Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

command/operator_debug: add pprof interval #11938

Merged
merged 10 commits into from
Apr 4, 2022
Merged

command/operator_debug: add pprof interval #11938

Show file tree
Hide file tree
Changes from 4 commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
29b4b06
command/operator_debug: add pprof interval
danishprakash Jan 9, 2022
657a041
command/operator_debug: update tests
danishprakash Feb 12, 2022
9123886
command/operator_debug: pass pprofIntervalCount for filename
danishprakash Apr 2, 2022
7401fe5
pad profile filename to match console output
tgross Apr 4, 2022
bc4077d
docs and changelog entry
tgross Apr 4, 2022
bd0d6b0
assert correct file name for Debug_CapturedFiles test
tgross Apr 4, 2022
48d0d78
run periodic pprof collection concurrently
tgross Apr 4, 2022
4d536bc
periodic pprof termination should not end the whole collection
tgross Apr 4, 2022
0a250b1
fixup for sigsegv test
tgross Apr 4, 2022
69a473f
clean up test directories
tgross Apr 4, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
65 changes: 58 additions & 7 deletions command/operator_debug.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,6 +38,7 @@ type OperatorDebugCommand struct {
collectDir string
duration time.Duration
interval time.Duration
pprofInterval time.Duration
pprofDuration time.Duration
logLevel string
maxNodes int
Expand Down Expand Up @@ -166,6 +167,10 @@ Debug Options:
The interval between snapshots of the Nomad state. Set interval equal to
duration to capture a single snapshot. Defaults to 30s.

-pprof-interval=<pprof-interval>
The interval between pprof collections. Set interval equal to
duration to capture a single snapshot. Defaults to 250ms.

-log-level=<level>
The log level to monitor. Defaults to DEBUG.

Expand Down Expand Up @@ -334,7 +339,7 @@ func (c *OperatorDebugCommand) Run(args []string) int {
flags := c.Meta.FlagSet(c.Name(), FlagSetClient)
flags.Usage = func() { c.Ui.Output(c.Help()) }

var duration, interval, output, pprofDuration, eventTopic string
var duration, interval, pprofInterval, output, pprofDuration, eventTopic string
var eventIndex int64
var nodeIDs, serverIDs string
var allowStale bool
Expand All @@ -351,6 +356,7 @@ func (c *OperatorDebugCommand) Run(args []string) int {
flags.BoolVar(&allowStale, "stale", false, "")
flags.StringVar(&output, "output", "", "")
flags.StringVar(&pprofDuration, "pprof-duration", "1s", "")
flags.StringVar(&pprofInterval, "pprof-interval", "250ms", "")
flags.BoolVar(&c.verbose, "verbose", false, "")

c.consul = &external{tls: &api.TLSConfig{}}
Expand Down Expand Up @@ -423,6 +429,20 @@ func (c *OperatorDebugCommand) Run(args []string) int {
}
c.index = uint64(eventIndex)

// Parse the pprof capture interval
pi, err := time.ParseDuration(pprofInterval)
if err != nil {
c.Ui.Error(fmt.Sprintf("Error parsing pprof-interval: %s: %s", pprofInterval, err.Error()))
return 1
}
c.pprofInterval = pi

// Validate interval
if pi.Seconds() > pd.Seconds() {
c.Ui.Error(fmt.Sprintf("pprof-interval %s must be less than pprof-duration %s", pprofInterval, pprofDuration))
return 1
}

// Verify there are no extra arguments
args = flags.Args()
if l := len(args); l != 0 {
Expand Down Expand Up @@ -595,6 +615,7 @@ func (c *OperatorDebugCommand) Run(args []string) int {
}
c.Ui.Output(fmt.Sprintf(" Interval: %s", interval))
c.Ui.Output(fmt.Sprintf(" Duration: %s", duration))
c.Ui.Output(fmt.Sprintf(" Pprof Interval: %s", pprofInterval))
if c.pprofDuration.Seconds() != 1 {
c.Ui.Output(fmt.Sprintf(" pprof Duration: %s", c.pprofDuration))
}
Expand Down Expand Up @@ -663,7 +684,7 @@ func (c *OperatorDebugCommand) collect(client *api.Client) error {
c.collectVault(clusterDir, vaultAddr)

c.collectAgentHosts(client)
c.collectPprofs(client)
c.collectPeriodicPprofs(client)

c.collectPeriodic(client)

Expand Down Expand Up @@ -876,19 +897,48 @@ func (c *OperatorDebugCommand) collectAgentHost(path, id string, client *api.Cli
c.writeJSON(path, "agent-host.json", host, err)
}

func (c *OperatorDebugCommand) collectPeriodicPprofs(client *api.Client) {
tgross marked this conversation as resolved.
Show resolved Hide resolved
duration := time.After(c.duration)
// Create a ticker to execute on every interval ticks
ticker := time.NewTicker(c.pprofInterval)

var pprofIntervalCount int
var name string

// Additionally, an out of loop execute to imitate first tick
c.collectPprofs(client, pprofIntervalCount)

for {
select {
case <-duration:
c.cancel()
return

case <-ticker.C:
name = fmt.Sprintf("%04d", pprofIntervalCount)
c.Ui.Output(fmt.Sprintf(" Capture pprofInterval %s", name))
c.collectPprofs(client, pprofIntervalCount)
pprofIntervalCount++

case <-c.ctx.Done():
return
}
}
}

// collectPprofs captures the /agent/pprof for each listed node
func (c *OperatorDebugCommand) collectPprofs(client *api.Client) {
func (c *OperatorDebugCommand) collectPprofs(client *api.Client, interval int) {
for _, n := range c.nodeIDs {
c.collectPprof(clientDir, n, client)
c.collectPprof(clientDir, n, client, interval)
}

for _, n := range c.serverIDs {
c.collectPprof(serverDir, n, client)
c.collectPprof(serverDir, n, client, interval)
}
}

// collectPprof captures pprof data for the node
func (c *OperatorDebugCommand) collectPprof(path, id string, client *api.Client) {
func (c *OperatorDebugCommand) collectPprof(path, id string, client *api.Client, interval int) {
pprofDurationSeconds := int(c.pprofDuration.Seconds())
opts := api.PprofOptions{Seconds: pprofDurationSeconds}
if path == serverDir {
Expand All @@ -911,7 +961,8 @@ func (c *OperatorDebugCommand) collectPprof(path, id string, client *api.Client)
return // only exit on 403
}
} else {
err := c.writeBytes(path, "profile.prof", bs)
filename := fmt.Sprintf("profile_%04d.prof", interval)
danishprakash marked this conversation as resolved.
Show resolved Hide resolved
err := c.writeBytes(path, filename, bs)
if err != nil {
c.Ui.Error(err.Error())
}
Expand Down
20 changes: 13 additions & 7 deletions command/operator_debug_test.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -211,12 +211,12 @@ func TestDebug_MultiRegion(t *testing.T) {
// Good
{
name: "no region - all servers, all clients",
args: []string{"-address", addrServer1, "-duration", "250ms", "-interval", "250ms", "-server-id", "all", "-node-id", "all", "-pprof-duration", "0"},
args: []string{"-address", addrServer1, "-duration", "250ms", "-interval", "250ms", "-server-id", "all", "-node-id", "all", "-pprof-duration", "1s", "-pprof-interval", "250ms"},
expectedCode: 0,
},
{
name: "region1 - server1 address",
args: []string{"-address", addrServer1, "-region", region1, "-duration", "50ms", "-interval", "50ms", "-server-id", "all", "-node-id", "all", "-pprof-duration", "0"},
args: []string{"-address", addrServer1, "-region", region1, "-duration", "50ms", "-interval", "50ms", "-server-id", "all", "-node-id", "all", "-pprof-duration", "1s", "-pprof-interval", "250ms"},
expectedCode: 0,
expectedOutputs: []string{
"Region: " + region1 + "\n",
Expand All @@ -227,7 +227,7 @@ func TestDebug_MultiRegion(t *testing.T) {
},
{
name: "region1 - client1 address",
args: []string{"-address", addrClient1, "-region", region1, "-duration", "50ms", "-interval", "50ms", "-server-id", "all", "-node-id", "all", "-pprof-duration", "0"},
args: []string{"-address", addrClient1, "-region", region1, "-duration", "50ms", "-interval", "50ms", "-server-id", "all", "-node-id", "all", "-pprof-duration", "1s", "-pprof-interval", "250ms"},
expectedCode: 0,
expectedOutputs: []string{
"Region: " + region1 + "\n",
Expand All @@ -238,7 +238,7 @@ func TestDebug_MultiRegion(t *testing.T) {
},
{
name: "region2 - server2 address",
args: []string{"-address", addrServer2, "-region", region2, "-duration", "50ms", "-interval", "50ms", "-server-id", "all", "-node-id", "all", "-pprof-duration", "0"},
args: []string{"-address", addrServer2, "-region", region2, "-duration", "50ms", "-interval", "50ms", "-server-id", "all", "-node-id", "all", "-pprof-duration", "1s", "-pprof-interval", "250ms"},
expectedCode: 0,
expectedOutputs: []string{
"Region: " + region2 + "\n",
Expand All @@ -249,7 +249,7 @@ func TestDebug_MultiRegion(t *testing.T) {
},
{
name: "region2 - client2 address",
args: []string{"-address", addrClient2, "-region", region2, "-duration", "50ms", "-interval", "50ms", "-server-id", "all", "-node-id", "all", "-pprof-duration", "0"},
args: []string{"-address", addrClient2, "-region", region2, "-duration", "50ms", "-interval", "50ms", "-server-id", "all", "-node-id", "all", "-pprof-duration", "1s", "-pprof-interval", "250ms"},
expectedCode: 0,
expectedOutputs: []string{
"Region: " + region2 + "\n",
Expand All @@ -262,7 +262,7 @@ func TestDebug_MultiRegion(t *testing.T) {
// Bad
{
name: "invalid region - all servers, all clients",
args: []string{"-address", addrServer1, "-region", "never", "-duration", "50ms", "-interval", "50ms", "-server-id", "all", "-node-id", "all", "-pprof-duration", "0"},
args: []string{"-address", addrServer1, "-region", "never", "-duration", "50ms", "-interval", "50ms", "-server-id", "all", "-node-id", "all", "-pprof-duration", "1s", "-pprof-interval", "250ms"},
expectedCode: 1,
expectedError: "500 (No path to region)",
},
Expand Down Expand Up @@ -347,6 +347,11 @@ func TestDebug_Failures(t *testing.T) {
args: []string{"-pprof-duration", "baz"},
expectedCode: 1,
},
{
name: "Fails bad pprof interval",
args: []string{"-pprof-interval", "bar"},
expectedCode: 1,
},
{
name: "Fails bad address",
args: []string{"-address", url + "bogus"},
Expand Down Expand Up @@ -482,7 +487,8 @@ func TestDebug_CapturedFiles(t *testing.T) {
"-node-id", clientID,
"-duration", duration.String(),
"-interval", interval.String(),
"-pprof-duration", "0",
"-pprof-duration", "1s",
"-pprof-interval", "250ms",
})

// Get capture directory
Expand Down