ADBDEV-8057: Fix pg_rewind leaving replica on old timeline

[silent-observer]

Remove bogus assertion in pg_rewind

It's entirely possible that the calculated "point of divergence" is
before the end of WAL on the target, because when we calculate the
point of divergence, we assume that the target's WAL extends to
infinity on its current timeline. It might be more clear to calculate
the end of WAL first, and if the target is a direct ancestor of the
source, consider the end of target WAL as the point of
divergence. However, it doesn't change the result: the target is a
direct ancestor of the source, and doesn't need to be rewound.

Add tests to cover the "no rewind required" cases, both when the
target's end of WAL is exactly at the point of divergence, which is
what the removed assertion assumed (case B in the test), and the case
where the assertion failed (case A).

Changes from original patch: test was made more general to prepare for more
detailed tests in the next commits. In particular, we pg_rewind from a
running source node, not from a stopped one, and the source node is passed as
an argument. This doesn't affect the test validity for this patch.

Discussion: https://www.postgresql.org/message-id/18575-1b9b3b60e2a480de@postgresql.org

---

Alternatively, perhaps just copying the history file would be enough?

Changes from original patch:

• Move sourceHistfile initialization to before rewind_parseTimeLineHistory,
  since rewind_parseTimeLineHistory actually modifies it.
• Don't copy timeline history file when dry_run is specified. Dry run
  shouldn't make changes that can potentially break the target server when
  applied incorrectly.
• Fix 107_empty_conf.pl test to preserve primary_conninfo setting.

---

Fix pg_rewind edge case when checkpoint is at switchpoint

In one specific edge case, when the source server makes a checkpoint exactly
after switching to a new timeline, switchpoint LSN exactly matches latest
checkpoint LSN. This can happen, for example, if the source has just performed
archive recovery. Then setting minRecoveryPoint to the checkpoint LSN in
pg_rewind is illegal and causes an error at startup, since minRecoveryPoint
actually means the minimum allowed end of WAL, and we want that to be after
the switch occurs, not at the end of the previous timeline.

As a fix, add +1 to the checkpoint LSN when setting minRecoveryPoint, which
now correctly handles this case. Also add a test scenario for the source node
that has performed recovery.

---

Note: Do not squash to preserve authorship

[KnightMurloc]

Use checkpoint+1 to account for the edge case when the checkpoint exactly
matches switchpoint LSN.

Is this not the same case as described in the first commit?

Add the appropriate description in the second commit.

[KnightMurloc]

If switches to the first commit, the tests pass with unchanged to pg_rewind.

[silent-observer]

I split the patch into three commits now.

• The first contains the assert removal and an updated test that can be adapted to second and third commits with minimal changes.
• The second contains the bulk of the patch by Heikki, with minor fixes, and updates to the test that only pass after the second commit.
• The third commit contains a fix for minRecoveryPoint edge case, and extends the test more to cover it.
  Now the test changes are fully sequential: the tests pass on each of the three commits, but only with corresponding changes to pg_rewind present.

[RekGRpth]

Refactor sourceHistfile to be an out parameter instead of global state.

pgindent these your changes

[silent-observer]

Is it expected that changes in the src/bin/pg_rewind/t/107_empty_conf.pl test work without a patch?

Yes, changes to 107_empty_conf.pl only fix config line being accidentally deleted, they don't change the nature of the test. 010_target_is_ancestor.pl is the main test for this patch.