[Fix] Sync Tablet::_cumulative_point with TabletMeta::_cumulative_layer_point

[LemonCL]

Proposed changes

This PR fixes a synchronization issue between Tablet::_cumulative_point (in-memory) and TabletMeta::_cumulative_layer_point (persistent in RocksDB). Although there is an auto-recovery mechanism (calculate_cumulative_point()), it creates a problem window period after every BE restart, causing compaction system to be completely disabled until the first compaction is triggered.

Problem description

Real Production Evidence

Production environment shows this exact issue (tablet_id=4676420):

Runtime value (correct):

$ curl http://10.92.104.1:8040/api/compaction/show?tablet_id=4676420
{
    "cumulative point": 57,  # ← Runtime value is correct
    "last cumulative success time": "2026-02-27 07:11:20.758",
    "rowsets": [
        "[0-56] 1 DATA NONOVERLAPPING ... 56.33 KB"
    ]
}

Persistent value (wrong):

$ curl http://10.92.104.1:8040/api/meta/header/4676420
{
    "tablet_id": 4676420,
    "cumulative_layer_point": -1,  # ← Persistent value is -1 (wrong!)
    "tablet_state": "PB_RUNNING"
}

Root Cause

There are two separate variables tracking the cumulative point:

• Tablet::_cumulative_point - Runtime value, updated by compaction
• TabletMeta::_cumulative_layer_point - Persistent value, stored in RocksDB

The core issue: save_meta() never syncs _cumulative_point to _cumulative_layer_point, causing persistent value to always remain at -1.

Data flow:

1. Compaction updates _cumulative_point = 57 ✅
2. save_meta() is called
3. ❌ Missing: _tablet_meta->set_cumulative_layer_point(_cumulative_point)
4. TabletMeta saves to RocksDB with cumulative_layer_point = -1 ❌
5. BE restart loads cumulative_layer_point = -1 from RocksDB
6. Tablet constructor initializes _cumulative_point = -1 ❌

What changes were proposed in this pull request?

This PR adds bidirectional synchronization to eliminate the problem window:

1. Save Path - Sync to TabletMeta before persistence

File: be/src/olap/tablet.cpp:341

void Tablet::save_meta() {
    check_table_size_correctness();
    // NEW: Sync in-memory value to TabletMeta before persisting
    _tablet_meta->set_cumulative_layer_point(_cumulative_point);
    auto res = _tablet_meta->save_meta(_data_dir);
    // ...
}

Effect: Ensures cumulative_layer_point is always synchronized to RocksDB.

2. Load Path - Load from TabletMeta on construction

File: be/src/olap/tablet.cpp:260

// Before: hardcoded to -1
_cumulative_point(K_INVALID_CUMULATIVE_POINT),

// After: load from TabletMeta (which was deserialized from RocksDB)
_cumulative_point(_tablet_meta->cumulative_layer_point()),

Effect: Correctly restores cumulative_point from RocksDB on BE restart.

Before vs After

Before This Fix

Compaction completes:
  → _cumulative_point = 57 (runtime)
  → cumulative_layer_point = -1 (persistent) ❌

BE restart:
  → Load from RocksDB: cumulative_layer_point = -1
  → _cumulative_point = -1
  → Problem window: compaction disabled for minutes/hours ⚠️
  → Auto-recovery: calculate_cumulative_point() recalculates
  → _cumulative_point = 57 (runtime only, not persisted) ⚠️

Next BE restart:
  → Same problem repeats (cyclic issue) ❌

After This Fix

Compaction completes:
  → _cumulative_point = 57 (runtime)
  → save_meta() syncs to TabletMeta
  → cumulative_layer_point = 57 (persistent) ✅

BE restart:
  → Load from RocksDB: cumulative_layer_point = 57
  → _cumulative_point = 57 ✅
  → No problem window: compaction immediately available ✅
  → No need for auto-recovery ✅

Next BE restart:
  → Works correctly (problem eliminated) ✅

Verification

After this fix, for tablet_id=4676420:

• Runtime value: cumulative_point = 57
• Persistent value: cumulative_layer_point = 57 ✅ (synced!)
• After BE restart: cumulative_point = 57 ✅ (immediately restored)
• No problem window period ✅

Benefits

1. ✅ Eliminate problem window: Compaction works immediately after restart
2. ✅ Eliminate cyclic issue: Fix persists across all restarts
3. ✅ Improve query performance: No performance degradation window
4. ✅ Reduce operational cost: No need to monitor or manually trigger compaction
5. ✅ Support clone/restore: Correctly transfer cumulative point across replicas

Types of changes

• Bugfix (non-breaking change which fixes an issue)

Checklist

• I have read the CLA Document and made sure all commits are signed-off.
• I have added necessary documentation (if appropriate)
• Any dependent changes have been merged

[hello-stephen]

Thank you for your contribution to Apache Doris.
Don't know what should be done next? See How to process your PR.

Please clearly describe your PR:

1. What problem was fixed (it's best to include specific error reporting information). How it was fixed.
2. Which behaviors were modified. What was the previous behavior, what is it now, why was it modified, and what possible impacts might there be.
3. What features were added. Why was this function added?
4. Which code was refactored and why was this part of the code refactored?
5. Which functions were optimized and what is the difference before and after the optimization?

[lide-reed]

LGTM

[github-actions]

PR approved by at least one committer and no changes requested.

[github-actions]

PR approved by anyone and no changes requested.