88419: ui: add connected components for insights r=ericharmeling a=ericharmeling

This commit adds connected components for the workload and schema insights pages, for use in the CC Console.

Fixes #87693.

https://www.loom.com/share/08163a7c125948119ca71ac097099a29

Release note: None

88510: ui: move `latestQuery`, `latestFormattedQuery` from redux to local state r=xinhaoz a=xinhaoz

Previously, the fields `latestQuery` and `latestFormattedQuery` representing the latest non-empty query string for a statement viewed from the detaisl page was being stored in redux. The purpose of these fields was to preserve the query when changing tabs in the stmt details page. Saving this in the redux store was unnecessary and so this commit moves these fields to the stmt details local state.

Release note: None


https://www.loom.com/share/a28d412fb83a429391210935982404de

88594: cloud: add version gate for auth via assume role in AWS and GCP stora… r=rhu713 a=rhu713

…ge and KMS

Add a version gate for auth via assume role in AWS and GCP storage and KMS to prevent this type of auth until all nodes in the cluster has been upgraded to 22.2. The gate prevents a class of job failures where sometimes a job can succeed with assume role auth if its processors happen to all be on 22.2 nodes, but fail at times when one of its processor nodes don't support assume role. This version gate preempts the issue by preventing this type of auth until the cluster has been finalized on 22.2 and gives a better error message of why the auth cannot be used.

It's important to note that this gate does not prevent a user from creating a BACKUP job that uses assume role auth, e.g. via the DETACHED option, because the destination storage is not accessed during planning. This is inline with existing behavior for other types of auth errors, e.g. if the user enters incorrect credentials. The BACKUP job will still fail with the version gate error when it eventually executes.

Release note: None

88596: kv: use RaftAppliedIndexTerm to generate SnapshotMetadata, don't scan log r=nvanbenschoten a=nvanbenschoten

This commit replaces the call to `Term(raftAppliedIndex)` with direct use of the new `RaftAppliedIndexTerm` field (added in c3bc064) when generating a `SnapshotMetadata` in service of the `raft.Storage.Snapshot` interface. As of v22.2, this field has been fully migrated in.

First and foremost, this is a code simplification. However, it also helps with projects like #87050, where async Raft log writes make it possible for a Raft leader to apply an entry before it has been appended to the leader's own log. Such flexibility[^1] would help smooth out tail latency in any single replica's local log writes, even if that replica is the leader itself. This is an important characteristic of quorum systems that we fail to provide because of the tight coupling between the Raft leader's own log writes and the Raft leader's acknowledgment of committed proposals.

Release justification: None. Don't backport to release-22.2.

Release note: None.

[^1]: if safe, I haven't convinced myself that it is in all cases. It certainly is not for operations like non-loosely coupled log truncation.

88800: kvserver: use type-safe atomics in raftSendQueue r=nvanbenschoten a=pavelkalinnikov

Go 1.19 introduced atomic types that enforce atomic access to variables, which in many situation is less error-prone. This commit resolves a TODO to take advantage of these types.

Release note: None

Co-authored-by: Eric Harmeling <[email protected]>
Co-authored-by: Xin Hao Zhang <[email protected]>
Co-authored-by: Rui Hu <[email protected]>
Co-authored-by: Nathan VanBenschoten <[email protected]>
Co-authored-by: Pavel Kalinnikov <[email protected]>