status-go/healthmanager/providers_health_manager_te...

214 lines
8.0 KiB
Go
Raw Normal View History

package healthmanager
import (
"context"
"errors"
"fmt"
"sync"
"testing"
"time"
"github.com/stretchr/testify/suite"
"github.com/status-im/status-go/healthmanager/rpcstatus"
)
type ProvidersHealthManagerSuite struct {
suite.Suite
phm *ProvidersHealthManager
}
// SetupTest initializes the ProvidersHealthManager before each test
func (s *ProvidersHealthManagerSuite) SetupTest() {
s.phm = NewProvidersHealthManager(1)
}
// Helper method to update providers and wait for a notification on the given channel
func (s *ProvidersHealthManagerSuite) updateAndWait(ch <-chan struct{}, statuses []rpcstatus.RpcProviderCallStatus, expectedChainStatus rpcstatus.StatusType, timeout time.Duration) {
s.phm.Update(context.Background(), statuses)
select {
case <-ch:
// Received notification
case <-time.After(timeout):
s.Fail("Timeout waiting for chain status update")
}
s.assertChainStatus(expectedChainStatus)
}
// Helper method to update providers and wait for a notification on the given channel
func (s *ProvidersHealthManagerSuite) updateAndExpectNoNotification(ch <-chan struct{}, statuses []rpcstatus.RpcProviderCallStatus, expectedChainStatus rpcstatus.StatusType, timeout time.Duration) {
s.phm.Update(context.Background(), statuses)
select {
case <-ch:
s.Fail("Unexpected status update")
case <-time.After(timeout):
// No notification as expected
}
s.assertChainStatus(expectedChainStatus)
}
// Helper method to assert the current chain status
func (s *ProvidersHealthManagerSuite) assertChainStatus(expected rpcstatus.StatusType) {
actual := s.phm.Status().Status
s.Equal(expected, actual, fmt.Sprintf("Expected chain status to be %s", expected))
}
func (s *ProvidersHealthManagerSuite) TestInitialStatus() {
s.assertChainStatus(rpcstatus.StatusDown)
}
func (s *ProvidersHealthManagerSuite) TestUpdateProviderStatuses() {
s.updateAndWait(s.phm.Subscribe(), []rpcstatus.RpcProviderCallStatus{
{Name: "Provider1", Timestamp: time.Now(), Err: nil},
{Name: "Provider2", Timestamp: time.Now(), Err: errors.New("connection error")},
}, rpcstatus.StatusUp, time.Second)
statusMap := s.phm.GetStatuses()
s.Len(statusMap, 2, "Expected 2 provider statuses")
s.Equal(rpcstatus.StatusUp, statusMap["Provider1"].Status, "Expected Provider1 status to be Up")
s.Equal(rpcstatus.StatusDown, statusMap["Provider2"].Status, "Expected Provider2 status to be Down")
}
func (s *ProvidersHealthManagerSuite) TestChainStatusUpdatesOnce() {
ch := s.phm.Subscribe()
s.assertChainStatus(rpcstatus.StatusDown)
// Update providers to Down
statuses := []rpcstatus.RpcProviderCallStatus{
{Name: "Provider1", Timestamp: time.Now(), Err: errors.New("error")},
{Name: "Provider2", Timestamp: time.Now(), Err: nil},
}
s.updateAndWait(ch, statuses, rpcstatus.StatusUp, time.Second)
s.updateAndExpectNoNotification(ch, statuses, rpcstatus.StatusUp, 10*time.Millisecond)
}
func (s *ProvidersHealthManagerSuite) TestSubscribeReceivesOnlyOnChange() {
ch := s.phm.Subscribe()
// Update provider to Up and wait for notification
upStatuses := []rpcstatus.RpcProviderCallStatus{
{Name: "Provider1", Timestamp: time.Now(), Err: nil},
}
s.updateAndWait(ch, upStatuses, rpcstatus.StatusUp, time.Second)
// Update provider to Down and wait for notification
downStatuses := []rpcstatus.RpcProviderCallStatus{
{Name: "Provider1", Timestamp: time.Now(), Err: errors.New("some critical error")},
}
s.updateAndWait(ch, downStatuses, rpcstatus.StatusDown, time.Second)
s.updateAndExpectNoNotification(ch, downStatuses, rpcstatus.StatusDown, 10*time.Millisecond)
}
func (s *ProvidersHealthManagerSuite) TestConcurrency() {
var wg sync.WaitGroup
providerCount := 1000
s.phm.Update(context.Background(), []rpcstatus.RpcProviderCallStatus{
{Name: "ProviderUp", Timestamp: time.Now(), Err: nil},
})
ctx := context.Background()
for i := 0; i < providerCount-1; i++ {
wg.Add(1)
go func(i int) {
defer wg.Done()
providerName := fmt.Sprintf("Provider%d", i)
var err error
if i%2 == 0 {
err = errors.New("error")
}
s.phm.Update(ctx, []rpcstatus.RpcProviderCallStatus{
{Name: providerName, Timestamp: time.Now(), Err: err},
})
}(i)
}
wg.Wait()
statuses := s.phm.GetStatuses()
s.Len(statuses, providerCount, "Expected 1000 provider statuses")
chainStatus := s.phm.Status().Status
s.Equal(chainStatus, rpcstatus.StatusUp, "Expected chain status to be either Up or Down")
}
func (s *BlockchainHealthManagerSuite) TestInterleavedChainStatusChanges() {
// Register providers for chains 1, 2, and 3
phm1 := NewProvidersHealthManager(1)
phm2 := NewProvidersHealthManager(2)
phm3 := NewProvidersHealthManager(3)
err := s.manager.RegisterProvidersHealthManager(s.ctx, phm1)
s.Require().NoError(err)
err = s.manager.RegisterProvidersHealthManager(s.ctx, phm2)
s.Require().NoError(err)
err = s.manager.RegisterProvidersHealthManager(s.ctx, phm3)
s.Require().NoError(err)
// Subscribe to status updates
ch := s.manager.Subscribe()
defer s.manager.Unsubscribe(ch)
// Initially, all chains are up
phm1.Update(s.ctx, []rpcstatus.RpcProviderCallStatus{{Name: "provider_chain1", Timestamp: time.Now(), Err: nil}})
phm2.Update(s.ctx, []rpcstatus.RpcProviderCallStatus{{Name: "provider_chain2", Timestamp: time.Now(), Err: nil}})
phm3.Update(s.ctx, []rpcstatus.RpcProviderCallStatus{{Name: "provider_chain3", Timestamp: time.Now(), Err: nil}})
// Wait for the status to propagate
s.waitForUpdate(ch, rpcstatus.StatusUp, 100*time.Millisecond)
// Now chain 1 goes down, and chain 3 goes down at the same time
phm1.Update(s.ctx, []rpcstatus.RpcProviderCallStatus{{Name: "provider_chain1", Timestamp: time.Now(), Err: errors.New("connection error")}})
phm3.Update(s.ctx, []rpcstatus.RpcProviderCallStatus{{Name: "provider_chain3", Timestamp: time.Now(), Err: errors.New("connection error")}})
// Wait for the status to reflect the changes
s.waitForUpdate(ch, rpcstatus.StatusUp, 100*time.Millisecond)
// Check that short status correctly reflects the mixed state
shortStatus := s.manager.GetStatusPerChain()
s.Equal(rpcstatus.StatusUp, shortStatus.Status.Status)
s.Equal(rpcstatus.StatusDown, shortStatus.StatusPerChain[1].Status) // Chain 1 is down
s.Equal(rpcstatus.StatusUp, shortStatus.StatusPerChain[2].Status) // Chain 2 is still up
s.Equal(rpcstatus.StatusDown, shortStatus.StatusPerChain[3].Status) // Chain 3 is down
}
func (s *BlockchainHealthManagerSuite) TestDelayedChainUpdate() {
// Register providers for chains 1 and 2
phm1 := NewProvidersHealthManager(1)
phm2 := NewProvidersHealthManager(2)
err := s.manager.RegisterProvidersHealthManager(s.ctx, phm1)
s.Require().NoError(err)
err = s.manager.RegisterProvidersHealthManager(s.ctx, phm2)
s.Require().NoError(err)
// Subscribe to status updates
ch := s.manager.Subscribe()
defer s.manager.Unsubscribe(ch)
// Initially, both chains are up
phm1.Update(s.ctx, []rpcstatus.RpcProviderCallStatus{{Name: "provider1_chain1", Timestamp: time.Now(), Err: nil}})
phm2.Update(s.ctx, []rpcstatus.RpcProviderCallStatus{{Name: "provider1_chain2", Timestamp: time.Now(), Err: nil}})
s.waitForUpdate(ch, rpcstatus.StatusUp, 100*time.Millisecond)
// Chain 2 goes down
phm2.Update(s.ctx, []rpcstatus.RpcProviderCallStatus{{Name: "provider1_chain2", Timestamp: time.Now(), Err: errors.New("connection error")}})
s.waitForUpdate(ch, rpcstatus.StatusUp, 100*time.Millisecond)
// Chain 1 goes down after a delay
phm1.Update(s.ctx, []rpcstatus.RpcProviderCallStatus{{Name: "provider1_chain1", Timestamp: time.Now(), Err: errors.New("connection error")}})
s.waitForUpdate(ch, rpcstatus.StatusDown, 100*time.Millisecond)
// Check that short status reflects the final state where both chains are down
shortStatus := s.manager.GetStatusPerChain()
s.Equal(rpcstatus.StatusDown, shortStatus.Status.Status)
s.Equal(rpcstatus.StatusDown, shortStatus.StatusPerChain[1].Status) // Chain 1 is down
s.Equal(rpcstatus.StatusDown, shortStatus.StatusPerChain[2].Status) // Chain 2 is down
}
func TestProvidersHealthManagerSuite(t *testing.T) {
suite.Run(t, new(ProvidersHealthManagerSuite))
}