server

How can you run a browser on the EC2-instance?

Hi CumbersomeFox70 , I think they are ordered by iteration/metric/variant

OK, I'll make sure we put it into the list 🙂

Hi OddShrimp85 , do you mean that in the login screen, when you enter a name, you get an error?

Can you provide some more details - did you use a Helm chart to deploy (if so, which one?), and how did you configure the system?

Hi PleasantGiraffe85 - welcome to ClearML 😄
The ElasticSearch is used to index various aspects of the experiment and devops process, e.g. index experiment-specific metrics and events, execution agents performance reports etc.

Actually, if you use our helm charts, the ES part itself is not that complicated (since it uses the official ES chart), but I get why using k8s can be a bit more than simply using an existing ES installation 🙂

SharpDove45 , can you try this instead:
TRAINS__HOSTS__MONGO__BACKEND__HOST: "\"mongodb://myDBReader:D1fficultP%40ssw0rd@mongodb0.example.com:27017/?authSource=admin\"" TRAINS__HOSTS__MONGO__AUTH__HOST: "\"mongodb://myDBReader:D1fficultP%40ssw0rd@mongodb0.example.com:27017/?authSource=admin\""

Updating that the helm chart has been updated - thanks JuicyFox94 !

You can use the offline mode and later sync the run with the server

not really, any agent can do that. The only thing different about a services agent is that it's usually running several tasks at the same time (glue agent does that anyway), and is usually configured to use cpu-only and not GPU (which you can control in the glue agent as well)

Let me check...

So it's good now?

Hi WackyRabbit7 , thanks for the kind words! We're not nearly perfect, but we try to improve all the time 🙂
As for API reference, see https://clear.ml/docs/latest/docs/references/sdk/task

Can you share an example of a working URL and a non-working URL?

So the basic authentication header is different for the problematic call 🙂

And the question is - who intervenes and adds them? 😕

Yeah, I see what you mean, probably because the output_uri initialization is handled by the Task.init() . Can you please open a GitHub issue?

The Token is for the password, obviously 🙂 (just making sure there's no misunderstanding)

Hi GiddyTurkey39 , are we talking about two different issues? One is the Trains Server on GCP, the other is installing trains on your machine?

Of course, no matter which path you choose, you'll need to make sure it's possible to access Trains Server ports ( 8080 , 8081 , 8008 ) using the server's external IP.
In GCP, you can do that by choosing " Allow HTTP Access " when creating a new instance (easy, but less secure), or set up a more detailed and secure firewall rules that will only allow access to ports 8080 , 8081 and 8008 (you can consult GCP documentation for that).

Yeah, you should run trains-init on your development machine to set up the Trains SDK for using your new server

Usually pands has different versions for different python versions, so most likely the pandas version specified in the requirements can't be located for python 3.7