tests: fix br_tikv_outage learner bug

[Leavrth]

What problem does this PR solve?

fix issue #1050
In the case outage-after-request, there is a voter peer on tikv1 removed, and then a learner peer on tikv1 added.
But then quickly in the next case outage-at-finegrained (don't restart tikv while doing backup), tikv3 is killed.
So there is only 1 peer and 1 learner for region 257, lead to test got stuck.

What is changed and how it works?

It seems the case outage-after-request have a bad influence on the next case, so this PR try to exchange the sequence of these cases.

Check List

Related changes

• Need to cherry-pick to the release branch

Release note

-No release note

[Leavrth]

/run-integration-test

[Leavrth]

/run-integration-test

[Leavrth]

/run-integration-test

[Leavrth]

/run-integration-test

[Leavrth]

/run-integration-test

[Leavrth]

/run-integration-test

[Leavrth]

/run-integration-test

[Leavrth]

/run-integration-test

[Leavrth]

@YuJuncen PTAL

[Leavrth]

/run-integration-test

[glorv]

This sounds weird to me, I remember the scripts will recover the cluster to normal state before run each case, thus cases won't have side effect for each other. Are we sure this is the root cause? This case had already been fixed for several times 😭, is this the last time

[Leavrth]

This sounds weird to me, I remember the scripts will recover the cluster to normal state before run each case, thus cases won't have side effect for each other. Are we sure this is the root cause? This case had already been fixed for several times 😭, is this the last time

This test has many bugs which all can make test failed, And I think this is the last bug.
In fact, the recover in scripts only make sure that the cluster is already initialized before running each case. (This is fixed in other PR #1160 )
As shown above, even though this killed node has been restarted and initialized, there are still some peers in this node hasn't been synchronized with other peers in other nodes, which make these peers keep learner state.
At this point, if we kill another node, these learner state peers might be failed to synchronize itself, which makes the test timeout.

[ti-chi-bot]

[REVIEW NOTIFICATION]

This pull request has been approved by:

• glorv
• kennytm

To complete the pull request process, please ask the reviewers in the list to review by filling /cc @reviewer in the comment.
After your PR has acquired the required number of LGTMs, you can assign this pull request to the committer in the list by filling /assign @committer in the comment to help you merge this pull request.

The full list of commands accepted by this bot can be found here.

Reviewer can indicate their review by submitting an approval review.
Reviewer can cancel approval by submitting a request changes review.

[kennytm]

/merge

[ti-chi-bot]

This pull request has been accepted and is ready to merge.

Commit hash: 0ce3aa1

[ti-chi-bot]

In response to a cherrypick label: new pull request created: #1281.

[ti-chi-bot]

In response to a cherrypick label: new pull request created: #1282.

[ti-chi-bot]

In response to a cherrypick label: new pull request created: #1283.