CXX-2700 Add search index management e2e testing against Atlas #1002
Conversation
src/mongocxx/search_index_view.hpp
Outdated
| /// | ||
| /// This is a convenience method for creating a single search index. | ||
| /// | ||
| /// @param definition | ||
| /// The document describing the search index to be created. | ||
| /// | ||
| /// @return The name of the created search index. | ||
| /// | ||
| std::string create_one(bsoncxx::document::view_or_value definition); | ||
|
|
||
| /// | ||
| /// This is a convenience method for creating a single search index. | ||
| /// | ||
| /// @param definition | ||
| /// The document describing the search index to be created. | ||
| /// | ||
| /// @return The name of the created search index. | ||
| /// | ||
| std::string create_one(const client_session& session, | ||
| bsoncxx::document::view_or_value definition); | ||
|
|
There was a problem hiding this comment.
These two overloads were added due a change made in DRIVERS-2630 that allows for create_one to be called while only passing through a definition, no name. I kept the other overloads of create_one because the drivers ticket specified that a name should still be able to be passed through, but in the options of create_one. Because CXX allows function overloads, we don't have a need to have an "options" parameter.
| @@ -65,23 +74,24 @@ std::string search_index_view::create_one(const client_session& session, | |||
| return _get_impl().create_one(&session, model); | |||
| } | |||
|
|
|||
| std::vector<bsoncxx::string::view_or_value> search_index_view::create_many( | |||
| std::vector<std::string> search_index_view::create_many( | |||
There was a problem hiding this comment.
I ran into some issues with reading the values of the std::vector<bsoncxx::string::view_or_value> returned from this function due to ownership semantics. After talking to @eramongodb, we came to the conclusion that it would be easier and more clear to just make this a std::vector<std::string>. This minor change seen across this PR.
Summary
This PR will add 5 prose tests that will test search index management against a live Atlas cluster, modify the existing Evergreen config to automatically run these tests in the CI pipeline, and make small changes to the existing spec tests to sync with the drivers ticket.
Background
This ticket (mainly the changes made in the README) gives the vast majority of context needed to understand what prose tests should be added, why they should be added, and how to run them in Evergreen.
How to Test Locally
Being that running the prose tests could take around 20 minutes in total, there was not a lot of emphasis placed on making them easy to run locally. However, if you do wish to run them locally, here are the steps that you'll have to take:
setup-atlas-cluster.shscript indrivers-evergreen-tools/.evergreen/atlas. You'll need toexportsome of the environment variables mentioned in the comments of the script, they can be found on the Evergreen variables page. Spawning the cluster can take 7-10 minutes.MONGODB_URIenvironment variable as the cluster URI (printed toatlas-expansion.ymlby the script)atlas search indexes prose tests. This will take 5-10 minutes.teardown-atlas-cluster.sh. You'll need some environment variables for this one too, it's all mentioned in the comments of the script.Evergreen
We chose to use a task group to run the search index prose tests. What's unique about task groups is that they have a
setup_groupand ateardown_group, which are commands that will run before the tasks attached to the group are run (setup) and after they are run (teardown). Both the setup commands and the teardown commands will run regardless of the outcome of the actual tests. This is important in this use case because we need to ensure that the Atlas cluster gets torn down even if the tests don't pass. Failure to teardown the Atlas cluster is inefficient and will cause problems the next time that we try to spawn an Atlas cluster.What's New
create_onethat only takes in a definition (see comment for explanation)test/search_index_view.cpp