Databricks Community

Use Databricks runtime 8.3 ML or later.

Does this help? "No Public IPs": https://docs.microsoft.com/en-us/azure/databricks/security/secure-cluster-connectivity

Standard tiers are allowed to have 1000 saved jobs. Premium tiers have a higher limit at 1500. Some clouds have an enterprise tier which has a saved job limit of 2000. A workspace is limited to 1000 concurrent job runs. A 429 Too Many Requests respon...

By default, the state data (streaming aggregation query) is maintained in the JVM memory of the executors and large number of state objects could put memory pressure on the JVM causing high GC pauses. If you have stateful operations in your streamin...

DBFS is the "Databricks File System", but really it's just a shim / wrapper on top of distributed storage, that makes files in S3 or ADLS look like local files under the path /dbfs/... This can be really useful when working with libraries that do not...

Using the same instance type is a fine default. If you know that you need very large workers, but little happens on the driver, maybe you can save money with a smaller driver. Conversely, you may know that some parts of your notebook involve a lot of...

Yes, still limited to 1000 rows / data points. However, when your visualization involves things like sums or averages of a Spark DataFrame's result, those will be performed on the cluster, so would involve maybe many more than 1000 data points, even ...

Single-node, like the name implies, is a single machine. It still has Spark, just a local cluster. This is a good choice if you are running a workload that does not use Spark, or only needs it for data access. One good example is a small deep learnin...

You don't have to. If you don't have a huge data set, there may not be much value in Spark ML over anything else. There are also other distributed modeling libraries that work on Spark like xgboost, and Horovod + TF, Keras, Pytorch. Spark ML is a goo...

> What determines when the cluster autoscaling activates to add and remove workersDuring scale-down, the service removes a worker only if it is idle and does not contain any shuffle data. This allows aggressive resizing without killing tasks or recom...

Mounts are global to all clusters but as a best practice, you can use IAM roles to prevent access tot he underlying data. To take this one step further, you can use IAM credential passthrough rather than instance profile because instance profile can ...