open-nomad/nomad/leader_test.go

1591 lines
40 KiB
Go
Raw Normal View History

package nomad
2015-06-05 21:54:45 +00:00
import (
"errors"
"fmt"
2020-10-21 04:16:25 +00:00
"sort"
"strconv"
2015-06-05 21:54:45 +00:00
"testing"
2015-08-05 23:32:46 +00:00
"time"
2015-06-05 21:54:45 +00:00
2020-02-16 21:54:28 +00:00
"github.com/hashicorp/consul/sdk/testutil/retry"
"github.com/hashicorp/go-hclog"
2017-02-08 05:22:48 +00:00
memdb "github.com/hashicorp/go-memdb"
"github.com/hashicorp/go-version"
"github.com/hashicorp/nomad/helper"
"github.com/hashicorp/nomad/nomad/mock"
"github.com/hashicorp/nomad/nomad/state"
"github.com/hashicorp/nomad/nomad/structs"
2015-06-05 21:54:45 +00:00
"github.com/hashicorp/nomad/testutil"
2018-04-10 00:51:55 +00:00
"github.com/hashicorp/raft"
"github.com/hashicorp/serf/serf"
"github.com/stretchr/testify/assert"
2018-02-20 23:05:16 +00:00
"github.com/stretchr/testify/require"
2015-06-05 21:54:45 +00:00
)
func TestLeader_LeftServer(t *testing.T) {
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
s1, cleanupS1 := TestServer(t, func(c *Config) {
c.BootstrapExpect = 3
})
defer cleanupS1()
2015-06-05 21:54:45 +00:00
s2, cleanupS2 := TestServer(t, func(c *Config) {
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
c.BootstrapExpect = 3
2015-06-05 21:54:45 +00:00
})
defer cleanupS2()
2015-06-05 21:54:45 +00:00
s3, cleanupS3 := TestServer(t, func(c *Config) {
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
c.BootstrapExpect = 3
2015-06-05 21:54:45 +00:00
})
defer cleanupS3()
2015-06-05 21:54:45 +00:00
servers := []*Server{s1, s2, s3}
2018-01-12 01:00:30 +00:00
TestJoin(t, s1, s2, s3)
2015-06-05 21:54:45 +00:00
for _, s := range servers {
testutil.WaitForResult(func() (bool, error) {
peers, _ := s.numPeers()
2017-02-03 01:50:06 +00:00
return peers == 3, nil
2015-06-05 21:54:45 +00:00
}, func(err error) {
t.Fatalf("should have 3 peers")
})
}
// Kill any server
2017-10-19 21:16:27 +00:00
var peer *Server
for _, s := range servers {
if !s.IsLeader() {
peer = s
break
}
}
if peer == nil {
t.Fatalf("Should have a non-leader")
}
peer.Shutdown()
name := fmt.Sprintf("%s.%s", peer.config.NodeName, peer.config.Region)
2015-06-05 21:54:45 +00:00
testutil.WaitForResult(func() (bool, error) {
2017-10-19 21:16:27 +00:00
for _, s := range servers {
if s == peer {
continue
}
// Force remove the non-leader (transition to left state)
if err := s.RemoveFailedNode(name); err != nil {
return false, err
}
2015-06-05 21:54:45 +00:00
peers, _ := s.numPeers()
2017-02-03 01:50:06 +00:00
return peers == 2, errors.New(fmt.Sprintf("%v", peers))
2015-06-05 21:54:45 +00:00
}
return true, nil
}, func(err error) {
t.Fatalf("err: %s", err)
})
}
func TestLeader_LeftLeader(t *testing.T) {
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
s1, cleanupS1 := TestServer(t, func(c *Config) {
c.BootstrapExpect = 3
})
defer cleanupS1()
2015-06-05 21:54:45 +00:00
s2, cleanupS2 := TestServer(t, func(c *Config) {
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
c.BootstrapExpect = 3
2015-06-05 21:54:45 +00:00
})
defer cleanupS2()
2015-06-05 21:54:45 +00:00
s3, cleanupS3 := TestServer(t, func(c *Config) {
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
c.BootstrapExpect = 3
2015-06-05 21:54:45 +00:00
})
defer cleanupS3()
2015-06-05 21:54:45 +00:00
servers := []*Server{s1, s2, s3}
2018-01-12 01:00:30 +00:00
TestJoin(t, s1, s2, s3)
2015-06-05 21:54:45 +00:00
for _, s := range servers {
testutil.WaitForResult(func() (bool, error) {
peers, _ := s.numPeers()
2017-02-03 01:50:06 +00:00
return peers == 3, nil
2015-06-05 21:54:45 +00:00
}, func(err error) {
t.Fatalf("should have 3 peers")
})
}
// Kill the leader!
leader := waitForStableLeadership(t, servers)
2015-06-05 21:54:45 +00:00
leader.Leave()
leader.Shutdown()
for _, s := range servers {
if s == leader {
continue
}
testutil.WaitForResult(func() (bool, error) {
peers, _ := s.numPeers()
2017-02-03 01:50:06 +00:00
return peers == 2, errors.New(fmt.Sprintf("%v", peers))
2015-06-05 21:54:45 +00:00
}, func(err error) {
t.Fatalf("should have 2 peers: %v", err)
})
}
}
func TestLeader_MultiBootstrap(t *testing.T) {
s1, cleanupS1 := TestServer(t, nil)
defer cleanupS1()
2015-06-05 21:54:45 +00:00
s2, cleanupS2 := TestServer(t, nil)
defer cleanupS2()
2015-06-05 21:54:45 +00:00
servers := []*Server{s1, s2}
2018-01-12 01:00:30 +00:00
TestJoin(t, s1, s2)
2015-06-05 21:54:45 +00:00
for _, s := range servers {
testutil.WaitForResult(func() (bool, error) {
peers := s.Members()
return len(peers) == 2, nil
}, func(err error) {
t.Fatalf("should have 2 peers")
})
}
// Ensure we don't have multiple raft peers
for _, s := range servers {
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
peers, err := s.numPeers()
if err != nil {
t.Fatalf("failed: %v", err)
}
2017-02-03 01:50:06 +00:00
if peers != 1 {
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
t.Fatalf("should only have 1 raft peer! %v", peers)
2015-06-05 21:54:45 +00:00
}
}
}
2015-08-05 23:32:46 +00:00
func TestLeader_PlanQueue_Reset(t *testing.T) {
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
s1, cleanupS1 := TestServer(t, func(c *Config) {
c.BootstrapExpect = 3
})
defer cleanupS1()
2015-08-05 23:32:46 +00:00
s2, cleanupS2 := TestServer(t, func(c *Config) {
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
c.BootstrapExpect = 3
2015-08-05 23:32:46 +00:00
})
defer cleanupS2()
2015-08-05 23:32:46 +00:00
s3, cleanupS3 := TestServer(t, func(c *Config) {
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
c.BootstrapExpect = 3
2015-08-05 23:32:46 +00:00
})
defer cleanupS3()
2015-08-05 23:32:46 +00:00
servers := []*Server{s1, s2, s3}
2018-01-12 01:00:30 +00:00
TestJoin(t, s1, s2, s3)
2015-08-05 23:32:46 +00:00
leader := waitForStableLeadership(t, servers)
2015-08-05 23:32:46 +00:00
if !leader.planQueue.Enabled() {
t.Fatalf("should enable plan queue")
}
for _, s := range servers {
if !s.IsLeader() && s.planQueue.Enabled() {
t.Fatalf("plan queue should not be enabled")
}
}
2015-08-05 23:32:46 +00:00
// Kill the leader
leader.Shutdown()
time.Sleep(100 * time.Millisecond)
// Wait for a new leader
leader = nil
testutil.WaitForResult(func() (bool, error) {
for _, s := range servers {
if s.IsLeader() {
leader = s
return true, nil
}
}
return false, nil
}, func(err error) {
t.Fatalf("should have leader")
})
// Check that the new leader has a pending GC expiration
testutil.WaitForResult(func() (bool, error) {
return leader.planQueue.Enabled(), nil
}, func(err error) {
t.Fatalf("should enable plan queue")
})
}
func TestLeader_EvalBroker_Reset(t *testing.T) {
s1, cleanupS1 := TestServer(t, func(c *Config) {
c.NumSchedulers = 0
})
defer cleanupS1()
s2, cleanupS2 := TestServer(t, func(c *Config) {
c.NumSchedulers = 0
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
c.BootstrapExpect = 3
})
defer cleanupS2()
s3, cleanupS3 := TestServer(t, func(c *Config) {
c.NumSchedulers = 0
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
c.BootstrapExpect = 3
})
defer cleanupS3()
servers := []*Server{s1, s2, s3}
2018-01-12 01:00:30 +00:00
TestJoin(t, s1, s2, s3)
leader := waitForStableLeadership(t, servers)
// Inject a pending eval
req := structs.EvalUpdateRequest{
Evals: []*structs.Evaluation{mock.Eval()},
}
_, _, err := leader.raftApply(structs.EvalUpdateRequestType, req)
if err != nil {
t.Fatalf("err: %v", err)
}
// Kill the leader
leader.Shutdown()
time.Sleep(100 * time.Millisecond)
// Wait for a new leader
leader = nil
testutil.WaitForResult(func() (bool, error) {
for _, s := range servers {
if s.IsLeader() {
leader = s
return true, nil
}
}
return false, nil
}, func(err error) {
t.Fatalf("should have leader")
})
// Check that the new leader has a pending evaluation
testutil.WaitForResult(func() (bool, error) {
stats := leader.evalBroker.Stats()
return stats.TotalReady == 1, nil
}, func(err error) {
t.Fatalf("should have pending evaluation")
})
}
2015-08-15 22:15:00 +00:00
func TestLeader_PeriodicDispatcher_Restore_Adds(t *testing.T) {
s1, cleanupS1 := TestServer(t, func(c *Config) {
c.NumSchedulers = 0
})
defer cleanupS1()
s2, cleanupS2 := TestServer(t, func(c *Config) {
c.NumSchedulers = 0
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
c.BootstrapExpect = 3
})
defer cleanupS2()
s3, cleanupS3 := TestServer(t, func(c *Config) {
c.NumSchedulers = 0
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
c.BootstrapExpect = 3
})
defer cleanupS3()
servers := []*Server{s1, s2, s3}
2018-01-12 01:00:30 +00:00
TestJoin(t, s1, s2, s3)
leader := waitForStableLeadership(t, servers)
// Inject a periodic job, a parameterized periodic job and a non-periodic job
periodic := mock.PeriodicJob()
nonPeriodic := mock.Job()
parameterizedPeriodic := mock.PeriodicJob()
parameterizedPeriodic.ParameterizedJob = &structs.ParameterizedJobConfig{}
for _, job := range []*structs.Job{nonPeriodic, periodic, parameterizedPeriodic} {
req := structs.JobRegisterRequest{
Job: job,
2017-09-07 23:56:15 +00:00
WriteRequest: structs.WriteRequest{
Namespace: job.Namespace,
},
}
_, _, err := leader.raftApply(structs.JobRegisterRequestType, req)
if err != nil {
t.Fatalf("err: %v", err)
}
}
// Kill the leader
leader.Shutdown()
time.Sleep(100 * time.Millisecond)
// Wait for a new leader
leader = nil
testutil.WaitForResult(func() (bool, error) {
for _, s := range servers {
if s.IsLeader() {
leader = s
return true, nil
}
}
return false, nil
}, func(err error) {
t.Fatalf("should have leader")
})
2017-09-07 23:56:15 +00:00
tuplePeriodic := structs.NamespacedID{
ID: periodic.ID,
Namespace: periodic.Namespace,
}
tupleNonPeriodic := structs.NamespacedID{
ID: nonPeriodic.ID,
Namespace: nonPeriodic.Namespace,
}
tupleParameterized := structs.NamespacedID{
ID: parameterizedPeriodic.ID,
Namespace: parameterizedPeriodic.Namespace,
}
// Check that the new leader is tracking the periodic job only
testutil.WaitForResult(func() (bool, error) {
test: fix race and nil panic in nomad/ tests Race was test only and due to unlocked map access. Panic was test only and due to checking a field on a struct even when we knew the struct was nil. Race output that was fixed: ``` ================== WARNING: DATA RACE Read at 0x00c000697dd0 by goroutine 768: runtime.mapaccess2() /usr/local/go/src/runtime/map.go:439 +0x0 github.com/hashicorp/nomad/nomad.TestLeader_PeriodicDispatcher_Restore_Adds.func8() /home/schmichael/go/src/github.com/hashicorp/nomad/nomad/leader_test.go:402 +0xe6 github.com/hashicorp/nomad/testutil.WaitForResultRetries() /home/schmichael/go/src/github.com/hashicorp/nomad/testutil/wait.go:30 +0x5a github.com/hashicorp/nomad/testutil.WaitForResult() /home/schmichael/go/src/github.com/hashicorp/nomad/testutil/wait.go:22 +0x57 github.com/hashicorp/nomad/nomad.TestLeader_PeriodicDispatcher_Restore_Adds() /home/schmichael/go/src/github.com/hashicorp/nomad/nomad/leader_test.go:401 +0xb53 testing.tRunner() /usr/local/go/src/testing/testing.go:827 +0x162 Previous write at 0x00c000697dd0 by goroutine 569: runtime.mapassign() /usr/local/go/src/runtime/map.go:549 +0x0 github.com/hashicorp/nomad/nomad.(*PeriodicDispatch).Add() /home/schmichael/go/src/github.com/hashicorp/nomad/nomad/periodic.go:224 +0x2eb github.com/hashicorp/nomad/nomad.(*Server).restorePeriodicDispatcher() /home/schmichael/go/src/github.com/hashicorp/nomad/nomad/leader.go:394 +0x29a github.com/hashicorp/nomad/nomad.(*Server).establishLeadership() /home/schmichael/go/src/github.com/hashicorp/nomad/nomad/leader.go:234 +0x593 github.com/hashicorp/nomad/nomad.(*Server).leaderLoop() /home/schmichael/go/src/github.com/hashicorp/nomad/nomad/leader.go:117 +0x82e github.com/hashicorp/nomad/nomad.(*Server).monitorLeadership.func1() /home/schmichael/go/src/github.com/hashicorp/nomad/nomad/leader.go:72 +0x6c Goroutine 768 (running) created at: testing.(*T).Run() /usr/local/go/src/testing/testing.go:878 +0x650 testing.runTests.func1() /usr/local/go/src/testing/testing.go:1119 +0xa8 testing.tRunner() /usr/local/go/src/testing/testing.go:827 +0x162 testing.runTests() /usr/local/go/src/testing/testing.go:1117 +0x4ee testing.(*M).Run() /usr/local/go/src/testing/testing.go:1034 +0x2ee main.main() _testmain.go:1150 +0x221 Goroutine 569 (running) created at: github.com/hashicorp/nomad/nomad.(*Server).monitorLeadership() /home/schmichael/go/src/github.com/hashicorp/nomad/nomad/leader.go:70 +0x269 ================== ```
2018-12-19 21:05:12 +00:00
leader.periodicDispatcher.l.Lock()
defer leader.periodicDispatcher.l.Unlock()
2017-09-07 23:56:15 +00:00
if _, tracked := leader.periodicDispatcher.tracked[tuplePeriodic]; !tracked {
return false, fmt.Errorf("periodic job not tracked")
}
2017-09-07 23:56:15 +00:00
if _, tracked := leader.periodicDispatcher.tracked[tupleNonPeriodic]; tracked {
return false, fmt.Errorf("non periodic job tracked")
}
2017-09-07 23:56:15 +00:00
if _, tracked := leader.periodicDispatcher.tracked[tupleParameterized]; tracked {
return false, fmt.Errorf("parameterized periodic job tracked")
}
return true, nil
}, func(err error) {
t.Fatalf(err.Error())
})
}
func TestLeader_PeriodicDispatcher_Restore_NoEvals(t *testing.T) {
s1, cleanupS1 := TestServer(t, func(c *Config) {
c.NumSchedulers = 0
})
defer cleanupS1()
2015-12-05 00:53:36 +00:00
testutil.WaitForLeader(t, s1.RPC)
// Inject a periodic job that will be triggered soon.
launch := time.Now().Add(1 * time.Second)
job := testPeriodicJob(launch)
req := structs.JobRegisterRequest{
Job: job,
2017-09-07 23:56:15 +00:00
WriteRequest: structs.WriteRequest{
Namespace: job.Namespace,
},
}
_, _, err := s1.raftApply(structs.JobRegisterRequestType, req)
if err != nil {
t.Fatalf("err: %v", err)
}
// Flush the periodic dispatcher, ensuring that no evals will be created.
2015-12-05 00:53:36 +00:00
s1.periodicDispatcher.SetEnabled(false)
// Get the current time to ensure the launch time is after this once we
// restore.
now := time.Now()
// Sleep till after the job should have been launched.
time.Sleep(3 * time.Second)
// Restore the periodic dispatcher.
2015-12-05 00:53:36 +00:00
s1.periodicDispatcher.SetEnabled(true)
s1.restorePeriodicDispatcher()
// Ensure the job is tracked.
2017-09-07 23:56:15 +00:00
tuple := structs.NamespacedID{
ID: job.ID,
Namespace: job.Namespace,
}
if _, tracked := s1.periodicDispatcher.tracked[tuple]; !tracked {
t.Fatalf("periodic job not restored")
}
// Check that an eval was made.
2017-02-08 05:22:48 +00:00
ws := memdb.NewWatchSet()
2017-09-07 23:56:15 +00:00
last, err := s1.fsm.State().PeriodicLaunchByID(ws, job.Namespace, job.ID)
if err != nil || last == nil {
t.Fatalf("failed to get periodic launch time: %v", err)
}
if last.Launch.Before(now) {
t.Fatalf("restorePeriodicDispatcher did not force launch: last %v; want after %v", last.Launch, now)
}
}
func TestLeader_PeriodicDispatcher_Restore_Evals(t *testing.T) {
s1, cleanupS1 := TestServer(t, func(c *Config) {
c.NumSchedulers = 0
})
defer cleanupS1()
2015-12-05 00:53:36 +00:00
testutil.WaitForLeader(t, s1.RPC)
// Inject a periodic job that triggered once in the past, should trigger now
// and once in the future.
now := time.Now()
past := now.Add(-1 * time.Second)
future := now.Add(10 * time.Second)
job := testPeriodicJob(past, now, future)
req := structs.JobRegisterRequest{
Job: job,
2017-09-07 23:56:15 +00:00
WriteRequest: structs.WriteRequest{
Namespace: job.Namespace,
},
}
_, _, err := s1.raftApply(structs.JobRegisterRequestType, req)
if err != nil {
t.Fatalf("err: %v", err)
}
// Create an eval for the past launch.
s1.periodicDispatcher.createEval(job, past)
// Flush the periodic dispatcher, ensuring that no evals will be created.
2015-12-05 00:53:36 +00:00
s1.periodicDispatcher.SetEnabled(false)
// Sleep till after the job should have been launched.
2015-12-05 00:53:36 +00:00
time.Sleep(3 * time.Second)
// Restore the periodic dispatcher.
2015-12-05 00:53:36 +00:00
s1.periodicDispatcher.SetEnabled(true)
s1.restorePeriodicDispatcher()
// Ensure the job is tracked.
2017-09-07 23:56:15 +00:00
tuple := structs.NamespacedID{
ID: job.ID,
Namespace: job.Namespace,
}
if _, tracked := s1.periodicDispatcher.tracked[tuple]; !tracked {
t.Fatalf("periodic job not restored")
}
// Check that an eval was made.
2017-02-08 05:22:48 +00:00
ws := memdb.NewWatchSet()
2017-09-07 23:56:15 +00:00
last, err := s1.fsm.State().PeriodicLaunchByID(ws, job.Namespace, job.ID)
if err != nil || last == nil {
t.Fatalf("failed to get periodic launch time: %v", err)
}
if last.Launch == past {
t.Fatalf("restorePeriodicDispatcher did not force launch")
}
}
2015-08-15 22:15:00 +00:00
func TestLeader_PeriodicDispatch(t *testing.T) {
s1, cleanupS1 := TestServer(t, func(c *Config) {
2015-08-15 22:15:00 +00:00
c.NumSchedulers = 0
c.EvalGCInterval = 5 * time.Millisecond
})
defer cleanupS1()
2015-08-15 22:15:00 +00:00
// Wait for a periodic dispatch
testutil.WaitForResult(func() (bool, error) {
stats := s1.evalBroker.Stats()
bySched, ok := stats.ByScheduler[structs.JobTypeCore]
if !ok {
return false, nil
}
return bySched.Ready > 0, nil
}, func(err error) {
t.Fatalf("should pending job")
})
}
2015-08-16 18:10:18 +00:00
func TestLeader_ReapFailedEval(t *testing.T) {
s1, cleanupS1 := TestServer(t, func(c *Config) {
2015-08-16 18:10:18 +00:00
c.NumSchedulers = 0
c.EvalDeliveryLimit = 1
})
defer cleanupS1()
2015-08-16 18:10:18 +00:00
testutil.WaitForLeader(t, s1.RPC)
// Wait for a periodic dispatch
eval := mock.Eval()
s1.evalBroker.Enqueue(eval)
2015-08-16 18:10:18 +00:00
// Dequeue and Nack
out, token, err := s1.evalBroker.Dequeue(defaultSched, time.Second)
if err != nil {
t.Fatalf("err: %v", err)
}
s1.evalBroker.Nack(out.ID, token)
// Wait for an updated and followup evaluation
2015-08-16 18:10:18 +00:00
state := s1.fsm.State()
testutil.WaitForResult(func() (bool, error) {
2017-02-08 05:22:48 +00:00
ws := memdb.NewWatchSet()
out, err := state.EvalByID(ws, eval.ID)
2015-08-16 18:10:18 +00:00
if err != nil {
return false, err
}
if out == nil {
return false, fmt.Errorf("expect original evaluation to exist")
}
if out.Status != structs.EvalStatusFailed {
return false, fmt.Errorf("got status %v; want %v", out.Status, structs.EvalStatusFailed)
}
if out.NextEval == "" {
return false, fmt.Errorf("got empty NextEval")
}
// See if there is a followup
2017-09-07 23:56:15 +00:00
evals, err := state.EvalsByJob(ws, eval.Namespace, eval.JobID)
if err != nil {
return false, err
}
if l := len(evals); l != 2 {
return false, fmt.Errorf("got %d evals, want 2", l)
}
for _, e := range evals {
if e.ID == eval.ID {
continue
}
if e.Status != structs.EvalStatusPending {
return false, fmt.Errorf("follow up eval has status %v; want %v",
e.Status, structs.EvalStatusPending)
}
if e.ID != out.NextEval {
return false, fmt.Errorf("follow up eval id is %v; orig eval NextEval %v",
e.ID, out.NextEval)
}
2017-04-14 22:24:55 +00:00
if e.Wait < s1.config.EvalFailedFollowupBaselineDelay ||
e.Wait > s1.config.EvalFailedFollowupBaselineDelay+s1.config.EvalFailedFollowupDelayRange {
return false, fmt.Errorf("bad wait: %v", e.Wait)
}
if e.TriggeredBy != structs.EvalTriggerFailedFollowUp {
return false, fmt.Errorf("follow up eval TriggeredBy %v; want %v",
e.TriggeredBy, structs.EvalTriggerFailedFollowUp)
}
}
return true, nil
2015-08-16 18:10:18 +00:00
}, func(err error) {
t.Fatalf("err: %v", err)
})
}
func TestLeader_ReapDuplicateEval(t *testing.T) {
s1, cleanupS1 := TestServer(t, func(c *Config) {
c.NumSchedulers = 0
})
defer cleanupS1()
testutil.WaitForLeader(t, s1.RPC)
// Create a duplicate blocked eval
eval := mock.Eval()
2018-11-07 19:59:24 +00:00
eval.CreateIndex = 100
eval2 := mock.Eval()
eval2.JobID = eval.JobID
2018-11-07 19:59:24 +00:00
eval2.CreateIndex = 102
s1.blockedEvals.Block(eval)
s1.blockedEvals.Block(eval2)
// Wait for the evaluation to marked as cancelled
state := s1.fsm.State()
testutil.WaitForResult(func() (bool, error) {
2017-02-08 05:22:48 +00:00
ws := memdb.NewWatchSet()
2018-11-07 19:59:24 +00:00
out, err := state.EvalByID(ws, eval.ID)
if err != nil {
return false, err
}
return out != nil && out.Status == structs.EvalStatusCancelled, nil
}, func(err error) {
t.Fatalf("err: %v", err)
})
}
func TestLeader_revokeVaultAccessorsOnRestore(t *testing.T) {
s1, cleanupS1 := TestServer(t, func(c *Config) {
c.NumSchedulers = 0
})
defer cleanupS1()
testutil.WaitForLeader(t, s1.RPC)
// Insert a vault accessor that should be revoked
fsmState := s1.fsm.State()
va := mock.VaultAccessor()
if err := fsmState.UpsertVaultAccessor(100, []*structs.VaultAccessor{va}); err != nil {
t.Fatalf("bad: %v", err)
}
// Swap the Vault client
tvc := &TestVaultClient{}
s1.vault = tvc
// Do a restore
if err := s1.revokeVaultAccessorsOnRestore(); err != nil {
t.Fatalf("Failed to restore: %v", err)
}
if len(tvc.RevokedTokens) != 1 && tvc.RevokedTokens[0].Accessor != va.Accessor {
t.Fatalf("Bad revoked accessors: %v", tvc.RevokedTokens)
}
}
func TestLeader_revokeSITokenAccessorsOnRestore(t *testing.T) {
t.Parallel()
r := require.New(t)
s1, cleanupS1 := TestServer(t, func(c *Config) {
c.NumSchedulers = 0
})
defer cleanupS1()
testutil.WaitForLeader(t, s1.RPC)
// replace consul ACLs API with a mock for tracking calls in tests
var consulACLsAPI mockConsulACLsAPI
s1.consulACLs = &consulACLsAPI
// Insert a SI token accessor that should be revoked
fsmState := s1.fsm.State()
accessor := mock.SITokenAccessor()
err := fsmState.UpsertSITokenAccessors(100, []*structs.SITokenAccessor{accessor})
r.NoError(err)
// Do a restore
err = s1.revokeSITokenAccessorsOnRestore()
r.NoError(err)
// Check the accessor was revoked
exp := []revokeRequest{{
accessorID: accessor.AccessorID,
committed: true,
}}
r.ElementsMatch(exp, consulACLsAPI.revokeRequests)
}
func TestLeader_ClusterID(t *testing.T) {
t.Parallel()
s1, cleanupS1 := TestServer(t, func(c *Config) {
c.NumSchedulers = 0
c.Build = minClusterIDVersion.String()
})
defer cleanupS1()
testutil.WaitForLeader(t, s1.RPC)
clusterID, err := s1.ClusterID()
require.NoError(t, err)
require.True(t, helper.IsUUID(clusterID))
}
func TestLeader_ClusterID_upgradePath(t *testing.T) {
t.Parallel()
before := version.Must(version.NewVersion("0.10.1")).String()
after := minClusterIDVersion.String()
type server struct {
s *Server
cleanup func()
}
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
outdated := func() server {
s, cleanup := TestServer(t, func(c *Config) {
c.NumSchedulers = 0
c.Build = before
c.BootstrapExpect = 3
c.Logger.SetLevel(hclog.Trace)
})
return server{s: s, cleanup: cleanup}
}
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
upgraded := func() server {
s, cleanup := TestServer(t, func(c *Config) {
c.NumSchedulers = 0
c.Build = after
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
c.BootstrapExpect = 0
c.Logger.SetLevel(hclog.Trace)
})
return server{s: s, cleanup: cleanup}
}
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
servers := []server{outdated(), outdated(), outdated()}
// fallback shutdown attempt in case testing fails
defer servers[0].cleanup()
defer servers[1].cleanup()
defer servers[2].cleanup()
upgrade := func(i int) {
previous := servers[i]
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
servers[i] = upgraded()
TestJoin(t, servers[i].s, servers[(i+1)%3].s, servers[(i+2)%3].s)
testutil.WaitForLeader(t, servers[i].s.RPC)
require.NoError(t, previous.s.Leave())
require.NoError(t, previous.s.Shutdown())
}
// Join the servers before doing anything
TestJoin(t, servers[0].s, servers[1].s, servers[2].s)
// Wait for servers to settle
for i := 0; i < len(servers); i++ {
testutil.WaitForLeader(t, servers[i].s.RPC)
}
// A check that ClusterID is not available yet
noIDYet := func() {
for _, s := range servers {
retry.Run(t, func(r *retry.R) {
if _, err := s.s.ClusterID(); err == nil {
r.Error("expected error")
}
})
}
}
// Replace first old server with new server
upgrade(0)
defer servers[0].cleanup()
noIDYet() // ClusterID should not work yet, servers: [new, old, old]
// Replace second old server with new server
upgrade(1)
defer servers[1].cleanup()
noIDYet() // ClusterID should not work yet, servers: [new, new, old]
// Replace third / final old server with new server
upgrade(2)
defer servers[2].cleanup()
// Wait for old servers to really be gone
for _, s := range servers {
testutil.WaitForResult(func() (bool, error) {
peers, _ := s.s.numPeers()
return peers == 3, nil
}, func(_ error) {
t.Fatalf("should have 3 peers")
})
}
// Now we can tickle the leader into making a cluster ID
leaderID := ""
for _, s := range servers {
if s.s.IsLeader() {
id, err := s.s.ClusterID()
require.NoError(t, err)
leaderID = id
break
}
}
require.True(t, helper.IsUUID(leaderID))
// Now every participating server has been upgraded, each one should be
// able to get the cluster ID, having been plumbed all the way through.
agreeClusterID(t, []*Server{servers[0].s, servers[1].s, servers[2].s})
}
func TestLeader_ClusterID_noUpgrade(t *testing.T) {
t.Parallel()
type server struct {
s *Server
cleanup func()
}
s1, cleanupS1 := TestServer(t, func(c *Config) {
c.Logger.SetLevel(hclog.Trace)
c.NumSchedulers = 0
c.Build = minClusterIDVersion.String()
c.BootstrapExpect = 3
})
defer cleanupS1()
s2, cleanupS2 := TestServer(t, func(c *Config) {
c.Logger.SetLevel(hclog.Trace)
c.NumSchedulers = 0
c.Build = minClusterIDVersion.String()
c.BootstrapExpect = 3
})
defer cleanupS2()
s3, cleanupS3 := TestServer(t, func(c *Config) {
c.Logger.SetLevel(hclog.Trace)
c.NumSchedulers = 0
c.Build = minClusterIDVersion.String()
c.BootstrapExpect = 3
})
defer cleanupS3()
servers := []*Server{s1, s2, s3}
// Join the servers before doing anything
TestJoin(t, servers[0], servers[1], servers[2])
// Wait for servers to settle
for i := 0; i < len(servers); i++ {
testutil.WaitForLeader(t, servers[i].RPC)
}
// Each server started at the minimum version, check there should be only 1
// cluster ID they all agree on.
agreeClusterID(t, []*Server{servers[0], servers[1], servers[2]})
}
func agreeClusterID(t *testing.T, servers []*Server) {
retries := &retry.Timer{Timeout: 60 * time.Second, Wait: 1 * time.Second}
ids := make([]string, 3)
for i, s := range servers {
retry.RunWith(retries, t, func(r *retry.R) {
id, err := s.ClusterID()
if err != nil {
r.Error(err.Error())
return
}
if !helper.IsUUID(id) {
r.Error("not a UUID")
return
}
ids[i] = id
})
}
require.True(t, ids[0] == ids[1] && ids[1] == ids[2], "ids[0] %s, ids[1] %s, ids[2] %s", ids[0], ids[1], ids[2])
}
func TestLeader_ReplicateACLPolicies(t *testing.T) {
t.Parallel()
s1, root, cleanupS1 := TestACLServer(t, func(c *Config) {
c.Region = "region1"
c.AuthoritativeRegion = "region1"
c.ACLEnabled = true
})
defer cleanupS1()
s2, _, cleanupS2 := TestACLServer(t, func(c *Config) {
c.Region = "region2"
c.AuthoritativeRegion = "region1"
c.ACLEnabled = true
c.ReplicationBackoff = 20 * time.Millisecond
c.ReplicationToken = root.SecretID
})
defer cleanupS2()
2018-01-12 01:00:30 +00:00
TestJoin(t, s1, s2)
testutil.WaitForLeader(t, s1.RPC)
testutil.WaitForLeader(t, s2.RPC)
// Write a policy to the authoritative region
p1 := mock.ACLPolicy()
if err := s1.State().UpsertACLPolicies(structs.MsgTypeTestSetup, 100, []*structs.ACLPolicy{p1}); err != nil {
t.Fatalf("bad: %v", err)
}
// Wait for the policy to replicate
testutil.WaitForResult(func() (bool, error) {
state := s2.State()
out, err := state.ACLPolicyByName(nil, p1.Name)
return out != nil, err
}, func(err error) {
t.Fatalf("should replicate policy")
})
}
func TestLeader_DiffACLPolicies(t *testing.T) {
t.Parallel()
2017-10-13 21:36:02 +00:00
state := state.TestStateStore(t)
// Populate the local state
p1 := mock.ACLPolicy()
p2 := mock.ACLPolicy()
p3 := mock.ACLPolicy()
assert.Nil(t, state.UpsertACLPolicies(structs.MsgTypeTestSetup, 100, []*structs.ACLPolicy{p1, p2, p3}))
// Simulate a remote list
p2Stub := p2.Stub()
p2Stub.ModifyIndex = 50 // Ignored, same index
p3Stub := p3.Stub()
p3Stub.ModifyIndex = 100 // Updated, higher index
p3Stub.Hash = []byte{0, 1, 2, 3}
p4 := mock.ACLPolicy()
remoteList := []*structs.ACLPolicyListStub{
p2Stub,
p3Stub,
p4.Stub(),
}
delete, update := diffACLPolicies(state, 50, remoteList)
// P1 does not exist on the remote side, should delete
assert.Equal(t, []string{p1.Name}, delete)
// P2 is un-modified - ignore. P3 modified, P4 new.
assert.Equal(t, []string{p3.Name, p4.Name}, update)
}
2017-08-13 23:45:13 +00:00
func TestLeader_ReplicateACLTokens(t *testing.T) {
t.Parallel()
s1, root, cleanupS1 := TestACLServer(t, func(c *Config) {
2017-08-13 23:45:13 +00:00
c.Region = "region1"
c.AuthoritativeRegion = "region1"
c.ACLEnabled = true
})
defer cleanupS1()
s2, _, cleanupS2 := TestACLServer(t, func(c *Config) {
2017-08-13 23:45:13 +00:00
c.Region = "region2"
c.AuthoritativeRegion = "region1"
c.ACLEnabled = true
c.ReplicationBackoff = 20 * time.Millisecond
c.ReplicationToken = root.SecretID
2017-08-13 23:45:13 +00:00
})
defer cleanupS2()
2018-01-12 01:00:30 +00:00
TestJoin(t, s1, s2)
2017-08-13 23:45:13 +00:00
testutil.WaitForLeader(t, s1.RPC)
testutil.WaitForLeader(t, s2.RPC)
// Write a token to the authoritative region
p1 := mock.ACLToken()
p1.Global = true
if err := s1.State().UpsertACLTokens(structs.MsgTypeTestSetup, 100, []*structs.ACLToken{p1}); err != nil {
2017-08-13 23:45:13 +00:00
t.Fatalf("bad: %v", err)
}
// Wait for the token to replicate
testutil.WaitForResult(func() (bool, error) {
state := s2.State()
out, err := state.ACLTokenByAccessorID(nil, p1.AccessorID)
return out != nil, err
}, func(err error) {
t.Fatalf("should replicate token")
})
}
func TestLeader_DiffACLTokens(t *testing.T) {
t.Parallel()
2017-10-13 21:36:02 +00:00
state := state.TestStateStore(t)
2017-08-13 23:45:13 +00:00
// Populate the local state
p0 := mock.ACLToken()
p1 := mock.ACLToken()
p1.Global = true
p2 := mock.ACLToken()
p2.Global = true
p3 := mock.ACLToken()
p3.Global = true
assert.Nil(t, state.UpsertACLTokens(structs.MsgTypeTestSetup, 100, []*structs.ACLToken{p0, p1, p2, p3}))
2017-08-13 23:45:13 +00:00
// Simulate a remote list
p2Stub := p2.Stub()
p2Stub.ModifyIndex = 50 // Ignored, same index
p3Stub := p3.Stub()
p3Stub.ModifyIndex = 100 // Updated, higher index
p3Stub.Hash = []byte{0, 1, 2, 3}
2017-08-13 23:45:13 +00:00
p4 := mock.ACLToken()
p4.Global = true
remoteList := []*structs.ACLTokenListStub{
p2Stub,
p3Stub,
p4.Stub(),
}
delete, update := diffACLTokens(state, 50, remoteList)
// P0 is local and should be ignored
// P1 does not exist on the remote side, should delete
assert.Equal(t, []string{p1.AccessorID}, delete)
// P2 is un-modified - ignore. P3 modified, P4 new.
assert.Equal(t, []string{p3.AccessorID, p4.AccessorID}, update)
}
func TestLeader_UpgradeRaftVersion(t *testing.T) {
t.Parallel()
s1, cleanupS1 := TestServer(t, func(c *Config) {
2018-01-12 01:00:30 +00:00
c.Datacenter = "dc1"
c.RaftConfig.ProtocolVersion = 2
})
defer cleanupS1()
s2, cleanupS2 := TestServer(t, func(c *Config) {
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
c.BootstrapExpect = 3
c.RaftConfig.ProtocolVersion = 1
})
defer cleanupS2()
s3, cleanupS3 := TestServer(t, func(c *Config) {
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
c.BootstrapExpect = 3
c.RaftConfig.ProtocolVersion = 2
})
defer cleanupS3()
servers := []*Server{s1, s2, s3}
// Try to join
2018-01-12 01:00:30 +00:00
TestJoin(t, s1, s2, s3)
for _, s := range servers {
testutil.WaitForResult(func() (bool, error) {
peers, _ := s.numPeers()
return peers == 3, nil
}, func(err error) {
t.Fatalf("should have 3 peers")
})
}
// Kill the v1 server
if err := s2.Leave(); err != nil {
t.Fatal(err)
}
for _, s := range []*Server{s1, s3} {
minVer, err := s.autopilot.MinRaftProtocol()
if err != nil {
t.Fatal(err)
}
if got, want := minVer, 2; got != want {
t.Fatalf("got min raft version %d want %d", got, want)
}
}
// Replace the dead server with one running raft protocol v3
s4, cleanupS4 := TestServer(t, func(c *Config) {
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
c.BootstrapExpect = 3
c.Datacenter = "dc1"
c.RaftConfig.ProtocolVersion = 3
})
defer cleanupS4()
2018-01-12 01:00:30 +00:00
TestJoin(t, s1, s4)
servers[1] = s4
// Make sure we're back to 3 total peers with the new one added via ID
for _, s := range servers {
testutil.WaitForResult(func() (bool, error) {
addrs := 0
ids := 0
future := s.raft.GetConfiguration()
if err := future.Error(); err != nil {
return false, err
}
for _, server := range future.Configuration().Servers {
if string(server.ID) == string(server.Address) {
addrs++
} else {
ids++
}
}
if got, want := addrs, 2; got != want {
return false, fmt.Errorf("got %d server addresses want %d", got, want)
}
if got, want := ids, 1; got != want {
return false, fmt.Errorf("got %d server ids want %d", got, want)
}
return true, nil
}, func(err error) {
t.Fatal(err)
})
}
}
func TestLeader_Reelection(t *testing.T) {
raftProtocols := []int{1, 2, 3}
for _, p := range raftProtocols {
t.Run(fmt.Sprintf("Leader Election - Protocol version %d", p), func(t *testing.T) {
leaderElectionTest(t, raft.ProtocolVersion(p))
})
}
}
func leaderElectionTest(t *testing.T, raftProtocol raft.ProtocolVersion) {
s1, cleanupS1 := TestServer(t, func(c *Config) {
2018-04-10 00:51:55 +00:00
c.BootstrapExpect = 3
c.RaftConfig.ProtocolVersion = raftProtocol
})
defer cleanupS1()
s2, cleanupS2 := TestServer(t, func(c *Config) {
2018-04-10 00:51:55 +00:00
c.BootstrapExpect = 3
c.RaftConfig.ProtocolVersion = raftProtocol
})
defer cleanupS2()
s3, cleanupS3 := TestServer(t, func(c *Config) {
2018-04-10 00:51:55 +00:00
c.BootstrapExpect = 3
c.RaftConfig.ProtocolVersion = raftProtocol
})
defer cleanupS3() // todo(shoenig) added this, should be here right??
servers := []*Server{s1, s2, s3}
// Try to join
TestJoin(t, s1, s2, s3)
2018-04-10 00:51:55 +00:00
testutil.WaitForLeader(t, s1.RPC)
2018-04-10 00:51:55 +00:00
testutil.WaitForResult(func() (bool, error) {
future := s1.raft.GetConfiguration()
if err := future.Error(); err != nil {
return false, err
}
2018-04-10 00:51:55 +00:00
for _, server := range future.Configuration().Servers {
if server.Suffrage == raft.Nonvoter {
return false, fmt.Errorf("non-voter %v", server)
}
}
return true, nil
}, func(err error) {
t.Fatal(err)
})
2018-04-10 00:51:55 +00:00
var leader, nonLeader *Server
for _, s := range servers {
if s.IsLeader() {
leader = s
} else {
nonLeader = s
}
}
// Shutdown the leader
leader.Shutdown()
2018-04-10 13:16:50 +00:00
// Wait for new leader to elect
testutil.WaitForLeader(t, nonLeader.RPC)
}
func TestLeader_RollRaftServer(t *testing.T) {
t.Parallel()
s1, cleanupS1 := TestServer(t, func(c *Config) {
c.RaftConfig.ProtocolVersion = 2
})
defer cleanupS1()
s2, cleanupS2 := TestServer(t, func(c *Config) {
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
c.BootstrapExpect = 3
c.RaftConfig.ProtocolVersion = 2
})
defer cleanupS2()
s3, cleanupS3 := TestServer(t, func(c *Config) {
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
c.BootstrapExpect = 3
c.RaftConfig.ProtocolVersion = 2
})
defer cleanupS3()
servers := []*Server{s1, s2, s3}
// Try to join
2018-01-12 01:00:30 +00:00
TestJoin(t, s1, s2, s3)
for _, s := range servers {
retry.Run(t, func(r *retry.R) { r.Check(wantPeers(s, 3)) })
}
// Kill the first v2 server
s1.Shutdown()
for _, s := range []*Server{s1, s3} {
retry.Run(t, func(r *retry.R) {
minVer, err := s.autopilot.MinRaftProtocol()
if err != nil {
r.Fatal(err)
}
if got, want := minVer, 2; got != want {
r.Fatalf("got min raft version %d want %d", got, want)
}
})
}
// Replace the dead server with one running raft protocol v3
s4, cleanupS4 := TestServer(t, func(c *Config) {
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
c.BootstrapExpect = 3
c.RaftConfig.ProtocolVersion = 3
})
defer cleanupS4()
TestJoin(t, s4, s2)
servers[0] = s4
// Kill the second v2 server
s2.Shutdown()
for _, s := range []*Server{s3, s4} {
retry.Run(t, func(r *retry.R) {
minVer, err := s.autopilot.MinRaftProtocol()
if err != nil {
r.Fatal(err)
}
if got, want := minVer, 2; got != want {
r.Fatalf("got min raft version %d want %d", got, want)
}
})
}
// Replace another dead server with one running raft protocol v3
s5, cleanupS5 := TestServer(t, func(c *Config) {
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
c.BootstrapExpect = 3
c.RaftConfig.ProtocolVersion = 3
})
defer cleanupS5()
TestJoin(t, s5, s4)
servers[1] = s5
// Kill the last v2 server, now minRaftProtocol should be 3
s3.Shutdown()
for _, s := range []*Server{s4, s5} {
retry.Run(t, func(r *retry.R) {
minVer, err := s.autopilot.MinRaftProtocol()
if err != nil {
r.Fatal(err)
}
if got, want := minVer, 3; got != want {
r.Fatalf("got min raft version %d want %d", got, want)
}
})
}
// Replace the last dead server with one running raft protocol v3
s6, cleanupS6 := TestServer(t, func(c *Config) {
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
c.BootstrapExpect = 3
c.RaftConfig.ProtocolVersion = 3
})
defer cleanupS6()
TestJoin(t, s6, s4)
servers[2] = s6
// Make sure all the dead servers are removed and we're back to 3 total peers
for _, s := range servers {
retry.Run(t, func(r *retry.R) {
addrs := 0
ids := 0
future := s.raft.GetConfiguration()
if err := future.Error(); err != nil {
r.Fatal(err)
}
for _, server := range future.Configuration().Servers {
if string(server.ID) == string(server.Address) {
addrs++
} else {
ids++
}
}
if got, want := addrs, 0; got != want {
r.Fatalf("got %d server addresses want %d", got, want)
}
if got, want := ids, 3; got != want {
r.Fatalf("got %d server ids want %d", got, want)
}
})
}
}
2018-02-20 23:05:16 +00:00
func TestLeader_RevokeLeadership_MultipleTimes(t *testing.T) {
s1, cleanupS1 := TestServer(t, nil)
defer cleanupS1()
2018-02-20 23:05:16 +00:00
testutil.WaitForLeader(t, s1.RPC)
testutil.WaitForResult(func() (bool, error) {
return s1.evalBroker.Enabled(), nil
}, func(err error) {
t.Fatalf("should have finished establish leader loop")
})
require.Nil(t, s1.revokeLeadership())
require.Nil(t, s1.revokeLeadership())
require.Nil(t, s1.revokeLeadership())
}
func TestLeader_TransitionsUpdateConsistencyRead(t *testing.T) {
s1, cleanupS1 := TestServer(t, nil)
defer cleanupS1()
testutil.WaitForLeader(t, s1.RPC)
testutil.WaitForResult(func() (bool, error) {
return s1.isReadyForConsistentReads(), nil
}, func(err error) {
require.Fail(t, "should have finished establish leader loop")
})
require.Nil(t, s1.revokeLeadership())
require.False(t, s1.isReadyForConsistentReads())
ch := make(chan struct{})
require.Nil(t, s1.establishLeadership(ch))
require.True(t, s1.isReadyForConsistentReads())
}
// TestLeader_PausingWorkers asserts that scheduling workers are paused
// (and unpaused) upon leader elections (and step downs).
func TestLeader_PausingWorkers(t *testing.T) {
s1, cleanupS1 := TestServer(t, func(c *Config) {
c.NumSchedulers = 12
})
defer cleanupS1()
testutil.WaitForLeader(t, s1.RPC)
require.Len(t, s1.workers, 12)
pausedWorkers := func() int {
c := 0
for _, w := range s1.workers {
w.pauseLock.Lock()
if w.paused {
c++
}
w.pauseLock.Unlock()
}
return c
}
// pause 3/4 of the workers
require.Equal(t, 9, pausedWorkers())
err := s1.revokeLeadership()
require.NoError(t, err)
require.Zero(t, pausedWorkers())
}
2018-05-30 18:05:15 +00:00
// Test doing an inplace upgrade on a server from raft protocol 2 to 3
// This verifies that removing the server and adding it back with a uuid works
// even if the server's address stays the same.
func TestServer_ReconcileMember(t *testing.T) {
t.Parallel()
// Create a three node cluster
s1, cleanupS1 := TestServer(t, func(c *Config) {
c.BootstrapExpect = 2
c.RaftConfig.ProtocolVersion = 3
})
defer cleanupS1()
s2, cleanupS2 := TestServer(t, func(c *Config) {
c.BootstrapExpect = 2
c.RaftConfig.ProtocolVersion = 3
})
defer cleanupS2()
TestJoin(t, s1, s2)
testutil.WaitForLeader(t, s1.RPC)
// test relies on s3 not being the leader, so adding it
// after leadership has been established to reduce
s3, cleanupS3 := TestServer(t, func(c *Config) {
Simplify Bootstrap logic in tests This change updates tests to honor `BootstrapExpect` exclusively when forming test clusters and removes test only knobs, e.g. `config.DevDisableBootstrap`. Background: Test cluster creation is fragile. Test servers don't follow the BootstapExpected route like production clusters. Instead they start as single node clusters and then get rejoin and may risk causing brain split or other test flakiness. The test framework expose few knobs to control those (e.g. `config.DevDisableBootstrap` and `config.Bootstrap`) that control whether a server should bootstrap the cluster. These flags are confusing and it's unclear when to use: their usage in multi-node cluster isn't properly documented. Furthermore, they have some bad side-effects as they don't control Raft library: If `config.DevDisableBootstrap` is true, the test server may not immediately attempt to bootstrap a cluster, but after an election timeout (~50ms), Raft may force a leadership election and win it (with only one vote) and cause a split brain. The knobs are also confusing as Bootstrap is an overloaded term. In BootstrapExpect, we refer to bootstrapping the cluster only after N servers are connected. But in tests and the knobs above, it refers to whether the server is a single node cluster and shouldn't wait for any other server. Changes: This commit makes two changes: First, it relies on `BootstrapExpected` instead of `Bootstrap` and/or `DevMode` flags. This change is relatively trivial. Introduce a `Bootstrapped` flag to track if the cluster is bootstrapped. This allows us to keep `BootstrapExpected` immutable. Previously, the flag was a config value but it gets set to 0 after cluster bootstrap completes.
2020-03-02 15:29:24 +00:00
c.BootstrapExpect = 0
c.RaftConfig.ProtocolVersion = 2
})
defer cleanupS3()
TestJoin(t, s1, s3)
// Create a memberlist object for s3, with raft protocol upgraded to 3
upgradedS3Member := serf.Member{
Name: s3.config.NodeName,
Addr: s3.config.RPCAddr.IP,
Status: serf.StatusAlive,
Tags: make(map[string]string),
}
upgradedS3Member.Tags["role"] = "nomad"
upgradedS3Member.Tags["id"] = s3.config.NodeID
upgradedS3Member.Tags["region"] = s3.config.Region
upgradedS3Member.Tags["dc"] = s3.config.Datacenter
upgradedS3Member.Tags["rpc_addr"] = "127.0.0.1"
upgradedS3Member.Tags["port"] = strconv.Itoa(s3.config.RPCAddr.Port)
upgradedS3Member.Tags["build"] = "0.8.0"
upgradedS3Member.Tags["vsn"] = "2"
upgradedS3Member.Tags["mvn"] = "1"
upgradedS3Member.Tags["raft_vsn"] = "3"
findLeader := func(t *testing.T) *Server {
t.Helper()
for _, s := range []*Server{s1, s2, s3} {
if s.IsLeader() {
t.Logf("found leader: %v %v", s.config.NodeID, s.config.RPCAddr)
return s
}
}
t.Fatalf("no leader found")
return nil
}
// Find the leader so that we can call reconcile member on it
leader := findLeader(t)
if err := leader.reconcileMember(upgradedS3Member); err != nil {
t.Fatalf("failed to reconcile member: %v", err)
}
// This should remove s3 from the config and potentially cause a leader election
testutil.WaitForLeader(t, s1.RPC)
// Figure out the new leader and call reconcile again, this should add s3 with the new ID format
leader = findLeader(t)
if err := leader.reconcileMember(upgradedS3Member); err != nil {
t.Fatalf("failed to reconcile member: %v", err)
}
testutil.WaitForLeader(t, s1.RPC)
future := s2.raft.GetConfiguration()
if err := future.Error(); err != nil {
t.Fatal(err)
}
addrs := 0
ids := 0
for _, server := range future.Configuration().Servers {
if string(server.ID) == string(server.Address) {
addrs++
} else {
ids++
}
}
// After this, all three servers should have IDs in raft
if got, want := addrs, 0; got != want {
t.Fatalf("got %d server addresses want %d", got, want)
}
if got, want := ids, 3; got != want {
t.Fatalf("got %d server ids want %d: %#v", got, want, future.Configuration().Servers)
}
}
2020-10-21 04:16:25 +00:00
func TestLeader_ReplicateNamespaces(t *testing.T) {
t.Parallel()
assert := assert.New(t)
s1, root, cleanupS1 := TestACLServer(t, func(c *Config) {
c.Region = "region1"
c.AuthoritativeRegion = "region1"
c.ACLEnabled = true
})
defer cleanupS1()
s2, _, cleanupS2 := TestACLServer(t, func(c *Config) {
c.Region = "region2"
c.AuthoritativeRegion = "region1"
c.ACLEnabled = true
c.ReplicationBackoff = 20 * time.Millisecond
c.ReplicationToken = root.SecretID
})
defer cleanupS2()
TestJoin(t, s1, s2)
testutil.WaitForLeader(t, s1.RPC)
testutil.WaitForLeader(t, s2.RPC)
// Write a namespace to the authoritative region
ns1 := mock.Namespace()
assert.Nil(s1.State().UpsertNamespaces(100, []*structs.Namespace{ns1}))
// Wait for the namespace to replicate
testutil.WaitForResult(func() (bool, error) {
state := s2.State()
out, err := state.NamespaceByName(nil, ns1.Name)
return out != nil, err
}, func(err error) {
t.Fatalf("should replicate namespace")
})
// Delete the namespace at the authoritative region
assert.Nil(s1.State().DeleteNamespaces(200, []string{ns1.Name}))
// Wait for the namespace deletion to replicate
testutil.WaitForResult(func() (bool, error) {
state := s2.State()
out, err := state.NamespaceByName(nil, ns1.Name)
return out == nil, err
}, func(err error) {
t.Fatalf("should replicate namespace deletion")
})
}
func TestLeader_DiffNamespaces(t *testing.T) {
t.Parallel()
state := state.TestStateStore(t)
// Populate the local state
ns1 := mock.Namespace()
ns2 := mock.Namespace()
ns3 := mock.Namespace()
assert.Nil(t, state.UpsertNamespaces(100, []*structs.Namespace{ns1, ns2, ns3}))
// Simulate a remote list
rns2 := ns2.Copy()
rns2.ModifyIndex = 50 // Ignored, same index
rns3 := ns3.Copy()
rns3.ModifyIndex = 100 // Updated, higher index
rns3.Hash = []byte{0, 1, 2, 3}
ns4 := mock.Namespace()
remoteList := []*structs.Namespace{
rns2,
rns3,
ns4,
}
delete, update := diffNamespaces(state, 50, remoteList)
sort.Strings(delete)
// ns1 does not exist on the remote side, should delete
assert.Equal(t, []string{structs.DefaultNamespace, ns1.Name}, delete)
// ns2 is un-modified - ignore. ns3 modified, ns4 new.
assert.Equal(t, []string{ns3.Name, ns4.Name}, update)
}
// waitForStableLeadership waits until a leader is elected and all servers
// get promoted as voting members, returns the leader
func waitForStableLeadership(t *testing.T, servers []*Server) *Server {
nPeers := len(servers)
// wait for all servers to discover each other
for _, s := range servers {
testutil.WaitForResult(func() (bool, error) {
peers, _ := s.numPeers()
if peers != nPeers {
return false, fmt.Errorf("should find %d peers but found %d", nPeers, peers)
}
return true, nil
}, func(err error) {
require.NoError(t, err)
})
}
// wait for leader
var leader *Server
testutil.WaitForResult(func() (bool, error) {
for _, s := range servers {
if s.IsLeader() {
leader = s
return true, nil
}
}
return false, fmt.Errorf("no leader found")
}, func(err error) {
require.NoError(t, err)
})
// wait for all servers get marked as voters
testutil.WaitForResult(func() (bool, error) {
future := leader.raft.GetConfiguration()
if err := future.Error(); err != nil {
return false, fmt.Errorf("failed to get raft config: %v", future.Error())
}
ss := future.Configuration().Servers
if len(ss) != len(servers) {
return false, fmt.Errorf("raft doesn't contain all servers. Expected %d but found %d", len(servers), len(ss))
}
for _, s := range ss {
if s.Suffrage != raft.Voter {
return false, fmt.Errorf("configuration has non voting server: %v", s)
}
}
return true, nil
}, func(err error) {
require.NoError(t, err)
})
return leader
}