Port over all OS tasks

tools/sgcollect/task_types.go

@@ -21,36 +21,28 @@ type SGCollectTask interface {
 	Run(ctx context.Context, opts *SGCollectOptions, out io.Writer) error
 }
 
-// SGCollectTaskEx wraps a SGCollectTask with information about how it should be run.
-type SGCollectTaskEx interface {
+type SGCollectTaskEx struct {
 	SGCollectTask
-	ShouldRun(platform string) bool
-	RequiresRoot() bool
-	NumSamples() int
-	Interval() time.Duration
-	Timeout() time.Duration
+	platforms          []string
+	root               bool
+	samples            int
+	interval           time.Duration
+	timeout            time.Duration
+	outputFileOverride string
+	noHeader           bool
 }
 
 // TaskEx wraps the given SGCollectTask in a SGCollectTaskEx if necessary.
 func TaskEx(t SGCollectTask) SGCollectTaskEx {
 	if ex, ok := t.(SGCollectTaskEx); ok {
 		return ex
 	}
-	return ex{
+	return SGCollectTaskEx{
 		SGCollectTask: t,
 	}
 }
 
-type ex struct {
-	SGCollectTask
-	platforms []string
-	root      bool
-	samples   int
-	interval  time.Duration
-	timeout   time.Duration
-}
-
-func (e ex) ShouldRun(platform string) bool {
+func (e SGCollectTaskEx) ShouldRun(platform string) bool {
 	if len(e.platforms) == 0 {
 		return true
 	}
@@ -62,57 +54,93 @@ func (e ex) ShouldRun(platform string) bool {
 	return false
 }
 
-func (e ex) RequiresRoot() bool {
+func (e SGCollectTaskEx) RequiresRoot() bool {
 	return e.root
 }
 
-func (e ex) NumSamples() int {
+func (e SGCollectTaskEx) NumSamples() int {
 	return e.samples
 }
 
-func (e ex) Interval() time.Duration {
+func (e SGCollectTaskEx) Interval() time.Duration {
 	return e.interval
 }
 
-func (e ex) Timeout() time.Duration {
+func (e SGCollectTaskEx) Timeout() time.Duration {
 	return e.timeout
 }
 
+func (e SGCollectTaskEx) Header() string {
+	if e.noHeader {
+		return ""
+	}
+	return e.SGCollectTask.Header()
+}
+
+func (e SGCollectTaskEx) OutputFile() string {
+	if e.outputFileOverride != "" {
+		return e.outputFileOverride
+	}
+	return e.SGCollectTask.OutputFile()
+}
+
 func Sample(t SGCollectTask, samples int, interval time.Duration) SGCollectTaskEx {
-	if ex, ok := t.(ex); ok {
+	if ex, ok := t.(SGCollectTaskEx); ok {
 		ex.samples = samples
 		ex.interval = interval
 		return ex
 	}
-	return ex{
+	return SGCollectTaskEx{
 		SGCollectTask: t,
 		samples:       samples,
 		interval:      interval,
 	}
 }
 
 func Timeout(t SGCollectTask, timeout time.Duration) SGCollectTaskEx {
-	if ex, ok := t.(ex); ok {
+	if ex, ok := t.(SGCollectTaskEx); ok {
 		ex.timeout = timeout
 		return ex
 	}
-	return ex{
+	return SGCollectTaskEx{
 		SGCollectTask: t,
 		timeout:       timeout,
 	}
 }
 
 func Privileged(t SGCollectTask) SGCollectTaskEx {
-	if ex, ok := t.(ex); ok {
+	if ex, ok := t.(SGCollectTaskEx); ok {
 		ex.root = true
 		return ex
 	}
-	return ex{
+	return SGCollectTaskEx{
 		SGCollectTask: t,
 		root:          true,
 	}
 }
 
+func OverrideOutput(t SGCollectTask, out string) SGCollectTaskEx {
+	if ex, ok := t.(SGCollectTaskEx); ok {
+		ex.outputFileOverride = out
+		return ex
+	}
+	return SGCollectTaskEx{
+		SGCollectTask:      t,
+		outputFileOverride: out,
+	}
+}
+
+func NoHeader(t SGCollectTask) SGCollectTaskEx {
+	if ex, ok := t.(SGCollectTaskEx); ok {
+		ex.noHeader = true
+		return ex
+	}
+	return SGCollectTaskEx{
+		SGCollectTask: t,
+		noHeader:      true,
+	}
+}
+
 type URLTask struct {
 	name       string
 	url        string
@@ -301,17 +329,26 @@ func (o *osTask) ShouldRun(platform string) bool {
 }
 
 // OSTask is a helper to return a task for the given platform, which can be one of
-// "unix", "linux, "windows", "solaris", or "" for all platforms.
+// "unix", "linux, "windows", "darwin", "solaris", or "" for all platforms.
 func OSTask(platform string, name, cmd string) SGCollectTask {
 	switch platform {
 	case "":
 		return &OSCommandTask{name, cmd, ""}
-	case "linux", "windows":
-		return &osTask{OSCommandTask{name, cmd, ""}, []string{platform}}
+	case "linux", "windows", "darwin":
+		return SGCollectTaskEx{
+			SGCollectTask: &OSCommandTask{name, cmd, ""},
+			platforms:     []string{platform},
+		}
 	case "unix":
-		return &osTask{OSCommandTask{name, cmd, ""}, []string{"linux", "darwin", "freebsd", "netbsd", "openbsd"}}
+		return SGCollectTaskEx{
+			SGCollectTask: &OSCommandTask{name, cmd, ""},
+			platforms:     []string{"linux", "darwin", "freebsd", "netbsd", "openbsd"},
+		}
 	case "solaris":
-		return &osTask{OSCommandTask{name, cmd, ""}, []string{"illumos", "solaris"}}
+		return SGCollectTaskEx{
+			SGCollectTask: &OSCommandTask{name, cmd, ""},
+			platforms:     []string{"illumos", "solaris"},
+		}
 	default:
 		panic(fmt.Sprintf("unknown platform %s", platform))
 	}

tools/sgcollect/tasks.go

@@ -49,6 +49,72 @@ func makeOSTasks() []SGCollectTask {
 		OSTask("linux", "Extended iostat", "iostat -x -p ALL 1 10 || iostat -x 1 10"),
 		OSTask("linux", "Core dump settings", "find /proc/sys/kernel -type f -name '*core*' -print -exec cat '{}' ';'"),
 		OSTask("unix", "sysctl settings", "sysctl -a"),
+		OSTask("linux", "lsof output", "echo sync_gateway | xargs -n1 pgrep | xargs -n1 -r -- lsof -n -p"),
+		OSTask("linux", "LVM info", "lvdisplay"),
+		OSTask("linux", "LVM info", "vgdisplay"),
+		OSTask("linux", "LVM info", "pvdisplay"),
+		OSTask("darwin", "Process list snapshot", "top -l 1"),
+		OSTask("darwin", "Disk activity", "iostat 1 10"),
+		OSTask("darwin", "Process list", "ps -Aww -o user,pid,lwp,ppid,nlwp,pcpu,pri,nice,vsize,rss,tty,stat,wchan:12,start,bsdtime,command"),
+		OSTask("windows", "Installed software", "wmic product get name, version"),
+		OSTask("windows", "Service list", "wmic service where state=\"running\" GET caption, name, state"),
+		OSTask("windows", "Process list", "wmic process"),
+		OSTask("windows", "Process usage", "tasklist /V /fo list"),
+		OSTask("windows", "Swap settings", "wmic pagefile"),
+		OSTask("windows", "Disk partition", "wmic partition"),
+		OSTask("windows", "Disk volumes", "wmic volume"),
+		Sample(OSTask("unix", "Network configuration", "ifconfig -a"), 2, 10*time.Second),
+		OSTask("linux", "Network configuration", "echo link addr neigh rule route netns | xargs -n1 -- sh -x -c 'ip $1 list' --"),
+		Sample(OSTask("windows", "Network configuration", "ipconfig /all"), 2, 10*time.Second),
+		OSTask("linux", "Raw /proc/net/dev", "cat /proc/net/dev"),
+		OSTask("linux", "Network link statistics", "ip -s link"),
+		OSTask("unix", "Network status", "netstat -anp || netstat -an"),
+		OSTask("windows", "Network status", "netstat -ano"),
+		OSTask("unix", "Network routing table", "netstat -rn"),
+		OSTask("linux", "Network socket statistics", "ss -an"),
+		OSTask("linux", "Extended socket statistics", "ss -an --info --processes"),
+		OSTask("unix", "Arp cache", "arp -na"),
+		OSTask("linux", "Iptables dump", "iptables-save"),
+		OSTask("unix", "Raw /etc/hosts", "cat /etc/hosts"),
+		OSTask("unix", "Raw /etc/resolv.conf", "cat /etc/resolv.conf"),
+		OSTask("unix", "Raw /etc/nsswitch.conf", "cat /etc/nsswitch.conf"),
+		OSTask("windows", "Arp cache", "arp -a"),
+		OSTask("windows", "Network Interface Controller", "wmic nic"),
+		OSTask("windows", "Network Adapter", "wmic nicconfig"),
+		OSTask("windows", "Active network connection", "wmic netuse"),
+		OSTask("windows", "Protocols", "wmic netprotocol"),
+		OSTask("windows", "Hosts file", `type %SystemRoot%\system32\drivers\etc\hosts`),
+		OSTask("windows", "Cache memory", "wmic memcache"),
+		OSTask("windows", "Physical memory", "wmic memphysical"),
+		OSTask("windows", "Physical memory chip info", "wmic memorychip"),
+		OSTask("windows", "Local storage devices", "wmic logicaldisk"),
+		OSTask("unix", "Filesystem", "df -ha"),
+		OSTask("unix", "System activity reporter", "sar 1 10"),
+		OSTask("unix", "System paging activity", "vmstat 1 10"),
+		OSTask("unix", "System uptime", "uptime"),
+		OSTask("unix", "couchbase user definition", "getent passwd couchbase"),
+		Privileged(OSTask("unix", "couchbase user limits", `su couchbase -c "ulimit -a"`)),
+		OSTask("unix", "sync_gateway user definition", "getent passwd sync_gateway"),
+		Privileged(OSTask("unix", "sync_gateway user limits", `su sync_gateway -c "ulimit -a"`)),
+		OSTask("unix", "Interrupt status", "intrstat 1 10"),
+		OSTask("unix", "Processor status", "mpstat 1 10"),
+		OSTask("unix", "System log", "cat /var/adm/messages"),
+		OSTask("linux", "Raw /proc/uptime", "cat /proc/uptime"),
+		NoHeader(OverrideOutput(OSTask("linux", "Systemd journal", "journalctl 2>&1 | gzip -c"), "systemd_journal.gz")),
+		NoHeader(OverrideOutput(OSTask("linux", "All logs", "tar cz /var/log/syslog* /var/log/dmesg /var/log/messages* /var/log/daemon* /var/log/debug* /var/log/kern.log* 2>/dev/null"), "syslog.tar.gz")),
+		OSTask("linux", "Relevant proc data", "echo sync_gateway | xargs -n1 pgrep | xargs -n1 -- sh -c 'echo $1; cat /proc/$1/status; cat /proc/$1/limits; cat /proc/$1/smaps; cat /proc/$1/numa_maps; cat /proc/$1/task/*/sched; echo' --"),
+		OSTask("linux", "Processes' environment", "echo sync_gateway | xargs -n1 pgrep | xargs -n1 -- sh -c 'echo $1; ( cat /proc/$1/environ | tr \\0 \\n ); echo' --"),
+		OSTask("linux", "NUMA data", "numactl --hardware"),
+		OSTask("linux", "NUMA data", "numactl --show"),
+		OSTask("linux", "NUMA data", "cat /sys/devices/system/node/node*/numastat"),
+		OSTask("unix", "Kernel log buffer", "dmesg -H || dmesg"),
+		OSTask("linux", "Transparent Huge Pages data", "cat /sys/kernel/mm/transparent_hugepage/enabled"),
+		OSTask("linux", "Transparent Huge Pages data", "cat /sys/kernel/mm/transparent_hugepage/defrag"),
+		OSTask("linux", "Transparent Huge Pages data", "cat /sys/kernel/mm/redhat_transparent_hugepage/enabled"),
+		OSTask("linux", "Transparent Huge Pages data", "cat /sys/kernel/mm/redhat_transparent_hugepage/defrag"),
+		OSTask("linux", "Network statistics", "netstat -s"),
+		OSTask("linux", "Full raw netstat", "cat /proc/net/netstat"),
+		OSTask("linux", "CPU throttling info", "echo /sys/devices/system/cpu/cpu*/thermal_throttle/* | xargs -n1 -- sh -c 'echo $1; cat $1' --"),
 	}
 }