// rebuildCellSrvShard computes and writes the serving graph data to a
// single cell
func rebuildCellSrvShard(ctx context.Context, log logutil.Logger, ts topo.Server, si *topo.ShardInfo, cell string) (err error) {
log.Infof("rebuildCellSrvShard %v/%v in cell %v", si.Keyspace(), si.ShardName(), cell)
for {
select {
case <-ctx.Done():
return ctx.Err()
default:
}
// Read existing EndPoints node versions, so we know if any
// changes sneak in after we read the tablets.
versions, err := getEndPointsVersions(ctx, ts, cell, si.Keyspace(), si.ShardName())
// Get all tablets in this cell/shard.
tablets, err := ts.GetTabletMapForShardByCell(ctx, si.Keyspace(), si.ShardName(), []string{cell})
if err != nil {
if err != topo.ErrPartialResult {
return err
}
log.Warningf("Got ErrPartialResult from topo.GetTabletMapForShardByCell(%v), some tablets may not be added properly to serving graph", cell)
}
// Build up the serving graph from scratch.
serving := make(map[pb.TabletType]*pb.EndPoints)
for _, tablet := range tablets {
// Only add serving types.
if !tablet.IsInServingGraph() {
continue
}
// Check the Keyspace and Shard for the tablet are right.
if tablet.Keyspace != si.Keyspace() || tablet.Shard != si.ShardName() {
return fmt.Errorf("CRITICAL: tablet %v is in replication graph for shard %v/%v but belongs to shard %v:%v", tablet.Alias, si.Keyspace(), si.ShardName(), tablet.Keyspace, tablet.Shard)
}
// Add the tablet to the list.
endpoints, ok := serving[tablet.Type]
if !ok {
endpoints = topo.NewEndPoints()
serving[tablet.Type] = endpoints
}
entry, err := topo.TabletEndPoint(tablet.Tablet)
if err != nil {
log.Warningf("EndPointForTablet failed for tablet %v: %v", tablet.Alias, err)
continue
}
endpoints.Entries = append(endpoints.Entries, entry)
}
wg := sync.WaitGroup{}
fatalErrs := concurrency.AllErrorRecorder{}
retryErrs := concurrency.AllErrorRecorder{}
// Write nodes that should exist.
for tabletType, endpoints := range serving {
wg.Add(1)
go func(tabletType pb.TabletType, endpoints *pb.EndPoints) {
defer wg.Done()
log.Infof("saving serving graph for cell %v shard %v/%v tabletType %v", cell, si.Keyspace(), si.ShardName(), tabletType)
version, ok := versions[tabletType]
if !ok {
// This type didn't exist when we first checked.
// Try to create, but only if it still doesn't exist.
if err := ts.CreateEndPoints(ctx, cell, si.Keyspace(), si.ShardName(), tabletType, endpoints); err != nil {
log.Warningf("CreateEndPoints(%v, %v, %v) failed during rebuild: %v", cell, si, tabletType, err)
switch err {
case topo.ErrNodeExists:
retryErrs.RecordError(err)
default:
fatalErrs.RecordError(err)
}
}
return
}
// Update only if the version matches.
if err := ts.UpdateEndPoints(ctx, cell, si.Keyspace(), si.ShardName(), tabletType, endpoints, version); err != nil {
log.Warningf("UpdateEndPoints(%v, %v, %v) failed during rebuild: %v", cell, si, tabletType, err)
switch err {
case topo.ErrBadVersion, topo.ErrNoNode:
retryErrs.RecordError(err)
default:
fatalErrs.RecordError(err)
}
}
}(tabletType, endpoints)
}
// Delete nodes that shouldn't exist.
for tabletType, version := range versions {
if _, ok := serving[tabletType]; !ok {
wg.Add(1)
go func(tabletType pb.TabletType, version int64) {
defer wg.Done()
log.Infof("removing stale db type from serving graph: %v", tabletType)
if err := ts.DeleteEndPoints(ctx, cell, si.Keyspace(), si.ShardName(), tabletType, version); err != nil && err != topo.ErrNoNode {
log.Warningf("DeleteEndPoints(%v, %v, %v) failed during rebuild: %v", cell, si, tabletType, err)
switch err {
case topo.ErrNoNode:
// Someone else deleted it, which is fine.
case topo.ErrBadVersion:
retryErrs.RecordError(err)
default:
fatalErrs.RecordError(err)
}
}
}(tabletType, version)
}
}
// Update srvShard object
wg.Add(1)
go func() {
defer wg.Done()
log.Infof("updating shard serving graph in cell %v for %v/%v", cell, si.Keyspace(), si.ShardName())
if err := UpdateSrvShard(ctx, ts, cell, si); err != nil {
fatalErrs.RecordError(err)
log.Warningf("writing serving data in cell %v for %v/%v failed: %v", cell, si.Keyspace(), si.ShardName(), err)
}
}()
wg.Wait()
// If there are any fatal errors, give up.
if fatalErrs.HasErrors() {
return fatalErrs.Error()
}
// If there are any retry errors, try again.
if retryErrs.HasErrors() {
continue
}
// Otherwise, success!
return nil
}
}