check error

actually, all this code effectively repeats NewSession, the only difference is the keyspace. Can you move it into a helper createSession(keyspace string)?

The current implementation of newSessionPrerequisites kind of uses a hack to override cluster.Keyspace which was set in c.NewCluster(), so that the connection can be established without keyspace.
So, you are suggesting enclosing this override logic in createSession(keyspace string)?

Or something like I create a copy of the Configuration with different keyspace to do the work, so that we don't set cluster.keyspace = "" after creating it?

You can make the new function to depend on the config argument, and pass whichever config is needed in each instance, including cloning it to override keyspace

why do we need custom functions, can the conditions not be expressed directly via tags?

if we must have custom logic then just write it directly, with proper error messages - right now you are just returning a Boolean, so at best validator can say "property x did not validate".

why do we need custom functions, can the conditions not be expressed directly via tags?

Couldn't find such validations here: https://github.com/asaskevich/govalidator

if we must have custom logic then just write it directly, with proper error messages - right now you are just returning a Boolean, so at best validator can say "property x did not validate".

Fixed it.

how is this going to distinguish the two runs with SKIP_APPLY_SCHEMA=true/false?

I didn't quite get the question.

1. the CI workflow is not providing / varying SKIP_APPLY_SCHEMA, so by default it's false and the schema is being pre-created as before, i.e. none of your new code is being CI-tested
2. even if the CI workflow was running a matrix with SKIP_APPLY_SCHEMA varied true/false, the actual test is run with the above command that doesn't tell the test to pass any additional arguments to the code, meaning that the CreateSchema field in the config will still be false and won't execute your schema creation logic

I suggest that we can add the create: true in the config-cassandra.yaml, or do we still want to run integration tests with create: false?

For the workflow we can add the -s flag in the command based on skip-apply-schema : [true, false] in the matrix strategy.

Does this sound good?

we want to have integration tests that do both: (1) script-created schema (as we do today) and (2) in-code created schema (your fix)

We can use similar logic here: https://github.com/jaegertracing/jaeger/blob/58e08c5fcda8d07ac8f7306e40efe01766d117bc/cmd/jaeger/internal/integration/e2e_integration.go#L123C1-L126C3

Let me update it accordingly.

make fmt

it would make sense to trim spaces before checking for len=0.

I would do it in generateSchemaIfNotPresent against casQueries, not here.

There are too many functions in this file that are polluting the overall package namespace. I would prefer to introduce a helper struct

type schemaCreator struct {
  session cassandra.Session
  cfg *Schema
}

and define those functions on that struct (and minimize parameter passing)

not sure about this one