consider mounting data folder as sub-directory of `pacta-data/`

[cjyetman]

The current docker-compose.yml expects that you point to a local pacta-data/ directory which should have a sub-directory named for the desired timestamp/quarter, e.g. pacta-data/2021Q4. Given our current strategy of building PACTA data directories, I think it might be more natural to point directly to that sub-directory, since one might have a directory with numerous sub-directories each containing different builds/versions of PACTA data, but you really only want/need one, and it could be confusing which one is being used.

So instead of this in the docker-compose.yml https://github.com/RMI-PACTA/workflow.prepare.pacta.indices/blob/d93b71c58659f4ebdd894bf358fb3c5b110d01dc/docker-compose.yml#L8-L9

and this in the .env

PACTA_DATA_PATH=PATH/TO/pacta-data
R_CONFIG_ACTIVE=2022Q4

we might have this is the docker-compose.yml

    volumes:
      - ${PACTA_DATA_PATH}:/pacta-data/2022Q4

and this in the .env

PACTA_DATA_PATH=PATH/TO/pacta-data/2022Q4_20240215
R_CONFIG_ACTIVE=2022Q4

The only problem is that the /pacta-data/2022Q4 in the docker-compose.yml would need to adapt to whatever timestamp/quarter you're targeting, and I'm not sure how best to resolve that.

AB#10133

[cjyetman]

Reopening because #87 reveals this is still not adequately fixed. If you have R_CONFIG_ACTIVE=PA2024CH in your .env, that will activate the PA2024CH config, but it will also set the path that the PACTA data is mounted to as /pacta-data/PA2024CH/, based on target: /pacta-data/${R_CONFIG_ACTIVE} in the docker-compose.yml

[jdhoffa]

Hmm not ideal. I will open a hotfix to get it working again, and open an issue to track progress on designing a more appropriate solution.

[jdhoffa]

Ah so it turns out that this issue is already the "follow-up issue" i was going to open haha. Totally agree with the strategy proposed here.

(I think @AlexAxthelm closed this issue before I had a chance to read it XD)

[AlexAxthelm]

Hmm. I remember thinking when I saw that R_CONFIG_ACTIVE in a path that it would probably give us grief at some point. wasn't expecting it within a week.

[jdhoffa]

😂 for real I don't think such a big deal to solve though tbh

[cjyetman]

I think the logic is:

• the target path inside the image needs to be parameterized in the docker-compose.yml to whatever is the relevant quarter
• the docker-compose.yml only gets "input" from environment variables (although docker-compose CLI could likely take params as well)
• a new/additional param in the .env is necessary, e.g. PACTA_DATA_TIMESTAMP, so that it can be distinguished from R_CONFIG_ACTIVE (which could = e.g. 2023Q4, but could also = e.g. PA2024CH)

[jdhoffa]

The issue there is that it conflicts with the assumption that config.yml should set that information somehow

(e.g. it would be possible in that situation that PACTA_DATA_TIMESTAMP point to 2023Q4, R_CONFIG_ACTIVE points to PA2024CH BUT in config.yml that PA2024CH points to 2024Q4 or something like that)

seems like an unnecessary maintenance burden, so I am going to think it through more cautiously.

[cjyetman]

yeah, I guess that's what I meant... the PACTA data timestamp is necessary in the docker-compose step, and since that takes input from env vars not the config, the PACTA data timestamp kinda needs to be defined in the env vars not the config

[jdhoffa]

I think one of the tricky parts is that workflow.transition.monitor expects a directory called 2023Q4 but pacta-data directories are now named e.g. 2023Q4_20240229T200630Z

So already, I need to mount in the contents of a directory into a newly named directory. This is why I can't just mount 2023Q4_20240229T200630Z into pacta-data which I think would be the most preferable approach.

Thoughts?

[cjyetman]

I think one of the tricky parts is that workflow.transition.monitor expects a directory called 2023Q4 but pacta-data directories are now named e.g. 2023Q4_20240229T200630Z

yes, agree 💯, that is the crux of it

Though the additional thing that plays a role is, you may need the mounted directory to look like pacta-data/2022Q4 or pacta-data/2021Q4 depending on what quarter you run. That's a holdover from the olden days when we allowed a TM Docker image to be able to run multiple timestamps/quarters. Generally speaking, we're not doing that anymore, but the capability is still there. I think in the future, e.g. with workflow.pacta, that might not be an issue anymore, because I designed it to have the data outside the Docker image, so it can be mounted where/however you want and the input config determines where the code should look for it, e.g there's no strong restriction like pacta-data/2023Q4, it could be /2023Q4_20240229T200630Z as long as the config has "data_dir": "/2023Q4_20240229T200630Z"

[jdhoffa]

Yup totally. I think all of this points to prioritizing superseding it with workflow.pacta (i know you both agree, just saying it aloud).

I am tempted to not work much on this issue until after that is done/ after CH COP is delivered tbh (I don't want to jump through a bunch of hoops with an odd architecture if we have the opportunity to just update the architecture instead)