Env var values not being loaded without a config file definition

[joao-zanutto]

@wwoytenko I encountered a problem where environment variables are not being loaded if their config isn't defined in a config file. It seems that this is a known Viper issue: spf13/viper#584

I'll come up with a PR to solve this issue, changing the common.tmp_dir default definition.

We can also utilize this PR to set a default behavior for storage config. I see these possible scenarios

• Define default storage to directory storage on ~/dumps -- this would allow Docker users to map a volume to /home/greenmask/dumps and use it without needing to specify the storage.directory.path configuration
• Require storage configuration and error out in case none is provided

[wwoytenko]

@joao-zanutto thank you for your research. You're right. We can overcome this issue by manually registering ENV vars. It is completed in #51 . Rebase from the actual main branch and adapt your PR #49 last changes.

About the next points

Define default storage to directory storage on ~/dumps -- this would allow Docker users to map a volume to /home/greenmask/dumps and use it without needing to specify the storage.directory.path configuration

Agree. You can add it into #49

Require storage configuration and error out in case none is provided

Looks like I have done it in #51

[joao-zanutto]

@wwoytenko following my previous implementation and your own, we would need to statically bind each env to have it working, however, I was thinking in making the bind dynamically using the Config struct.

I tried a few solution, (like the one described in this comment spf13/viper#584 (comment)) but ended up finding a new build flag we can add to make it work out of the box.

There is a new private function bindStruct() that is being called inside the Unmarshall() function, but since it has some retrocompability issues, it was put behind this -tags=viper_bind_struct build arg. (See spf13/viper#1429)

I already built and test locally and it is working, however it doesn't work for the S3 storage parameters because it's defined as a pointer and the function can't parse it properly (See spf13/viper#1429 (comment)). Is there a chance we can remove the pointer definition se we don't have to statically define each S3 parameter as its own env and bind them individually?

[wwoytenko]

I think you can easily get rid of the pointers. It should work as expected, verify that all is okay and NewConfig function calls correctly

[wwoytenko]

Leave a comment in the code and Makefile as well so we would determine why it uses build flags

[joao-zanutto]

I think you can easily get rid of the pointers. It should work as expected, verify that all is okay and NewConfig function calls correctly

I did as you instructed but may have missed something because I cannot make it work correctly and I'm having difficult troubleshooting it. May I have your assistance in #55 ?