"math"
"os"
"runtime"
+ "sort"
"strings"
"sync"
"time"
}
// Run performs a balance operation using the given config and
-// runOptions. It should only be called once on a given Balancer
-// object. Typical usage:
+// runOptions, and returns RunOptions suitable for passing to a
+// subsequent balance operation.
//
-// err = (&Balancer{}).Run(config, runOptions)
-func (bal *Balancer) Run(config Config, runOptions RunOptions) (err error) {
+// Run should only be called once on a given Balancer object.
+//
+// Typical usage:
+//
+// runOptions, err = (&Balancer{}).Run(config, runOptions)
+func (bal *Balancer) Run(config Config, runOptions RunOptions) (nextRunOptions RunOptions, err error) {
+ nextRunOptions = runOptions
+
bal.Dumper = runOptions.Dumper
bal.Logger = runOptions.Logger
if bal.Logger == nil {
if err = bal.CheckSanityEarly(&config.Client); err != nil {
return
}
- if runOptions.CommitTrash {
+ rs := bal.rendezvousState()
+ if runOptions.CommitTrash && rs != runOptions.SafeRendezvousState {
+ if runOptions.SafeRendezvousState != "" {
+ bal.logf("notice: KeepServices list has changed since last run")
+ }
+ bal.logf("clearing existing trash lists, in case the new rendezvous order differs from previous run")
if err = bal.ClearTrashLists(&config.Client); err != nil {
return
}
+ // The current rendezvous state becomes "safe" (i.e.,
+ // OK to compute changes for that state without
+ // clearing existing trash lists) only now, after we
+ // succeed in clearing existing trash lists.
+ nextRunOptions.SafeRendezvousState = rs
}
if err = bal.GetCurrentState(&config.Client, config.CollectionBatchSize, config.CollectionBuffers); err != nil {
return
return nil
}
+// rendezvousState returns a fingerprint (e.g., a sorted list of
+// UUID+host+port) of the current set of keep services.
+func (bal *Balancer) rendezvousState() string {
+ srvs := make([]string, 0, len(bal.KeepServices))
+ for _, srv := range bal.KeepServices {
+ srvs = append(srvs, srv.String())
+ }
+ sort.Strings(srvs)
+ return strings.Join(srvs, "; ")
+}
+
// ClearTrashLists sends an empty trash list to each keep
// service. Calling this before GetCurrentState avoids races.
//
errs <- fmt.Errorf("%s: %v", srv, err)
return
}
+ if len(errs) > 0 {
+ // Some other goroutine encountered an
+ // error -- any further effort here
+ // will be wasted.
+ return
+ }
bal.logf("%s: add %d replicas to map", srv, len(idx))
bal.BlockStateMap.AddReplicas(srv, idx)
bal.logf("%s: done", srv)
}
}()
- go func() {
- // Send a nil error when all goroutines finish. If
- // this is the first error sent to errs, then
- // everything worked.
- wg.Wait()
- errs <- nil
- }()
- return <-errs
+ wg.Wait()
+ if len(errs) > 0 {
+ return <-errs
+ }
+ return nil
}
func (bal *Balancer) addCollection(coll arvados.Collection) error {