eyjhb/headscale
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 1

Replace database locks with transactions (#1701)

Browse source 
This commits removes the locks used to guard data integrity for the
database and replaces them with Transactions, turns out that SQL had
a way to deal with this all along.

This reduces the complexity we had with multiple locks that might stack
or recurse (database, nofitifer, mapper). All notifications and state
updates are now triggered _after_ a database change.


Signed-off-by: Kristoffer Dalby <kristoffer@tailscale.com>
This commit is contained in:
Kristoffer Dalby 2024-02-08 17:28:19 +01:00 • committed by GitHub
parent cbf57e27a7
commit 83769ba715
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
32 changed files with 1496 additions and 1128 deletions
Download patch file Download diff file Expand all files Collapse all files

4
integration/cli_test.go
View file

@ -1240,7 +1240,7 @@ func TestNodeRenameCommand(t *testing.T) {
assert.Contains(t, listAll[4].GetGivenName(), "node-5")
for idx := 0; idx < 3; idx++ {
_, err := headscale.Execute(
res, err := headscale.Execute(
[]string{
"headscale",
"nodes",
@ -1251,6 +1251,8 @@ func TestNodeRenameCommand(t *testing.T) {
},
)
assert.Nil(t, err)
assert.Contains(t, res, "Node renamed")
}
var listAllAfterRename []v1.Node

126
integration/general_test.go
View file

@ -26,6 +26,8 @@ func TestPingAllByIP(t *testing.T) {
assertNoErr(t, err)
defer scenario.Shutdown()
// TODO(kradalby): it does not look like the user thing works, only second
// get created? maybe only when many?
spec := map[string]int{
"user1": len(MustTestVersions),
"user2": len(MustTestVersions),
@ -321,7 +323,12 @@ func TestTaildrop(t *testing.T) {
t.Fatalf("failed to install curl on %s, err: %s", client.Hostname(), err)
}
}
curlCommand := []string{"curl", "--unix-socket", "/var/run/tailscale/tailscaled.sock", "http://local-tailscaled.sock/localapi/v0/file-targets"}
curlCommand := []string{
"curl",
"--unix-socket",
"/var/run/tailscale/tailscaled.sock",
"http://local-tailscaled.sock/localapi/v0/file-targets",
}
err = retry(10, 1*time.Second, func() error {
result, _, err := client.Execute(curlCommand)
if err != nil {
@ -338,13 +345,23 @@ func TestTaildrop(t *testing.T) {
for _, ft := range fts {
ftStr += fmt.Sprintf("\t%s\n", ft.Node.Name)
}
return fmt.Errorf("client %s does not have all its peers as FileTargets, got %d, want: %d\n%s", client.Hostname(), len(fts), len(allClients)-1, ftStr)
return fmt.Errorf(
"client %s does not have all its peers as FileTargets, got %d, want: %d\n%s",
client.Hostname(),
len(fts),
len(allClients)-1,
ftStr,
)
}
return err
})
if err != nil {
t.Errorf("failed to query localapi for filetarget on %s, err: %s", client.Hostname(), err)
t.Errorf(
"failed to query localapi for filetarget on %s, err: %s",
client.Hostname(),
err,
)
}
}
@ -434,72 +451,6 @@ func TestTaildrop(t *testing.T) {
}
}
func TestResolveMagicDNS(t *testing.T) {
IntegrationSkip(t)
t.Parallel()
scenario, err := NewScenario()
assertNoErr(t, err)
defer scenario.Shutdown()
spec := map[string]int{
"magicdns1": len(MustTestVersions),
"magicdns2": len(MustTestVersions),
}
err = scenario.CreateHeadscaleEnv(spec, []tsic.Option{}, hsic.WithTestName("magicdns"))
assertNoErrHeadscaleEnv(t, err)
allClients, err := scenario.ListTailscaleClients()
assertNoErrListClients(t, err)
err = scenario.WaitForTailscaleSync()
assertNoErrSync(t, err)
// Poor mans cache
_, err = scenario.ListTailscaleClientsFQDNs()
assertNoErrListFQDN(t, err)
_, err = scenario.ListTailscaleClientsIPs()
assertNoErrListClientIPs(t, err)
for _, client := range allClients {
for _, peer := range allClients {
// It is safe to ignore this error as we handled it when caching it
peerFQDN, _ := peer.FQDN()
command := []string{
"tailscale",
"ip", peerFQDN,
}
result, _, err := client.Execute(command)
if err != nil {
t.Fatalf(
"failed to execute resolve/ip command %s from %s: %s",
peerFQDN,
client.Hostname(),
err,
)
}
ips, err := peer.IPs()
if err != nil {
t.Fatalf(
"failed to get ips for %s: %s",
peer.Hostname(),
err,
)
}
for _, ip := range ips {
if !strings.Contains(result, ip.String()) {
t.Fatalf("ip %s is not found in \n%s\n", ip.String(), result)
}
}
}
}
}
func TestExpireNode(t *testing.T) {
IntegrationSkip(t)
t.Parallel()
@ -545,7 +496,7 @@ func TestExpireNode(t *testing.T) {
// TODO(kradalby): This is Headscale specific and would not play nicely
// with other implementations of the ControlServer interface
result, err := headscale.Execute([]string{
"headscale", "nodes", "expire", "--identifier", "0", "--output", "json",
"headscale", "nodes", "expire", "--identifier", "1", "--output", "json",
})
assertNoErr(t, err)
@ -576,16 +527,38 @@ func TestExpireNode(t *testing.T) {
assertNotNil(t, peerStatus.Expired)
assert.NotNil(t, peerStatus.KeyExpiry)
t.Logf("node %q should have a key expire before %s, was %s", peerStatus.HostName, now.String(), peerStatus.KeyExpiry)
t.Logf(
"node %q should have a key expire before %s, was %s",
peerStatus.HostName,
now.String(),
peerStatus.KeyExpiry,
)
if peerStatus.KeyExpiry != nil {
assert.Truef(t, peerStatus.KeyExpiry.Before(now), "node %q should have a key expire before %s, was %s", peerStatus.HostName, now.String(), peerStatus.KeyExpiry)
assert.Truef(
t,
peerStatus.KeyExpiry.Before(now),
"node %q should have a key expire before %s, was %s",
peerStatus.HostName,
now.String(),
peerStatus.KeyExpiry,
)
}
assert.Truef(t, peerStatus.Expired, "node %q should be expired, expired is %v", peerStatus.HostName, peerStatus.Expired)
assert.Truef(
t,
peerStatus.Expired,
"node %q should be expired, expired is %v",
peerStatus.HostName,
peerStatus.Expired,
)
_, stderr, _ := client.Execute([]string{"tailscale", "ping", node.GetName()})
if !strings.Contains(stderr, "node key has expired") {
t.Errorf("expected to be unable to ping expired host %q from %q", node.GetName(), client.Hostname())
t.Errorf(
"expected to be unable to ping expired host %q from %q",
node.GetName(),
client.Hostname(),
)
}
} else {
t.Errorf("failed to find node %q with nodekey (%s) in mapresponse, should be present even if it is expired", node.GetName(), expiredNodeKey)
@ -597,7 +570,7 @@ func TestExpireNode(t *testing.T) {
// NeedsLogin means that the node has understood that it is no longer
// valid.
assert.Equal(t, "NeedsLogin", status.BackendState)
assert.Equalf(t, "NeedsLogin", status.BackendState, "checking node %q", status.Self.HostName)
}
}
}
@ -690,7 +663,8 @@ func TestNodeOnlineLastSeenStatus(t *testing.T) {
assert.Truef(
t,
lastSeen.After(lastSeenThreshold),
"lastSeen (%v) was not %s after the threshold (%v)",
"node (%s) lastSeen (%v) was not %s after the threshold (%v)",
node.GetName(),
lastSeen,
keepAliveInterval,
lastSeenThreshold,

32
integration/route_test.go
View file

@ -88,9 +88,9 @@ func TestEnablingRoutes(t *testing.T) {
assert.Len(t, routes, 3)
for _, route := range routes {
assert.Equal(t, route.GetAdvertised(), true)
assert.Equal(t, route.GetEnabled(), false)
assert.Equal(t, route.GetIsPrimary(), false)
assert.Equal(t, true, route.GetAdvertised())
assert.Equal(t, false, route.GetEnabled())
assert.Equal(t, false, route.GetIsPrimary())
}
// Verify that no routes has been sent to the client,
@ -135,9 +135,9 @@ func TestEnablingRoutes(t *testing.T) {
assert.Len(t, enablingRoutes, 3)
for _, route := range enablingRoutes {
assert.Equal(t, route.GetAdvertised(), true)
assert.Equal(t, route.GetEnabled(), true)
assert.Equal(t, route.GetIsPrimary(), true)
assert.Equal(t, true, route.GetAdvertised())
assert.Equal(t, true, route.GetEnabled())
assert.Equal(t, true, route.GetIsPrimary())
}
time.Sleep(5 * time.Second)
@ -191,6 +191,8 @@ func TestEnablingRoutes(t *testing.T) {
})
assertNoErr(t, err)
time.Sleep(5 * time.Second)
var disablingRoutes []*v1.Route
err = executeAndUnmarshal(
headscale,
@ -209,16 +211,14 @@ func TestEnablingRoutes(t *testing.T) {
assert.Equal(t, true, route.GetAdvertised())
if route.GetId() == routeToBeDisabled.GetId() {
assert.Equal(t, route.GetEnabled(), false)
assert.Equal(t, route.GetIsPrimary(), false)
assert.Equal(t, false, route.GetEnabled())
assert.Equal(t, false, route.GetIsPrimary())
} else {
assert.Equal(t, route.GetEnabled(), true)
assert.Equal(t, route.GetIsPrimary(), true)
assert.Equal(t, true, route.GetEnabled())
assert.Equal(t, true, route.GetIsPrimary())
}
}
time.Sleep(5 * time.Second)
// Verify that the clients can see the new routes
for _, client := range allClients {
status, err := client.Status()
@ -294,7 +294,7 @@ func TestHASubnetRouterFailover(t *testing.T) {
// advertise HA route on node 1 and 2
// ID 1 will be primary
// ID 2 will be secondary
for _, client := range allClients {
for _, client := range allClients[:2] {
status, err := client.Status()
assertNoErr(t, err)
@ -306,6 +306,8 @@ func TestHASubnetRouterFailover(t *testing.T) {
}
_, _, err = client.Execute(command)
assertNoErrf(t, "failed to advertise route: %s", err)
} else {
t.Fatalf("failed to find route for Node %s (id: %s)", status.Self.HostName, status.Self.ID)
}
}
@ -328,6 +330,8 @@ func TestHASubnetRouterFailover(t *testing.T) {
assertNoErr(t, err)
assert.Len(t, routes, 2)
t.Logf("initial routes %#v", routes)
for _, route := range routes {
assert.Equal(t, true, route.GetAdvertised())
assert.Equal(t, false, route.GetEnabled())
@ -644,6 +648,8 @@ func TestHASubnetRouterFailover(t *testing.T) {
assertNoErr(t, err)
assert.Len(t, routesAfterDisabling1, 2)
t.Logf("routes after disabling1 %#v", routesAfterDisabling1)
// Node 1 is not primary
assert.Equal(t, true, routesAfterDisabling1[0].GetAdvertised())
assert.Equal(t, false, routesAfterDisabling1[0].GetEnabled())

4
integration/scenario.go
View file

@ -56,8 +56,8 @@ var (
"1.44": true, // CapVer: 63
"1.42": true, // CapVer: 61
"1.40": true, // CapVer: 61
"1.38": true, // CapVer: 58
"1.36": true, // Oldest supported version, CapVer: 56
"1.38": true, // Oldest supported version, CapVer: 58
"1.36": false, // CapVer: 56
"1.34": false, // CapVer: 51
"1.32": false, // CapVer: 46
"1.30": false,