// IsRunning attempts to determine whether this process is
// running on CI. This is done by checking any of:
//
// CI=true
// travis.IsRunning()
// jenkins.IsRunning()
//
func IsRunning() bool {
if os.Getenv(string(VarCI)) == "true" {
return true
}
return travis.IsRunning() || jenkins.IsRunning()
}
func TestDialBackoff(t *testing.T) {
// t.Skip("skipping for another test")
if travis.IsRunning() || jenkins.IsRunning() {
t.Skip("travis and jenkins will never have fun with this test")
}
t.Parallel()
ctx := context.Background()
swarms := makeSwarms(ctx, t, 2)
s1 := swarms[0]
s2 := swarms[1]
defer s1.Close()
defer s2.Close()
s1.dialT = time.Second // lower timeout for tests.
s2.dialT = time.Second // lower timeout for tests.
s2addrs, err := s2.InterfaceListenAddresses()
if err != nil {
t.Fatal(err)
}
s1.peers.AddAddrs(s2.local, s2addrs, peer.PermanentAddrTTL)
// dial to a non-existent peer.
s3p, s3addr, s3l := newSilentPeer(t)
go acceptAndHang(s3l)
defer s3l.Close()
s1.peers.AddAddr(s3p, s3addr, peer.PermanentAddrTTL)
// in this test we will:
// 1) dial 10x to each node.
// 2) all dials should hang
// 3) s1->s2 should succeed.
// 4) s1->s3 should not (and should place s3 on backoff)
// 5) disconnect entirely
// 6) dial 10x to each node again
// 7) s3 dials should all return immediately (except 1)
// 8) s2 dials should all hang, and succeed
// 9) last s3 dial ends, unsuccessful
dialOnlineNode := func(dst peer.ID, times int) <-chan bool {
ch := make(chan bool)
for i := 0; i < times; i++ {
go func() {
if _, err := s1.Dial(ctx, dst); err != nil {
t.Error("error dialing", dst, err)
ch <- false
} else {
ch <- true
}
}()
}
return ch
}
dialOfflineNode := func(dst peer.ID, times int) <-chan bool {
ch := make(chan bool)
for i := 0; i < times; i++ {
go func() {
if c, err := s1.Dial(ctx, dst); err != nil {
ch <- false
} else {
t.Error("succeeded in dialing", dst)
ch <- true
c.Close()
}
}()
}
return ch
}
{
// 1) dial 10x to each node.
N := 10
s2done := dialOnlineNode(s2.local, N)
s3done := dialOfflineNode(s3p, N)
// when all dials should be done by:
dialTimeout1x := time.After(s1.dialT)
// dialTimeout1Ax := time.After(s1.dialT * 2) // dialAttempts)
dialTimeout10Ax := time.After(s1.dialT * 2 * 10) // dialAttempts * 10)
// 2) all dials should hang
select {
case <-s2done:
t.Error("s2 should not happen immediately")
case <-s3done:
t.Error("s3 should not happen yet")
case <-time.After(time.Millisecond):
// s2 may finish very quickly, so let's get out.
}
// 3) s1->s2 should succeed.
for i := 0; i < N; i++ {
select {
case r := <-s2done:
if !r {
t.Error("s2 should not fail")
}
case <-s3done:
t.Error("s3 should not happen yet")
case <-dialTimeout1x:
t.Error("s2 took too long")
}
}
select {
case <-s2done:
t.Error("s2 should have no more")
case <-s3done:
t.Error("s3 should not happen yet")
case <-dialTimeout1x: // let it pass
}
// 4) s1->s3 should not (and should place s3 on backoff)
// N-1 should finish before dialTimeout1x * 2
for i := 0; i < N; i++ {
select {
case <-s2done:
t.Error("s2 should have no more")
case r := <-s3done:
if r {
t.Error("s3 should not succeed")
}
case <-(dialTimeout1x):
if i < (N - 1) {
t.Fatal("s3 took too long")
}
t.Log("dialTimeout1x * 1.3 hit for last peer")
case <-dialTimeout10Ax:
t.Fatal("s3 took too long")
}
}
// check backoff state
if s1.backf.Backoff(s2.local) {
t.Error("s2 should not be on backoff")
}
if !s1.backf.Backoff(s3p) {
t.Error("s3 should be on backoff")
}
// 5) disconnect entirely
for _, c := range s1.Connections() {
c.Close()
}
for i := 0; i < 100 && len(s1.Connections()) > 0; i++ {
<-time.After(time.Millisecond)
}
if len(s1.Connections()) > 0 {
t.Fatal("s1 conns must exit")
}
}
{
// 6) dial 10x to each node again
N := 10
s2done := dialOnlineNode(s2.local, N)
s3done := dialOfflineNode(s3p, N)
// when all dials should be done by:
dialTimeout1x := time.After(s1.dialT)
// dialTimeout1Ax := time.After(s1.dialT * 2) // dialAttempts)
dialTimeout10Ax := time.After(s1.dialT * 2 * 10) // dialAttempts * 10)
// 7) s3 dials should all return immediately (except 1)
for i := 0; i < N-1; i++ {
select {
case <-s2done:
t.Error("s2 should not succeed yet")
case r := <-s3done:
if r {
t.Error("s3 should not succeed")
}
case <-dialTimeout1x:
t.Fatal("s3 took too long")
}
}
// 8) s2 dials should all hang, and succeed
for i := 0; i < N; i++ {
select {
case r := <-s2done:
if !r {
t.Error("s2 should succeed")
}
// case <-s3done:
case <-(dialTimeout1x):
t.Fatal("s3 took too long")
}
}
// 9) the last s3 should return, failed.
select {
case <-s2done:
t.Error("s2 should have no more")
case r := <-s3done:
if r {
t.Error("s3 should not succeed")
}
case <-dialTimeout10Ax:
t.Fatal("s3 took too long")
}
// check backoff state (the same)
if s1.backf.Backoff(s2.local) {
t.Error("s2 should not be on backoff")
}
if !s1.backf.Backoff(s3p) {
t.Error("s3 should be on backoff")
}
}
}