cancel
Showing results forย 
Search instead forย 
Did you mean:ย 
Data Engineering
Join discussions on data engineering best practices, architectures, and optimization strategies within the Databricks Community. Exchange insights and solutions with fellow data engineers.
cancel
Showing results forย 
Search instead forย 
Did you mean:ย 

Serverless Compute connectivity issues with .com.br domains vs. Classic Clusters Spark hangs

ThiagoRosetti
Visitor

Hi everyone,

I'm facing two specific issues in my Databricks Premium workspace (AWS - sa-east-1).

  1. Serverless Connectivity Issue: When using Serverless compute, I can successfully call APIs ending in .com, but calls to .com.br domains fail with connection/DNS errors. The exact same code works fine when running on a Classic Cluster.

  • VPC Setup: Custom VPC with Unity Catalog enabled.

  • Security Groups: Outbound rules are open for port 443 (0.0.0.0/0).

  • Symptom: It feels like a DNS resolution or Egress filtering issue specific to Serverless.

  1. Classic Cluster Spark Hang: On the other hand, when I switch to a Classic Cluster to bypass the connectivity issue, any Spark command (e.g., spark.read or simple transformations) hangs indefinitely without starting the job.

Has anyone experienced this specific behavior where Serverless ignores certain TLDs or where Spark fails to initialize on Classic Clusters in the same VPC?

Thanks in advance!

(pt-br)

Olรก pessoal,

Estou enfrentando dois problemas distintos no meu workspace Premium (AWS - regiรฃo sa-east-1):

  1. Conectividade no Serverless: Nรฃo consigo consumir APIs que terminam em .com.br usando Serverless compute. Se a API for .com, funciona normalmente. O mesmo cรณdigo funciona em um Cluster Clรกssico, o que sugere que o Serverless estรก lidando com o DNS ou com a saรญda de rede de forma diferente.

  • Jรก verifiquei os Security Groups e a porta 443 estรก aberta para 0.0.0.0/0.

  1. Spark "carregando infinitamente" no Cluster: Para contornar o problema acima, tentei usar um Cluster comum. O cรณdigo de requisiรงรฃo API funciona, mas qualquer comando Spark (como ler um dataframe ou um simples count) fica processando infinitamente e nรฃo inicia o job.

Alguรฉm jรก passou por algo parecido ou sabe se existe alguma configuraรงรฃo de VPC/Unity Catalog que possa estar causando esse conflito entre o tipo de computaรงรฃo e a resoluรงรฃo de domรญnios?

Obrigado!

0 REPLIES 0