Definindo links DISCOURSE_S3_CDN_URL para ativos na URL do CDN S3

I had this problem before and decided that I was crazy, confused, or the database on the site was suspect, but this is on a brand new site. Also, I was on Digital Ocean spaces, so I thought it might be a problem somehow.

I’m trying again to figure out how to keep images on AWS S3 like I think the Big Boys do.

Here’s what I have in the env section:

  DISCOURSE_S3_ACCESS_KEY_ID: 'key'
  DISCOURSE_S3_SECRET_ACCESS_KEY: 'lock'
  DISCOURSE_BACKUP_LOCATION: 's3'
  DISCOURSE_S3_BUCKET: 'lc-xyz'
  DISCOURSE_S3_BUCKET_NAME: 'lc-xyz'
  DISCOURSE_S3_BACKUP_BUCKET: 'lc-xyz/backups'
  DISCOURSE_S3_UPLOAD_BUCKET: 'lc-xyz/uploads'
  DISCOURSE_S3_CDN_URL: 'https://lc-rbx.s3.amazonaws.com'

When I include the s3 cdn url the site breaks because all of the links to the assets are on S3 like https://lc-xyz.s3.amazonaws.com/assets/plugin-third-party-01ba4719c80b6fe911b091a7c05124b64eeece964e09c058ef8f9805daca546b.br.js.

Because if you define s3 cdn url Discourse looks for the assets on the s3_cdn_url.

I did a rebuild, but the assets are still missing. I can do a

rake s3:upload_assets

Is there an after_bundling_assets stanza that I could add that to? (I know about after_db_migrate and after_bundle_exec, but don’t know if those would work.)

Do assets get produced some other time? (It would seem like when themes get modified assets would change.)

If I also had a “push CDN like normal”, would that keep this from happening?

Is there some best practice that I’m missing?

You can use after_assets_precompile as the hook for that rake task.

Aha! Thanks very much for that. Where does the list of those things live?

I’m still confused why this is necessary, especially since you seemed confused before. (Or maybe I should be happy that I can push all this stuff to S3, now that I know how to do it–And I think that it was you who said that one could use CloudFlare’s free CDN to front an AWS bucket.)

I’m trying this again. I’ve got uploads going to s3. That works.

I configured cloudflare to CDN the site. If I enter https://lc-XXX.literatehosting.com/uploads in the s3_cdn_url site setting, new uploads go to the S3 bucket and the image is linked to the CDN url and it works.

BUT if I try to set the s3_cdn_url with an ENV in app.yml (or edit config/discourse.conf by hand and sv restart unicorn, all of the assets are loaded from S3 (where they aren’t). That might be OK, but if I try to rake s3:upload_assets, it complains that S3 isn’t configured.

I happened across this while figuring out some clashing settings names.

I think (and hope, because it means I understand this and didn’t write absolute nonsense in the thread linked above) that if you remove these two lines:

DISCOURSE_S3_BUCKET: 'lc-xyz'
DISCOURSE_S3_BUCKET_NAME: 'lc-xyz'

Then you can set DISCOURSE_S3_CDN_URL, and it will only be used for uploads, not assets.

I think that you do understand, and, thanks to you, I’m at least closer to understanding myself! Thanks very much!

Running into this and still very confused after reading this thread…

I’m trying to server s3 uploads (not compiled assets) from CDN (Cloudfront).

If I configure “s3 cdn url” via the settings screen, it works as expected (well…except for System upload not using s3 cdn url)

However, if I configure via DISCOURSE_S3_CDN_URL and rebuild, the frontend is broken because its trying to load compiled assets from my s3 cdn url.

Seems DISCOURSE_S3_CDN_URL / s3_cdn_url should only affect uploads, and DISCOURSE_CDN_URL should only affect assets.

That is my experience too. I ended up making a plugin to set the S3_CDN_URL.

@pfaffman seems like this should be filed as a bug then, yes?

Perhaps. It’s not a feature likely to be used by normal self hosters,so it’s going to be a low priority. Also, I think that there is soon to be a change in how global settings and shadowed by global work, so it’ll likely get worked out then.

Parece que o mesmo problema me rejeitou aqui.
@pfaffman, preciso da sua ajuda.
Você usa o CloudFront para alcançar a URL de CDN de domínio personalizado ou apenas usa o prefixo público dos objetos do bucket na URL da CDN?

Acabei de me deparar com isso. @pfaffman, você conseguiu resolver isso?

Consegui fazer os uploads funcionarem configurando manualmente o s3_cdn_url nas configurações do site, mas idealmente preciso poder definir a variável de ambiente globalmente ao fazer o deploy. Quando faço isso, tenho o mesmo problema: o Discourse tenta buscar os assets no s3_cdn_url, o que parece ser um bug para mim em discourse/lib/content_security_policy/default.rb at main · discourse/discourse · GitHub

Acho que acabei de definir o valor no banco de dados.

Minha suposição é que a solução seja encontrar a tarefa rake que enviará os ativos para o S3.

Estava me perguntando se alguém da @team poderia confirmar se esse é o comportamento esperado, de que não é possível usar uma CDN S3 (configurada globalmente) para uploads sem que o Discourse também procure os ativos lá. Parece um comportamento inesperado, a menos que eu esteja entendendo algo errado.

Sim, esse é o comportamento.

Configurar o CDN do S3 fará com que ele seja usado para tudo o que seja um arquivo estático, sejam uploads ou ativos JS.

Aqui estão algumas informações sobre isso; eu lidei com isso no mês passado.

Corrigi o problema configurando ambas as variáveis (DISCOURSE_S3_CDN_URL e DISCOURSE_CDN_URL) e criando duas distribuições do CloudFront: uma para os uploads, com origem no bucket S3, e outra para os ativos, com origem no servidor.

Aqui está o código que usamos para isso:

Aqui está nosso arquivo app.yml (que chamamos de web.yml), onde substituímos as variáveis no momento da construção: infra/modules/services/discourse/web.yml at master · debtcollective/infra · GitHub

Incrível. Muito obrigado. Parece não haver muita documentação sobre isso, e as configurações sugerem que você pode usar o S3 apenas para uploads, o que, creio eu, é o que está confundindo a todos nós.

Está na minha lista escrever um howto sobre isso. Espero fazer isso até o final de semana.

Sim, assim que você entende que precisa de duas distribuições do CloudFront, faz mais sentido. Além disso, lembre-se de fazer upload dos ativos para o S3 após cada reconstrução. Há um problema nas atualizações a partir do Docker Manager, onde é necessário executar bundle exec rake s3:upload_assets manualmente dentro do contêiner Docker. Se você fizer uma reconstrução em vez disso, deve funcionar se adicionar estas linhas ao seu app.yml

hooks:
  after_assets_precompile:
    - exec:
        cd: $home
        cmd:
          - 'bundle exec rake s3:upload_assets'

Essa tarefa está reclamando que eu não configurei o S3. Eu reduzi o problema a este método em global_setting.rb:

def self.use_s3?
    (@use_s3 ||=
      begin
        s3_bucket &&
        s3_region && (
          s3_use_iam_profile || (s3_access_key_id && s3_secret_access_key)
        ) ? :true : :false
      end) == :true
  end

Onde GlobalSetting.s3_bucket deve ser definido? Parece-me que precisamos definir as variáveis de ambiente DISCOURSE_S3_UPLOAD_BUCKET e DISCOURSE_S3_BUCKET. Qual é a diferença entre as duas?