Merge cockroachdb#26753 cockroachdb#26754

26753: storage: Add extra event to allocator rebalancing r=a-robinson a=a-robinson Helps make the output of simulated allocator runs less confusing, since otherwise it's not clear why we're considering removal from the range and why the replicas being considered for removal includes one that isn't even a real member of the range. Release note: None Would have made looking at the simulated allocator output from https://forum.cockroachlabs.com/t/how-to-enable-leaseholder-load-balancing/1732/3 a little more pleasant. 26754: Bump RocksDB pointer to grab facebook/rocksdb#3926 r=benesch,a-robinson a=petermattis RocksDB was violating an invariant that no 2 sstables in a level overlap. It isn't quite clear what the upshot of this violation is. At the very least it would cause the overlapping tables to be compacted together. It seems possible that it could lead to missing writes, but I haven't been able to verify that. Fixes cockroachdb#26693 Release note: None Co-authored-by: Alex Robinson <[email protected]> Co-authored-by: Peter Mattis <[email protected]>
yuzefovich · Jun 15, 2018 · a6e7b4a · a6e7b4a
3 parents 718f6bd + 903a068 + dd35942
commit a6e7b4a
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 1 deletion.
diff --git a/c-deps/rocksdb b/c-deps/rocksdb
diff --git a/pkg/storage/allocator.go b/pkg/storage/allocator.go
@@ -386,6 +386,7 @@ func (a Allocator) simulateRemoveTarget(
 	defer func() {
 		a.storePool.updateLocalStoreAfterRebalance(targetStore, rangeInfo, roachpb.REMOVE_REPLICA)
 	}()
+	log.VEventf(ctx, 3, "simulating which replica would be removed after adding s%d", targetStore)
 	return a.RemoveTarget(ctx, zone, candidates, rangeInfo, disableStatsBasedRebalancing)
 }
+5 −0		db/compaction_picker.cc
+62 −0		db/db_compaction_test.cc