nexus-db-queries: track blueprint table coverage cumulatively #8988
Conversation
|
Thanks for taking this one on. I just wanted to give you a heads-up that I probably won't be able to look for a week or two. (If anyone else at Oxide can take a look, go for it!) |
sarataha
force-pushed
the
blueprint-coverage-test
branch
from
bd8bde9 to
8bb5e6f
Compare
| // Now make a new blueprint (with no meaningful changes) to ensure we | ||
| // can delete the last test blueprint we generated above. |
There was a problem hiding this comment.
This comment doesn't seem right any more.
I think what you want to do is add a new lap before this one that includes the Clickhouse stuff. Then keep the existing lap that makes basically no changes.
| // Exception tables that may be empty in the test blueprints: | ||
| // - debug log for planner reports: only populated when the blueprint | ||
| // was produced by the planner (test blueprints generally aren't) | ||
| let exception_tables = ["debug_log_blueprint_planning"]; |
There was a problem hiding this comment.
Is this true @jgallagher? I forget where we came down on this.
There was a problem hiding this comment.
It is true; a blueprint is specified by
pub enum BlueprintSource {
/// The initial blueprint created by the rack setup service.
Rss,
/// A blueprint created by the planner, and we still have the associated
/// planning report.
Planner(Arc<PlanningReport>),
/// A blueprint created by the planner but loaded from the database, so we
/// no longer have the associated planning report.
PlannerLoadedFromDatabase,
/// This blueprint was created by one of `reconfigurator-cli`'s blueprint
/// editing subcommands.
ReconfiguratorCliEdit,
/// This blueprint was constructed by hand by an automated test.
Test,
}and when we go to insert the blueprint to the db, we only insert a row in debug_log_blueprint_planning if the source is BlueprintSource::Planner(some_report).
| // Check that all non-exception tables have at least one row | ||
| let empty_tables = self.empty_tables(); | ||
| let problematic_tables: Vec<_> = empty_tables | ||
| .into_iter() | ||
| .filter(|table| !exception_tables.contains(&table.as_str())) | ||
| .collect(); |
There was a problem hiding this comment.
I was wondering if it's worth keeping the previous list of exception tables, which would basically leave us with three categories:
debug_log_blueprint_planning: could be empty?- the previous exception tables: they should now all have one row (this is included in the check here)
- the previous non-exception tables: they should now all have
num_blueprintsrows
I think I'm coming down on what you've got here though. The real goal here is to make sure that we can successfully round-trip everything through the database, and that's what we've verified.
Co-authored-by: David Pacheco <[email protected]>
3 participants
This is a follow up PR to #8667 as suggested by @davepacheco in #8667 (comment)
Instead of verifying table coverage for individual blueprint insertions, this change tracks coverage cumulatively across all tests in
test_representative_blueprint().