// StartCluster starts a cluster from the relevant flags. All test clusters
// should be created through this command since it sets up the logging in a
// unified way.
func StartCluster(t *testing.T, cfg cluster.TestConfig) (c cluster.Cluster) {
var completed bool
defer func() {
if !completed && c != nil {
c.AssertAndStop(t)
}
}()
if !*flagRemote {
logDir := *flagLogDir
if logDir != "" {
logDir = func(d string) string {
for i := 1; i < 100; i++ {
_, _, fun := caller.Lookup(i)
if testFuncRE.MatchString(fun) {
return filepath.Join(d, fun)
}
}
panic("no caller matching Test(.*) in stack trace")
}(logDir)
}
l := cluster.CreateLocal(cfg, logDir, *flagPrivileged, stopper)
l.Start()
c = l
checkRangeReplication(t, l, 20*time.Second)
completed = true
return l
}
f := farmer(t, "")
c = f
if err := f.Resize(*flagNodes); err != nil {
t.Fatal(err)
}
if err := f.WaitReady(5 * time.Minute); err != nil {
if destroyErr := f.Destroy(t); destroyErr != nil {
t.Fatalf("could not destroy cluster after error %v: %v", err, destroyErr)
}
t.Fatalf("cluster not ready in time: %v", err)
}
checkRangeReplication(t, f, 20*time.Second)
completed = true
return f
}
// StartCluster starts a cluster from the relevant flags. All test clusters
// should be created through this command since it sets up the logging in a
// unified way.
func StartCluster(t *testing.T, cfg cluster.TestConfig) (c cluster.Cluster) {
var completed bool
defer func() {
if !completed && c != nil {
c.AssertAndStop(t)
}
}()
if *flagRemote {
f := farmer(t, "")
c = f
if err := f.Resize(*flagNodes); err != nil {
t.Fatal(err)
}
if err := f.WaitReady(5 * time.Minute); err != nil {
if destroyErr := f.Destroy(t); destroyErr != nil {
t.Fatalf("could not destroy cluster after error %s: %s", err, destroyErr)
}
t.Fatalf("cluster not ready in time: %s", err)
}
} else {
logDir := *flagLogDir
if logDir != "" {
logDir = func(d string) string {
for i := 1; i < 100; i++ {
_, _, fun := caller.Lookup(i)
if testFuncRE.MatchString(fun) {
return filepath.Join(d, fun)
}
}
panic("no caller matching Test(.*) in stack trace")
}(logDir)
}
l := cluster.CreateLocal(cfg, logDir, *flagPrivileged, stopper)
l.Start()
c = l
}
wantedReplicas := 3
if numNodes := c.NumNodes(); numNodes < wantedReplicas {
wantedReplicas = numNodes
}
// Looks silly, but we actually start zero-node clusters in the
// reference tests.
if wantedReplicas > 0 {
ctx := context.TODO()
log.Infof(ctx, "waiting for first range to have %d replicas", wantedReplicas)
util.SucceedsSoon(t, func() error {
select {
case <-stopper:
t.Fatal("interrupted")
case <-time.After(time.Second):
}
// Reconnect on every iteration; gRPC will eagerly tank the connection
// on transport errors. Always talk to node 0 because it's guaranteed
// to exist.
client, dbStopper := c.NewClient(t, 0)
defer dbStopper.Stop()
ctxWithTimeout, cancel := context.WithTimeout(ctx, 5*time.Second)
defer cancel()
var desc roachpb.RangeDescriptor
if err := client.GetProto(ctxWithTimeout, keys.RangeDescriptorKey(roachpb.RKeyMin), &desc); err != nil {
return err
}
foundReplicas := len(desc.Replicas)
if log.V(1) {
log.Infof(ctxWithTimeout, "found %d replicas", foundReplicas)
}
if foundReplicas < wantedReplicas {
return errors.Errorf("expected %d replicas, only found %d", wantedReplicas, foundReplicas)
}
return nil
})
}
completed = true
return c
}
// StartCluster starts a cluster from the relevant flags. All test clusters
// should be created through this command since it sets up the logging in a
// unified way.
func StartCluster(ctx context.Context, t *testing.T, cfg cluster.TestConfig) (c cluster.Cluster) {
var completed bool
defer func() {
if !completed && c != nil {
c.AssertAndStop(ctx, t)
}
}()
if *flagRemote {
f := MakeFarmer(t, "", stopper)
c = f
if err := f.Resize(*flagNodes); err != nil {
t.Fatal(err)
}
if err := f.WaitReady(5 * time.Minute); err != nil {
if destroyErr := f.Destroy(t); destroyErr != nil {
t.Fatalf("could not destroy cluster after error %s: %s", err, destroyErr)
}
t.Fatalf("cluster not ready in time: %s", err)
}
} else {
logDir := *flagLogDir
if logDir != "" {
logDir = func(d string) string {
for i := 1; i < 100; i++ {
_, _, fun := caller.Lookup(i)
if testFuncRE.MatchString(fun) {
return filepath.Join(d, fun)
}
}
panic("no caller matching Test(.*) in stack trace")
}(logDir)
}
l := cluster.CreateLocal(ctx, cfg, logDir, *flagPrivileged, stopper)
l.Start(ctx)
c = l
}
wantedReplicas := 3
if numNodes := c.NumNodes(); numNodes < wantedReplicas {
wantedReplicas = numNodes
}
// Looks silly, but we actually start zero-node clusters in the
// reference tests.
if wantedReplicas > 0 {
log.Infof(ctx, "waiting for first range to have %d replicas", wantedReplicas)
testutils.SucceedsSoon(t, func() error {
select {
case <-stopper.ShouldStop():
t.Fatal("interrupted")
case <-time.After(time.Second):
}
// Reconnect on every iteration; gRPC will eagerly tank the connection
// on transport errors. Always talk to node 0 because it's guaranteed
// to exist.
client, err := c.NewClient(ctx, 0)
if err != nil {
t.Fatal(err)
}
var desc roachpb.RangeDescriptor
if err := client.GetProto(ctx, keys.RangeDescriptorKey(roachpb.RKeyMin), &desc); err != nil {
return err
}
foundReplicas := len(desc.Replicas)
if log.V(1) {
log.Infof(ctx, "found %d replicas", foundReplicas)
}
if foundReplicas < wantedReplicas {
return errors.Errorf("expected %d replicas, only found %d", wantedReplicas, foundReplicas)
}
return nil
})
}
// Ensure that all nodes are serving SQL by making sure a simple
// read-only query succeeds.
for i := 0; i < c.NumNodes(); i++ {
testutils.SucceedsSoon(t, func() error {
db, err := gosql.Open("postgres", c.PGUrl(ctx, i))
if err != nil {
return err
}
if _, err := db.Exec("SHOW DATABASES;"); err != nil {
return err
}
return nil
})
}
completed = true
return c
}
