Finch and Raven not creating their wps_output under /bird_name

[tlvu]

Migrated from old PAVICS https://github.com/Ouranosinc/PAVICS/issues/151

Does not cause problem now but might cause problem later when we want to restrict access to some wps output from these birds (for privacy reason). Older birds (flyingpigeon, mallefowl, ...) are following this schema.

See discusstion in https://github.com/Ouranosinc/PAVICS/pull/150#discussion_r305084199

[huard]

Has this been fixed with the cookie-cutter template update ?

[tlvu]

Nope. It's a deployment config. It's probably directly in here, unless pywps do not have that config yet or the config do not support giving a initially non-existing dir.

[huard]

@dbyrns Is this required for cowbird ?

[dbyrns]

Cowbird will not handle process outputs so the quick answer is no. On the other hand wps_outputs is served by nginx and thredds so the structure could be "visible". We better have something well organized. Permissions management may be harder also if everything is mixed under the same folder. To fix this, you probably don't have to update the bird directly. Weaver has been "fixed" simply by updating the volume mount in compose by defining the bird name on the server side. See : https://github.com/bird-house/birdhouse-deploy/blob/feature/DAC-300-weaver-compose/birdhouse/components/weaver/default.env#L54

[huard]

Looking at the docker compose, here are the volumes that are mounted:

• malleefowl: wps_outputs:/opt/birdhouse/var/lib/pywps/outputs
• hummingbird: wps_outputs:/opt/birdhouse/var/lib/pywps/outputs
• flyingpigeon: wps_outputs:/data/wpsoutputs
• finch: wps_outputs:/data/wpsoutputs
• raven: wps_outputs:/data/wpsoutputs

I'm not seeing where malleefowl and hummingbird actually configure their outputs to their bird name directory.

[dbyrns]

@huard, old birds were using a buildout recipe. The output location is defined here : https://github.com/bird-house/birdhousebuilder.recipe.pywps/blob/master/birdhousebuilder/recipe/pywps/pywps.cfg#L28

[huard]

Ah thanks !

So am I right thinking that we could simply do something like this in the compose:

finch: 
...
volumes:
   - wps_outputs/finch:/data/wpsoutputs

[dbyrns]

wps_outputs is a named volume, I don't think you can append a directory to it. Since wps_outputs is assumed to be under /data (set by $DATA_PERSIST_ROOT) you could do the same as weaver : See https://github.com/bird-house/birdhouse-deploy/pull/114/files#diff-bf078310de48415c354a40f99e18c9a5d5145c603579e8d29110585977aa42bfR57 and https://github.com/bird-house/birdhouse-deploy/pull/114/files#diff-a633f61efffbef9b534cc7bbfa6c4bca0acf906deb9bd03908e4645e47a31000R70 and define

finch: 
...
volumes:
   - ${DATA_PERSIST_ROOT}/wps_outputs/finch:/data/wpsoutputs

[tlvu]

wps_outputs is a named volume, I don't think you can append a directory to it.

Exact !

Since wps_outputs is assumed to be under /data (set by $DATA_PERSIST_ROOT)

The /data/wpsoutputs part is inside the container, not on the host. $DATA_PERSIST_ROOT is on the host side.

Remember this persistance is using data-volume, not direct bind mount from the host.

I was lost with the buildout magic. But I think the cleanest is to somehow reproduce that magic.

I'll try to give it another shot, unless you know more about buildout that you already know that recipe can not be reproduced outside of buildout. Let me know.

[tlvu]

@huard, old birds were using a buildout recipe. The output location is defined here : https://github.com/bird-house/birdhousebuilder.recipe.pywps/blob/master/birdhousebuilder/recipe/pywps/pywps.cfg#L28

Oh wait, isn't this just a wps.cfg change?! That can easily be reproduced. Thanks for pointing to the right place inside that buildout recipe. It would probably cost me some additional time to trace to this.

[dbyrns]

The /data/wpsoutputs part is inside the container, not on the host. $DATA_PERSIST_ROOT is on the host side.

I know, the finch part doesn't need to exist inside the container, it's only relevant on the host side, where the data must exist along other birds. By doing it like this compose will create the finch directory on the host side and mount it at the same place it used to inside the container and your PR (https://github.com/bird-house/birdhouse-deploy/pull/203/files) will not be required at all.

[tlvu]

The /data/wpsoutputs part is inside the container, not on the host. $DATA_PERSIST_ROOT is on the host side.

I know, the finch part doesn't need to exist inside the container, it's only relevant on the host side, where the data must exist along other birds. By doing it like this compose will create the finch directory on the host side and mount it at the same place it used to inside the container and your PR (https://github.com/bird-house/birdhouse-deploy/pull/203/files) will not be required at all.

I understood that. But doing what you propose means changing the left side of the volume mount from a data-volume (wps_outputs:/data/wpsoutputs) to a direct bind mount on the host (${DATA_PERSIST_ROOT}/wps_outputs/finch:/data/wpsoutputs).

This means changing all the existing birds that currently use data-volume to direct bind mount. Also I think we previously choose data-volume instead of direct bind mount because there was no backup need for this kind of data.

This also means the current way Weaver tries to share it's wpsoutput is wrong since the Nginx proxy also use the data-volume https://github.com/bird-house/birdhouse-deploy/blob/d5fcda8a818d1083974e87d406275eed528e243c/birdhouse/docker-compose.yml#L27.

[dbyrns]

Thank you @tlvu, after some DM with you thursday, I agreeded that the wps_outputs volume is mounted under ${DATA_PERSIST_ROOT} only in thredds and we cannot mix and match volume and direct bind mount. The best solution is therefore https://github.com/bird-house/birdhouse-deploy/pull/203

I remembered talking about that with @fmigneault and we were saying that sharing the same volume with all birds is not a very good practice. Sadly some birds expect to create a directory having their bird name. Without that behavior having a named volume per-bird would have solved the issue and be a better practice by isolating each bird.

@fmigneault, after having talk with @tlvu I'm no longer confident that the solution used in Weaver will be correct. The direct bind mount ${DATA_PERSIST_ROOT}/wps_outputs/weaver will probably not be exposed in thredds or nginx since the wps_outputs directory will be replaced by the volume mount. Will need to check that later (another PR 😉).

[fmigneault]

This also means the current way Weaver tries to share it's wpsoutput is wrong since the Nginx proxy also use the data-volume

The direct bind mount ${DATA_PERSIST_ROOT}/wps_outputs/weaver will probably not be exposed in thredds or nginx since the wps_outputs directory will be replaced by the volume mount.

Not true. Have a try: https://host-140-154.rdext.crim.ca/wpsoutputs/weaver/2dd1d8ca-651c-4644-9ac6-b6bf32718457.log

Without that behavior having a named volume per-bird would have solved the issue

I still believe the fix should be on the bird's side. Some flag to create or not the named directory. It should not enforce a specific name, and they should definitely remain distinct volumes.

[tlvu]

This is not yet working, update here https://github.com/bird-house/birdhouse-deploy/pull/231#issuecomment-1029133290