Skip to content

storage: AdminSplit using descriptor with ReplicaID zero #31918

Closed
Closed
storage: AdminSplit using descriptor with ReplicaID zero#31918
Assignees
petermattis
Labels
C-bugCode not up to spec/doc, specs & docs deemed correct. Solution expected to change code/behavior.
Milestone
19.1
@tbg

Description

@tbg
tbg
opened on Oct 26, 2018

Got this failure:

--- FAIL: import/tpch/nodes=8 (9943.52s)
	test.go:639,cluster.go:1453,import.go:122: pq: split at key /Table/53/1/390157478/5 failed: replica 0 is invalid: ReplicaID must not be zero

This originates in

cockroach/pkg/roachpb/metadata.go

Lines 222 to 234 in a050d56

// Validate performs some basic validation of the contents of a replica descriptor.
func (r ReplicaDescriptor) Validate() error {
if r.NodeID == 0 {
return errors.Errorf("NodeID must not be zero")
}
if r.StoreID == 0 {
return errors.Errorf("StoreID must not be zero")
}
if r.ReplicaID == 0 {
return errors.Errorf("ReplicaID must not be zero")
}
return nil
}

So we have a replica that has a NodeID and StoreID but a zero replicaID either in

cockroach/pkg/storage/replica_command.go

Line 348 in 62ea444

if err := updateRangeDescriptor(b, rightDescKey, nil, rightDesc); err != nil {

or in the other call to updateRangeDescriptor. I'm taking a look at the logs now. Doubtful that it's at all related to this change.

PS merges are turned off in this testing.

My best bet for this error is that

cockroach/pkg/storage/replica_command.go

Line 173 in 62ea444

reply, lastErr = r.adminSplitWithDescriptor(ctx, args, r.Desc())
returns a descriptor with a replica that has a replica with replicaID zero.

Metadata

Metadata

Assignees

Labels

C-bugCode not up to spec/doc, specs & docs deemed correct. Solution expected to change code/behavior.

Type

No type

Projects

No projects

Milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions