feat(pageserver): repartition on L0-L1 boundary

[skyzh]

Problem

Reduce the read amplification when doing repartition.

Summary of changes

Compute the L0-L1 boundary LSN and do repartition here.

[github-actions]

7425 tests run: 7031 passed, 42 failed, 352 skipped (full report)

---

Failures on Postgres 17

• test_sharding_compaction[4-16-False]: debug-x86-64-without-lfc
• test_sharding_compaction[4-32768-True]: debug-x86-64-without-lfc
• test_issue_5878[default_generation]: release-x86-64-without-lfc, release-x86-64-with-lfc, debug-x86-64-without-lfc, release-arm64-without-lfc, release-arm64-with-lfc
• test_issue_5878[same_generation]: release-x86-64-without-lfc, release-x86-64-with-lfc, debug-x86-64-without-lfc, release-arm64-without-lfc, release-arm64-with-lfc
• test_scrubber_physical_gc_ancestors[None]: release-x86-64-without-lfc, release-x86-64-with-lfc, release-arm64-with-lfc, release-arm64-without-lfc, debug-x86-64-without-lfc
• test_scrubber_physical_gc_ancestors[2]: release-arm64-without-lfc, release-arm64-with-lfc, debug-x86-64-without-lfc

Failures on Postgres 16

• test_issue_5878[default_generation]: release-x86-64-with-lfc, release-arm64-with-lfc
• test_issue_5878[same_generation]: release-x86-64-with-lfc, release-arm64-with-lfc
• test_scrubber_physical_gc_ancestors[2]: release-arm64-with-lfc

Failures on Postgres 15

• test_sharding_compaction[4-32768-False]: release-x86-64-with-lfc
• test_sharding_compaction[4-32768-True]: release-x86-64-with-lfc
• test_issue_5878[default_generation]: release-x86-64-with-lfc, release-arm64-with-lfc
• test_issue_5878[same_generation]: release-x86-64-with-lfc, release-arm64-with-lfc
• test_scrubber_physical_gc_ancestors[None]: release-x86-64-with-lfc, release-arm64-with-lfc
• test_scrubber_physical_gc_ancestors[2]: release-arm64-with-lfc

Failures on Postgres 14

• test_issue_5878[default_generation]: release-x86-64-with-lfc, release-arm64-with-lfc
• test_issue_5878[same_generation]: release-x86-64-with-lfc, release-arm64-with-lfc
• test_scrubber_physical_gc_ancestors[2]: release-x86-64-with-lfc, release-arm64-with-lfc
• test_scrubber_physical_gc_ancestors[None]: release-x86-64-with-lfc, release-arm64-with-lfc

# Run all failed tests locally:
scripts/pytest -vv -n $(nproc) -k "test_issue_5878[release-pg14-default_generation] or test_issue_5878[release-pg14-default_generation] or test_issue_5878[release-pg14-same_generation] or test_issue_5878[release-pg14-same_generation] or test_scrubber_physical_gc_ancestors[release-pg14-2] or test_scrubber_physical_gc_ancestors[release-pg14-2] or test_scrubber_physical_gc_ancestors[release-pg14-None] or test_scrubber_physical_gc_ancestors[release-pg14-None] or test_sharding_compaction[release-pg15-4-32768-False] or test_sharding_compaction[release-pg15-4-32768-True] or test_issue_5878[release-pg15-default_generation] or test_issue_5878[release-pg15-default_generation] or test_issue_5878[release-pg15-same_generation] or test_issue_5878[release-pg15-same_generation] or test_scrubber_physical_gc_ancestors[release-pg15-None] or test_scrubber_physical_gc_ancestors[release-pg15-None] or test_scrubber_physical_gc_ancestors[release-pg15-2] or test_issue_5878[release-pg16-default_generation] or test_issue_5878[release-pg16-default_generation] or test_issue_5878[release-pg16-same_generation] or test_issue_5878[release-pg16-same_generation] or test_scrubber_physical_gc_ancestors[release-pg16-2] or test_sharding_compaction[debug-pg17-4-16-False] or test_sharding_compaction[debug-pg17-4-32768-True] or test_issue_5878[release-pg17-default_generation] or test_issue_5878[release-pg17-default_generation] or test_issue_5878[debug-pg17-default_generation] or test_issue_5878[release-pg17-default_generation] or test_issue_5878[release-pg17-default_generation] or test_issue_5878[release-pg17-same_generation] or test_issue_5878[release-pg17-same_generation] or test_issue_5878[debug-pg17-same_generation] or test_issue_5878[release-pg17-same_generation] or test_issue_5878[release-pg17-same_generation] or test_scrubber_physical_gc_ancestors[release-pg17-None] or test_scrubber_physical_gc_ancestors[release-pg17-None] or test_scrubber_physical_gc_ancestors[release-pg17-None] or test_scrubber_physical_gc_ancestors[release-pg17-None] or test_scrubber_physical_gc_ancestors[debug-pg17-None] or test_scrubber_physical_gc_ancestors[release-pg17-2] or test_scrubber_physical_gc_ancestors[release-pg17-2] or test_scrubber_physical_gc_ancestors[debug-pg17-2]"

Flaky tests (4)

Postgres 17

• test_scrubber_physical_gc_ancestors[2]: release-x86-64-without-lfc, release-x86-64-with-lfc

Postgres 15

• test_scrubber_physical_gc_ancestors[2]: release-x86-64-with-lfc

Postgres 14

• test_check_visibility_map: release-arm64-with-lfc

Test coverage report is not available

_{The comment gets automatically updated with the latest test results
36985cd at 2025-02-06T22:52:45.774Z :recycle:}

[problame]

This will conflict at the code level, but not semantics level, with #10547

---

I have been arguing for a long time to compact at disk_consistent_lsn() and have alway met resistance from @hlinnaka , although I never understood the argument.

This here is even more extreme, because l0_min_lsn is generally < disk_consistent_lsn.

I'm all for it, maybe hear Heikki's concerns one last time?

[skyzh]

cc @hlinnaka if you have any concerns -- I'd like to merge it now so that we can observe it in staging+preprod for a weekend, and I can resolve/revert it if you don't think this is a good idea.