Poll from memory before fetching task information from DB

[jtuglu1]

Description

When running high #s of tasks in a cluster, the Overlord can become bottlenecked by the endpoint /task/{taskid}/status.
Because taskQueryTool.getTaskInfo(taskid) currently only issues a direct I/O to the underlying task storage, this results in heavy I/O + serde from the metadata store to pull the associated task information.

This change stores/updates the TaskInfo information inside the activeTasks queue's TaskEntry type, allowing the requests to fetch from memory if the task entry exists, otherwise falling back to the DB.

Some lingering questions:

Code Style/Smell

I didn't really like the idea of shoving the TaskInfo into the TaskEntry, but it felt like the most convenient way of storing all the necessary information needed. We're really just relying on the TaskInfo for the creation time as well as some up-to-date value of the current TaskStatus. It might be worth standardizing on TaskInfo as Task and TaskStatus are often paired with each other.

Consistency

TaskStatus

Despite updating the task status on TaskQueue::notifyStatus callbacks, I wonder whether it may be possible for the respective states in task storage and memory to diverge for a given task's status.

• In normal state, my initial thought is that the value returned from TaskQueryTool::getActiveTaskInfo and TaskQueryTool::getTaskInfo should always return state that is at least as recent as the latest commit to the DB, since the status updates to the DB in TaskQueue::notifyStatus are in the same critical section as those to the TaskEntry value. There might be merit in updating the task status in the critical section once it's been committed to the DB, however.
• In a failure scenario, the overlord could receive requests prior to initially (or recently) syncing with DB, resulting in partial/incorrect results.

TaskQueryTool::getAllActiveTasks

Prior to this change, the function would use a dummy timestamp for both creation/queue insertion timestamps in the returned payload. Now that we can serve accurate timestamps, I believe we can still use the same timestamp for both fields.

Release note

Poll from memory before fetching task information from DB

---

Key changed/added classes in this PR

• MyFoo
• OurBar
• TheirBaz

---

This PR has:

• been self-reviewed.
  • using the concurrency checklist (Remove this item if the PR doesn't have any relation to concurrency.)
• added documentation for new or modified features or behaviors.
• a release note entry in the PR description.
• added Javadocs for most classes and all non-trivial methods. Linked related entities via Javadoc links.
• added or updated version, license, or notice information in licenses.yaml
• added comments explaining the "why" and the intent of the code wherever would not be obvious for an unfamiliar reader.
• added unit tests or modified existing tests to cover new code paths, ensuring the threshold for code coverage is met.
• added integration tests.
• been tested in a test Druid cluster.

[kfaraz]

Thanks, @jtuglu1 ! Patch looks good.
I have left some minor suggestions in a couple of places.

[maytasm]

LGTM. Please wait for @kfaraz and @samarthjain reviews too.

[kfaraz]

LGTM 👍🏻