// refreshTablet needs to be run after an action may have changed the current
// state of the tablet.
func (agent *ActionAgent) refreshTablet(ctx context.Context, reason string) error {
log.Infof("Executing post-action state refresh")
span := trace.NewSpanFromContext(ctx)
span.StartLocal("ActionAgent.refreshTablet")
span.Annotate("reason", reason)
defer span.Finish()
ctx = trace.NewContext(ctx, span)
// Save the old tablet so callbacks can have a better idea of
// the precise nature of the transition.
oldTablet := agent.Tablet().Tablet
// Actions should have side effects on the tablet, so reload the data.
if err := agent.readTablet(); err != nil {
log.Warningf("Failed rereading tablet after %v - services may be inconsistent: %v", reason, err)
return fmt.Errorf("Failed rereading tablet after %v: %v", reason, err)
}
if updatedTablet := agent.checkTabletMysqlPort(agent.Tablet()); updatedTablet != nil {
agent.mutex.Lock()
agent._tablet = updatedTablet
agent.mutex.Unlock()
}
if err := agent.updateState(ctx, oldTablet, reason); err != nil {
return err
}
log.Infof("Done with post-action state refresh")
return nil
}
// Update shard file with new master, replicas, etc.
//
// Re-read from TopologyServer to make sure we are using the side
// effects of all actions.
//
// This function locks individual SvrShard paths, so it doesn't need a lock
// on the shard.
func RebuildShard(ctx context.Context, log logutil.Logger, ts topo.Server, keyspace, shard string, cells []string, timeout time.Duration, interrupted chan struct{}) (*topo.ShardInfo, error) {
log.Infof("RebuildShard %v/%v", keyspace, shard)
span := trace.NewSpanFromContext(ctx)
span.StartLocal("topotools.RebuildShard")
defer span.Finish()
ctx = trace.NewContext(ctx, span)
// read the existing shard info. It has to exist.
shardInfo, err := ts.GetShard(keyspace, shard)
if err != nil {
return nil, err
}
// rebuild all cells in parallel
wg := sync.WaitGroup{}
rec := concurrency.AllErrorRecorder{}
for _, cell := range shardInfo.Cells {
// skip this cell if we shouldn't rebuild it
if !topo.InCellList(cell, cells) {
continue
}
// start with the master if it's in the current cell
tabletsAsMap := make(map[topo.TabletAlias]bool)
if shardInfo.MasterAlias.Cell == cell {
tabletsAsMap[shardInfo.MasterAlias] = true
}
wg.Add(1)
go func(cell string) {
defer wg.Done()
// Lock the SrvShard so we don't race with other rebuilds of the same
// shard in the same cell (e.g. from our peer tablets).
actionNode := actionnode.RebuildSrvShard()
lockPath, err := actionNode.LockSrvShard(ctx, ts, cell, keyspace, shard, timeout, interrupted)
if err != nil {
rec.RecordError(err)
return
}
// read the ShardReplication object to find tablets
sri, err := ts.GetShardReplication(cell, keyspace, shard)
if err != nil {
rec.RecordError(fmt.Errorf("GetShardReplication(%v, %v, %v) failed: %v", cell, keyspace, shard, err))
return
}
// add all relevant tablets to the map
for _, rl := range sri.ReplicationLinks {
tabletsAsMap[rl.TabletAlias] = true
if rl.Parent.Cell == cell {
tabletsAsMap[rl.Parent] = true
}
}
// convert the map to a list
aliases := make([]topo.TabletAlias, 0, len(tabletsAsMap))
for a := range tabletsAsMap {
aliases = append(aliases, a)
}
// read all the Tablet records
tablets, err := topo.GetTabletMap(ctx, ts, aliases)
switch err {
case nil:
// keep going, we're good
case topo.ErrPartialResult:
log.Warningf("Got ErrPartialResult from topo.GetTabletMap in cell %v, some tablets may not be added properly to serving graph", cell)
default:
rec.RecordError(fmt.Errorf("GetTabletMap in cell %v failed: %v", cell, err))
return
}
// write the data we need to
rebuildErr := rebuildCellSrvShard(ctx, log, ts, shardInfo, cell, tablets)
// and unlock
if err := actionNode.UnlockSrvShard(ctx, ts, cell, keyspace, shard, lockPath, rebuildErr); err != nil {
rec.RecordError(err)
}
}(cell)
}
wg.Wait()
return shardInfo, rec.Error()
}