DB: rename table/container/etc to collection

[lmolkova]

Changes

Renames db.sql.table, db.cassandra.table, db.mongodb.collection, db.cosmosdb.container to db.collection.name

TODO:

• Q: should collection name include db name (SELECT * from db.table) - Java uses db.table
  • A: if captured from the query, should match what's provided - don't attempt to parse qualified name
• Q: anonymous tables (select * from (select 'foo' as bar) ... ) - no table, _OTHER, first thing that looks like a table name?
  • A: not set. Not all queries are related to tables (create db, update user, etc), so it should be valid not to have any collection name.
• Q: when more then one - the first or the primary? What's the primary?
  • A: the first. there is no notion of primary, so the first, but we have enough room for system-specific semconv to override.

Merge requirement checklist

• CONTRIBUTING.md guidelines followed.
• Change log entry added, according to the guidelines in When to add a changelog entry.
  • If your PR does not need a change log, start the PR title with [chore]
• schema-next.yaml updated with changes to existing conventions.

[AlexanderWert]

What about db.redis.database_index? Wouldn't that be of a similar category? If that's the case how to deal with the fact that db.redis.database_index is an int instead of a String?

[gregkalapos]

Thoughts on the TODO part:

should collection name include db name (SELECT * from db.table) - Java uses db.table

I think this could be better addressed with a specific attribute. There is also schema, which could be part of it as well (like, if we include db name in collection.name, then why don't we include schema?). I personally would prefer if this would be only table name for SQL and then tech specific semantic conventions could define additional fields (like db name and schema). Other option could be that the spec says it's something like a fully qualified name - but then I'm afraid that's not 100% clear across all database systems - like is schema part of it? Is the db instance part of it?

anonymous tables (select * from (select 'foo' as bar) ... ) - no table, _OTHER, first thing that looks like a table name?

I'd vote for _OTHER (or some specific well known value representing the fact that there is no table in the query). Otherwise it's impossible to differentiate from the non-parsing case when it's not available. So if the instrumentation known that there is no table, I'd store this information.

[lmolkova]

What about db.redis.database_index? Wouldn't that be of a similar category? If that's the case how to deal with the fact that db.redis.database_index is an int instead of a String?

Redis is currently mapping db index to db.name (as a string). Redis doesn't have a notion similar to collection - could be a set/list name or a namespace for a key? But there is nothing reliable we can use (namespaces are just a convention).

[lmolkova]

Been thinking about

anonymous tables (select * from (select 'foo' as bar) ... ) - no table, _OTHER, first thing that looks like a table name?

given that there are operations that are not involving tables (create db, update user, etc), it should be valid to not have one.
And asking instrumentation parsers to detect anonymous tables specifically does not seem that useful.

Also, if parsing is enabled, users still get sanitized query.