Store geolocations of TES instances about protes

[limarkdcunha]

Yeah got it. Now I have good idea of what should be done.

[uniqueg]

Note that it is easy to set up another database collection via the FOCA-based app config used in proTES (it becomes basically a declarative issue).

[limarkdcunha]

Hi @uniqueg,I would to like to work on this issue but I have some doubts I would like to clear.
As I understood, after the first call we need to store the  geolocations to the database and afterwards simply fetch them from the db.
I am confused as to what is meant by first call here ? Is it first call of session initiated by a new user or the first call of day or of a particular timeframe. Knowing what a first call is will help simplify the process
Also for "fetching the missing geolocations from the remote service and adding them to the database collection" when should one check if we have missing geolocations.

[uniqueg]

Great, @limarkdcunha, please go ahead and assign yourself :)

As for your question: Just check for every domain if it's in the database collection. If it is, then use the stored geolocations. If it isn't, follow the current procedure and then store them in the database collection. So for now let's assume all domain-coordinate pairs are stored in the database forever and are never updated once created (unless the database is wiped). This is of course not suitable for a production setting, but it's pretty much the same configuration we have for tasks, and we can worry about life cycle management for these documents later on.

So you can just follow the way we are dealing with the tasks collection, including its registration via the app configuration in config.yaml that takes care of the boilerplate of setting up and configuring the collection.

[limarkdcunha]

Hi @uniqueg,thanks for the clarification.
From what I understand from the codebase there are two task_distribution functions (random and distance).If the input urls are present in request ,distance based task_distribution is called or else random one is called.

Random TD does nothing but reshuffles 5 domains present in the app context from config.yaml.Now here I am not able to figure out where are the geolocations fetched for a particular domain? As I am able to infer from create_task function of task_runs.py, url itself is used to create a task using tes.HTTPClient.

Also in your first comment you said "In this way, for the vast majority of calls, only input URI geolocations will need to be fetched from the remote service". I find this statement a little confusing as I previously stated that without input_urls distance based task_distribution is not called.Can you please shed some light on this as well.
A final thing I am not able to assign the issue to myself

[uniqueg]

I have assigned you @limarkdcunha.

Geolocations are only used for the distance-based distribution logic. The point is to choose a TES instance that is closest to the inputs. So if there are no inputs, it doesn't make sense to use it, and we just distribute tasks randomly. In that case, geolocations don't play a role.

So geolocations are ONLY used in the distance-based task distribution logic, meaning that for addressing this issue, you will not need to look at the random distribution logic at all.

Does that make it clear?

[uniqueg]

Mental note: Now that we provide the distance-based task distribution plugin (inside proTES, yet still apart), we need to think carefully how to set up the database so that it is not too tightly coupled with the main app db. Ideally, a collection for storing IP geolocations should be created automatically (and only) when the plugin is used.