Skip to content

Commit 8c15192

Browse files
ethanndicksonethanndickson
authored
feat(cli): add p2p diagnostics to ping (coder#14426)
First PR to address coder#14244. Adds common potential reasons as to why a direct connection to the workspace agent couldn't be established to `coder ping`: - If the Coder deployment administrator has blocked direction connections (`CODER_BLOCK_DIRECT`). - If the client has no STUN servers within it's DERP map. - If the client or agent appears to be behind a hard NAT, as per Tailscale `netInfo.MappingVariesByDestIP` Also adds a warning if the client or agent has a network interface below the 'safe' MTU for tailnet. This warning is always displayed at the end of a `coder ping`.
1 parent b36d979 commit 8c15192

File tree

8 files changed

+298
-5
lines changed

8 files changed

+298
-5
lines changed

‎agent/api.go

+1
Original file line numberDiff line numberDiff line change
@@ -37,6 +37,7 @@ func (a *agent) apiHandler() http.Handler {
3737
}
3838
promHandler := PrometheusMetricsHandler(a.prometheusRegistry, a.logger)
3939
r.Get("/api/v0/listening-ports", lp.handler)
40+
r.Get("/api/v0/netcheck", a.HandleNetcheck)
4041
r.Get("/debug/logs", a.HandleHTTPDebugLogs)
4142
r.Get("/debug/magicsock", a.HandleHTTPDebugMagicsock)
4243
r.Get("/debug/magicsock/debug-logging/{state}", a.HandleHTTPMagicsockDebugLoggingState)

‎agent/health.go

+31
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,31 @@
1+
package agent
2+
3+
import (
4+
"net/http"
5+
6+
"github.com/coder/coder/v2/coderd/healthcheck/health"
7+
"github.com/coder/coder/v2/coderd/httpapi"
8+
"github.com/coder/coder/v2/codersdk"
9+
"github.com/coder/coder/v2/codersdk/healthsdk"
10+
)
11+
12+
func (a *agent) HandleNetcheck(rw http.ResponseWriter, r *http.Request) {
13+
ni := a.TailnetConn().GetNetInfo()
14+
15+
ifReport, err := healthsdk.RunInterfacesReport()
16+
if err != nil {
17+
httpapi.Write(r.Context(), rw, http.StatusInternalServerError, codersdk.Response{
18+
Message: "Failed to run interfaces report",
19+
Detail: err.Error(),
20+
})
21+
return
22+
}
23+
24+
httpapi.Write(r.Context(), rw, http.StatusOK, healthsdk.AgentNetcheckReport{
25+
BaseReport: healthsdk.BaseReport{
26+
Severity: health.SeverityOK,
27+
},
28+
NetInfo: ni,
29+
Interfaces: ifReport,
30+
})
31+
}

‎cli/cliui/agent.go

+55
Original file line numberDiff line numberDiff line change
@@ -10,8 +10,11 @@ import (
1010

1111
"github.com/google/uuid"
1212
"golang.org/x/xerrors"
13+
"tailscale.com/tailcfg"
1314

1415
"github.com/coder/coder/v2/codersdk"
16+
"github.com/coder/coder/v2/codersdk/healthsdk"
17+
"github.com/coder/coder/v2/codersdk/workspacesdk"
1518
"github.com/coder/coder/v2/tailnet"
1619
)
1720

@@ -346,3 +349,55 @@ func PeerDiagnostics(w io.Writer, d tailnet.PeerDiagnostics) {
346349
_, _ = fmt.Fprint(w, "✘ Wireguard is not connected\n")
347350
}
348351
}
352+
353+
type ConnDiags struct {
354+
ConnInfo *workspacesdk.AgentConnectionInfo
355+
PingP2P bool
356+
DisableDirect bool
357+
LocalNetInfo *tailcfg.NetInfo
358+
LocalInterfaces *healthsdk.InterfacesReport
359+
AgentNetcheck *healthsdk.AgentNetcheckReport
360+
// TODO: More diagnostics
361+
}
362+
363+
func ConnDiagnostics(w io.Writer, d ConnDiags) {
364+
if d.AgentNetcheck != nil {
365+
for _, msg := range d.AgentNetcheck.Interfaces.Warnings {
366+
_, _ = fmt.Fprintf(w, "❗ Agent: %s\n", msg.Message)
367+
}
368+
}
369+
370+
if d.LocalInterfaces != nil {
371+
for _, msg := range d.LocalInterfaces.Warnings {
372+
_, _ = fmt.Fprintf(w, "❗ Client: %s\n", msg.Message)
373+
}
374+
}
375+
376+
if d.PingP2P {
377+
_, _ = fmt.Fprint(w, "✔ You are connected directly (p2p)\n")
378+
return
379+
}
380+
_, _ = fmt.Fprint(w, "❗ You are connected via a DERP relay, not directly (p2p)\n")
381+
382+
if d.DisableDirect {
383+
_, _ = fmt.Fprint(w, "❗ Direct connections are disabled locally, by `--disable-direct` or `CODER_DISABLE_DIRECT`\n")
384+
return
385+
}
386+
387+
if d.ConnInfo != nil && d.ConnInfo.DisableDirectConnections {
388+
_, _ = fmt.Fprint(w, "❗ Your Coder administrator has blocked direct connections\n")
389+
return
390+
}
391+
392+
if d.ConnInfo != nil && d.ConnInfo.DERPMap != nil && !d.ConnInfo.DERPMap.HasSTUN() {
393+
_, _ = fmt.Fprint(w, "✘ The DERP map is not configured to use STUN, which will prevent direct connections from starting outside of local networks\n")
394+
}
395+
396+
if d.LocalNetInfo != nil && d.LocalNetInfo.MappingVariesByDestIP.EqualBool(true) {
397+
_, _ = fmt.Fprint(w, "❗ Client is potentially behind a hard NAT, as multiple endpoints were retrieved from different STUN servers\n")
398+
}
399+
400+
if d.AgentNetcheck != nil && d.AgentNetcheck.NetInfo != nil && d.AgentNetcheck.NetInfo.MappingVariesByDestIP.EqualBool(true) {
401+
_, _ = fmt.Fprint(w, "❗ Agent is potentially behind a hard NAT, as multiple endpoints were retrieved from different STUN servers\n")
402+
}
403+
}

‎cli/cliui/agent_test.go

+129
Original file line numberDiff line numberDiff line change
@@ -20,8 +20,11 @@ import (
2020

2121
"github.com/coder/coder/v2/cli/clitest"
2222
"github.com/coder/coder/v2/cli/cliui"
23+
"github.com/coder/coder/v2/coderd/healthcheck/health"
2324
"github.com/coder/coder/v2/coderd/util/ptr"
2425
"github.com/coder/coder/v2/codersdk"
26+
"github.com/coder/coder/v2/codersdk/healthsdk"
27+
"github.com/coder/coder/v2/codersdk/workspacesdk"
2528
"github.com/coder/coder/v2/tailnet"
2629
"github.com/coder/coder/v2/testutil"
2730
"github.com/coder/serpent"
@@ -672,3 +675,129 @@ func TestPeerDiagnostics(t *testing.T) {
672675
})
673676
}
674677
}
678+
679+
func TestConnDiagnostics(t *testing.T) {
680+
t.Parallel()
681+
testCases := []struct {
682+
name string
683+
diags cliui.ConnDiags
684+
want []string
685+
}{
686+
{
687+
name: "Direct",
688+
diags: cliui.ConnDiags{
689+
ConnInfo: &workspacesdk.AgentConnectionInfo{},
690+
PingP2P: true,
691+
LocalNetInfo: &tailcfg.NetInfo{},
692+
},
693+
want: []string{
694+
`✔ You are connected directly (p2p)`,
695+
},
696+
},
697+
{
698+
name: "DirectBlocked",
699+
diags: cliui.ConnDiags{
700+
ConnInfo: &workspacesdk.AgentConnectionInfo{
701+
DisableDirectConnections: true,
702+
},
703+
},
704+
want: []string{
705+
`❗ You are connected via a DERP relay, not directly (p2p)`,
706+
`❗ Your Coder administrator has blocked direct connections`,
707+
},
708+
},
709+
{
710+
name: "NoStun",
711+
diags: cliui.ConnDiags{
712+
ConnInfo: &workspacesdk.AgentConnectionInfo{
713+
DERPMap: &tailcfg.DERPMap{},
714+
},
715+
LocalNetInfo: &tailcfg.NetInfo{},
716+
},
717+
want: []string{
718+
`❗ You are connected via a DERP relay, not directly (p2p)`,
719+
`✘ The DERP map is not configured to use STUN, which will prevent direct connections from starting outside of local networks`,
720+
},
721+
},
722+
{
723+
name: "ClientHardNat",
724+
diags: cliui.ConnDiags{
725+
LocalNetInfo: &tailcfg.NetInfo{
726+
MappingVariesByDestIP: "true",
727+
},
728+
},
729+
want: []string{
730+
`❗ You are connected via a DERP relay, not directly (p2p)`,
731+
`❗ Client is potentially behind a hard NAT, as multiple endpoints were retrieved from different STUN servers`,
732+
},
733+
},
734+
{
735+
name: "AgentHardNat",
736+
diags: cliui.ConnDiags{
737+
ConnInfo: &workspacesdk.AgentConnectionInfo{},
738+
PingP2P: false,
739+
LocalNetInfo: &tailcfg.NetInfo{},
740+
AgentNetcheck: &healthsdk.AgentNetcheckReport{
741+
NetInfo: &tailcfg.NetInfo{MappingVariesByDestIP: "true"},
742+
},
743+
},
744+
want: []string{
745+
`❗ You are connected via a DERP relay, not directly (p2p)`,
746+
`❗ Agent is potentially behind a hard NAT, as multiple endpoints were retrieved from different STUN servers`,
747+
},
748+
},
749+
{
750+
name: "AgentInterfaceWarnings",
751+
diags: cliui.ConnDiags{
752+
PingP2P: true,
753+
AgentNetcheck: &healthsdk.AgentNetcheckReport{
754+
Interfaces: healthsdk.InterfacesReport{
755+
BaseReport: healthsdk.BaseReport{
756+
Warnings: []health.Message{
757+
health.Messagef(health.CodeInterfaceSmallMTU, "network interface eth0 has MTU 1280, (less than 1378), which may cause problems with direct connections"),
758+
},
759+
},
760+
},
761+
},
762+
},
763+
want: []string{
764+
`❗ Agent: network interface eth0 has MTU 1280, (less than 1378), which may cause problems with direct connections`,
765+
`✔ You are connected directly (p2p)`,
766+
},
767+
},
768+
{
769+
name: "LocalInterfaceWarnings",
770+
diags: cliui.ConnDiags{
771+
PingP2P: true,
772+
LocalInterfaces: &healthsdk.InterfacesReport{
773+
BaseReport: healthsdk.BaseReport{
774+
Warnings: []health.Message{
775+
health.Messagef(health.CodeInterfaceSmallMTU, "network interface eth1 has MTU 1310, (less than 1378), which may cause problems with direct connections"),
776+
},
777+
},
778+
},
779+
},
780+
want: []string{
781+
`❗ Client: network interface eth1 has MTU 1310, (less than 1378), which may cause problems with direct connections`,
782+
`✔ You are connected directly (p2p)`,
783+
},
784+
},
785+
}
786+
for _, tc := range testCases {
787+
tc := tc
788+
t.Run(tc.name, func(t *testing.T) {
789+
t.Parallel()
790+
r, w := io.Pipe()
791+
go func() {
792+
defer w.Close()
793+
cliui.ConnDiagnostics(w, tc.diags)
794+
}()
795+
bytes, err := io.ReadAll(r)
796+
require.NoError(t, err)
797+
output := string(bytes)
798+
for _, want := range tc.want {
799+
require.Contains(t, output, want)
800+
}
801+
})
802+
}
803+
}

‎cli/ping.go

+41-4
Original file line numberDiff line numberDiff line change
@@ -2,7 +2,9 @@ package cli
22

33
import (
44
"context"
5+
"errors"
56
"fmt"
7+
"net/http"
68
"time"
79

810
"golang.org/x/xerrors"
@@ -14,6 +16,7 @@ import (
1416

1517
"github.com/coder/coder/v2/cli/cliui"
1618
"github.com/coder/coder/v2/codersdk"
19+
"github.com/coder/coder/v2/codersdk/healthsdk"
1720
"github.com/coder/coder/v2/codersdk/workspacesdk"
1821
"github.com/coder/serpent"
1922
)
@@ -61,7 +64,8 @@ func (r *RootCmd) ping() *serpent.Command {
6164
if !r.disableNetworkTelemetry {
6265
opts.EnableTelemetry = true
6366
}
64-
conn, err := workspacesdk.New(client).DialAgent(ctx, workspaceAgent.ID, opts)
67+
client := workspacesdk.New(client)
68+
conn, err := client.DialAgent(ctx, workspaceAgent.ID, opts)
6569
if err != nil {
6670
return err
6771
}
@@ -138,11 +142,44 @@ func (r *RootCmd) ping() *serpent.Command {
138142
)
139143

140144
if n == int(pingNum) {
141-
diags := conn.GetPeerDiagnostics()
142-
cliui.PeerDiagnostics(inv.Stdout, diags)
143-
return nil
145+
break
144146
}
145147
}
148+
ctx, cancel = context.WithTimeout(inv.Context(), 30*time.Second)
149+
defer cancel()
150+
diags := conn.GetPeerDiagnostics()
151+
cliui.PeerDiagnostics(inv.Stdout, diags)
152+
153+
connDiags := cliui.ConnDiags{
154+
PingP2P: didP2p,
155+
DisableDirect: r.disableDirect,
156+
LocalNetInfo: conn.GetNetInfo(),
157+
}
158+
connInfo, err := client.AgentConnectionInfoGeneric(ctx)
159+
if err == nil {
160+
connDiags.ConnInfo = &connInfo
161+
} else {
162+
_, _ = fmt.Fprintf(inv.Stdout, "Failed to retrieve connection info from server: %v\n", err)
163+
}
164+
ifReport, err := healthsdk.RunInterfacesReport()
165+
if err == nil {
166+
connDiags.LocalInterfaces = &ifReport
167+
} else {
168+
_, _ = fmt.Fprintf(inv.Stdout, "Failed to retrieve local interfaces report: %v\n", err)
169+
}
170+
agentNetcheck, err := conn.Netcheck(ctx)
171+
if err == nil {
172+
connDiags.AgentNetcheck = &agentNetcheck
173+
} else {
174+
var sdkErr *codersdk.Error
175+
if errors.As(err, &sdkErr) && sdkErr.StatusCode() == http.StatusNotFound {
176+
_, _ = fmt.Fprint(inv.Stdout, "Could not generate full connection report as the workspace agent is outdated\n")
177+
} else {
178+
_, _ = fmt.Fprintf(inv.Stdout, "Failed to retrieve connection report from agent: %v\n", err)
179+
}
180+
}
181+
cliui.ConnDiagnostics(inv.Stdout, connDiags)
182+
return nil
146183
},
147184
}
148185

‎codersdk/healthsdk/healthsdk.go

+7
Original file line numberDiff line numberDiff line change
@@ -273,3 +273,10 @@ type ClientNetcheckReport struct {
273273
DERP DERPHealthReport `json:"derp"`
274274
Interfaces InterfacesReport `json:"interfaces"`
275275
}
276+
277+
// @typescript-ignore AgentNetcheckReport
278+
type AgentNetcheckReport struct {
279+
BaseReport
280+
NetInfo *tailcfg.NetInfo `json:"net_info"`
281+
Interfaces InterfacesReport `json:"interfaces"`
282+
}

‎codersdk/workspacesdk/agentconn.go

+18
Original file line numberDiff line numberDiff line change
@@ -22,6 +22,7 @@ import (
2222

2323
"github.com/coder/coder/v2/coderd/tracing"
2424
"github.com/coder/coder/v2/codersdk"
25+
"github.com/coder/coder/v2/codersdk/healthsdk"
2526
"github.com/coder/coder/v2/tailnet"
2627
)
2728

@@ -241,6 +242,23 @@ func (c *AgentConn) ListeningPorts(ctx context.Context) (codersdk.WorkspaceAgent
241242
return resp, json.NewDecoder(res.Body).Decode(&resp)
242243
}
243244

245+
// Netcheck returns a network check report from the workspace agent.
246+
func (c *AgentConn) Netcheck(ctx context.Context) (healthsdk.AgentNetcheckReport, error) {
247+
ctx, span := tracing.StartSpan(ctx)
248+
defer span.End()
249+
res, err := c.apiRequest(ctx, http.MethodGet, "/api/v0/netcheck", nil)
250+
if err != nil {
251+
return healthsdk.AgentNetcheckReport{}, xerrors.Errorf("do request: %w", err)
252+
}
253+
defer res.Body.Close()
254+
if res.StatusCode != http.StatusOK {
255+
return healthsdk.AgentNetcheckReport{}, codersdk.ReadBodyAsError(res)
256+
}
257+
258+
var resp healthsdk.AgentNetcheckReport
259+
return resp, json.NewDecoder(res.Body).Decode(&resp)
260+
}
261+
244262
// DebugMagicsock makes a request to the workspace agent's magicsock debug endpoint.
245263
func (c *AgentConn) DebugMagicsock(ctx context.Context) ([]byte, error) {
246264
ctx, span := tracing.StartSpan(ctx)

0 commit comments

Comments
 (0)