Merge pull request #450 from weaveworks/scope-284

Optimize /proc traversal
This commit is contained in:
Alvaro
2015-09-14 14:28:49 +02:00
23 changed files with 991 additions and 366 deletions

View File

@@ -3,6 +3,7 @@ package docker
import (
"strconv"
"github.com/weaveworks/scope/probe/proc"
"github.com/weaveworks/scope/probe/process"
"github.com/weaveworks/scope/report"
)
@@ -16,18 +17,18 @@ const (
// These vars are exported for testing.
var (
NewProcessTreeStub = process.NewTree
NewProcessTreeStub = proc.NewTree
)
// Tagger is a tagger that tags Docker container information to process
// nodes that have a PID.
type Tagger struct {
registry Registry
procWalker process.Walker
procWalker proc.Reader
}
// NewTagger returns a usable Tagger.
func NewTagger(registry Registry, procWalker process.Walker) *Tagger {
func NewTagger(registry Registry, procWalker proc.Reader) *Tagger {
return &Tagger{
registry: registry,
procWalker: procWalker,
@@ -44,7 +45,7 @@ func (t *Tagger) Tag(r report.Report) (report.Report, error) {
return r, nil
}
func (t *Tagger) tag(tree process.Tree, topology *report.Topology) {
func (t *Tagger) tag(tree proc.Tree, topology *report.Topology) {
for nodeID, nodeMetadata := range topology.Nodes {
pidStr, ok := nodeMetadata.Metadata[process.PID]
if !ok {

View File

@@ -6,7 +6,7 @@ import (
"testing"
"github.com/weaveworks/scope/probe/docker"
"github.com/weaveworks/scope/probe/process"
"github.com/weaveworks/scope/probe/proc"
"github.com/weaveworks/scope/report"
"github.com/weaveworks/scope/test"
)
@@ -27,7 +27,7 @@ func TestTagger(t *testing.T) {
oldProcessTree := docker.NewProcessTreeStub
defer func() { docker.NewProcessTreeStub = oldProcessTree }()
docker.NewProcessTreeStub = func(_ process.Walker) (process.Tree, error) {
docker.NewProcessTreeStub = func(_ proc.Reader) (proc.Tree, error) {
return &mockProcessTree{map[int]int{2: 1}}, nil
}

View File

@@ -7,7 +7,7 @@ import (
"github.com/prometheus/client_golang/prometheus"
"github.com/weaveworks/procspy"
"github.com/weaveworks/scope/probe/proc"
"github.com/weaveworks/scope/probe/process"
"github.com/weaveworks/scope/report"
)
@@ -28,6 +28,7 @@ type Reporter struct {
conntracker *Conntracker
natmapper *natmapper
revResolver *ReverseResolver
procReader proc.Reader
}
// SpyDuration is an exported prometheus metric
@@ -42,12 +43,12 @@ var SpyDuration = prometheus.NewSummaryVec(
[]string{},
)
// NewReporter creates a new Reporter that invokes procspy.Connections to
// NewReporter creates a new Reporter that invokes Connections to
// generate a report.Report that contains every discovered (spied) connection
// on the host machine, at the granularity of host and port. That information
// is stored in the Endpoint topology. It optionally enriches that topology
// with process (PID) information.
func NewReporter(hostID, hostName string, includeProcesses bool, useConntrack bool) *Reporter {
func NewReporter(hostID, hostName string, includeProcesses bool, procReader proc.Reader, useConntrack bool) *Reporter {
var (
conntrackModulePresent = ConntrackModulePresent()
conntracker *Conntracker
@@ -73,6 +74,7 @@ func NewReporter(hostID, hostName string, includeProcesses bool, useConntrack bo
conntracker: conntracker,
natmapper: natmapper,
revResolver: NewReverseResolver(),
procReader: procReader,
}
}
@@ -94,12 +96,7 @@ func (r *Reporter) Report() (report.Report, error) {
}(time.Now())
rpt := report.MakeReport()
conns, err := procspy.Connections(r.includeProcesses)
if err != nil {
return rpt, err
}
for conn := conns.Next(); conn != nil; conn = conns.Next() {
err := r.procReader.Connections(r.includeProcesses, func(conn proc.Connection) {
var (
localPort = conn.LocalPort
remotePort = conn.RemotePort
@@ -107,12 +104,15 @@ func (r *Reporter) Report() (report.Report, error) {
remoteAddr = conn.RemoteAddress.String()
)
extraNodeInfo := report.MakeNode()
if conn.Proc.PID > 0 {
if conn.Process.PID > 0 {
extraNodeInfo = extraNodeInfo.WithMetadata(report.Metadata{
process.PID: strconv.FormatUint(uint64(conn.Proc.PID), 10),
process.PID: strconv.FormatUint(uint64(conn.Process.PID), 10),
})
}
r.addConnection(&rpt, localAddr, remoteAddr, localPort, remotePort, &extraNodeInfo, nil)
})
if err != nil {
return rpt, err
}
if r.conntracker != nil {

View File

@@ -5,9 +5,9 @@ import (
"strconv"
"testing"
"github.com/weaveworks/procspy"
"github.com/weaveworks/scope/probe/docker"
"github.com/weaveworks/scope/probe/endpoint"
"github.com/weaveworks/scope/probe/proc"
"github.com/weaveworks/scope/report"
)
@@ -17,10 +17,10 @@ var (
fixRemoteAddress = net.ParseIP("192.168.1.2")
fixRemotePort = uint16(12345)
fixRemotePortB = uint16(12346)
fixProcessPID = uint(4242)
fixProcessPID = int(4242)
fixProcessName = "nginx"
fixConnections = []procspy.Connection{
fixConnections = []proc.Connection{
{
Transport: "tcp",
LocalAddress: fixLocalAddress,
@@ -37,16 +37,16 @@ var (
},
}
fixConnectionsWithProcesses = []procspy.Connection{
fixConnectionsWithProcesses = []proc.Connection{
{
Transport: "tcp",
LocalAddress: fixLocalAddress,
LocalPort: fixLocalPort,
RemoteAddress: fixRemoteAddress,
RemotePort: fixRemotePort,
Proc: procspy.Proc{
Process: proc.Process{
PID: fixProcessPID,
Name: fixProcessName,
Comm: fixProcessName,
},
},
{
@@ -55,23 +55,22 @@ var (
LocalPort: fixLocalPort,
RemoteAddress: fixRemoteAddress,
RemotePort: fixRemotePort,
Proc: procspy.Proc{
Process: proc.Process{
PID: fixProcessPID,
Name: fixProcessName,
Comm: fixProcessName,
},
},
}
)
func TestSpyNoProcesses(t *testing.T) {
procspy.SetFixtures(fixConnections)
const (
nodeID = "heinz-tomato-ketchup" // TODO rename to hostID
nodeName = "frenchs-since-1904" // TODO rename to hostNmae
)
reporter := endpoint.NewReporter(nodeID, nodeName, false, false)
procReader := proc.MockedReader{Conns: fixConnections}
reporter := endpoint.NewReporter(nodeID, nodeName, false, &procReader, false)
r, _ := reporter.Report()
//buf, _ := json.MarshalIndent(r, "", " ")
//t.Logf("\n%s\n", buf)
@@ -100,14 +99,13 @@ func TestSpyNoProcesses(t *testing.T) {
}
func TestSpyWithProcesses(t *testing.T) {
procspy.SetFixtures(fixConnectionsWithProcesses)
const (
nodeID = "nikon" // TODO rename to hostID
nodeName = "fishermans-friend" // TODO rename to hostNmae
)
reporter := endpoint.NewReporter(nodeID, nodeName, true, false)
procReader := proc.MockedReader{Conns: fixConnectionsWithProcesses}
reporter := endpoint.NewReporter(nodeID, nodeName, true, &procReader, false)
r, _ := reporter.Report()
// buf, _ := json.MarshalIndent(r, "", " ") ; t.Logf("\n%s\n", buf)

View File

@@ -14,11 +14,11 @@ import (
"syscall"
"time"
"github.com/weaveworks/procspy"
"github.com/weaveworks/scope/probe/docker"
"github.com/weaveworks/scope/probe/endpoint"
"github.com/weaveworks/scope/probe/host"
"github.com/weaveworks/scope/probe/overlay"
"github.com/weaveworks/scope/probe/proc"
"github.com/weaveworks/scope/probe/process"
"github.com/weaveworks/scope/probe/sniff"
"github.com/weaveworks/scope/report"
@@ -67,8 +67,6 @@ func main() {
}
log.Printf("publishing to: %s", strings.Join(targets, ", "))
procspy.SetProcRoot(*procRoot)
if *httpListen != "" {
log.Printf("profiling data being exported to %s", *httpListen)
log.Printf("go tool pprof http://%s/debug/pprof/{profile,heap,block}", *httpListen)
@@ -111,16 +109,18 @@ func main() {
}
var (
endpointReporter = endpoint.NewReporter(hostID, hostName, *spyProcs, *useConntrack)
processCache = process.NewCachingWalker(process.NewWalker(*procRoot))
tickers = []Ticker{processCache}
procDir = proc.OSDir{Dir: *procRoot}
procReader = proc.NewCachingProcReader(proc.NewReader(procDir), *spyProcs)
tickers = []Ticker{procReader}
endpointReporter = endpoint.NewReporter(hostID, hostName, *spyProcs, procReader, *useConntrack)
reporters = []Reporter{
endpointReporter,
host.NewReporter(hostID, hostName, localNets),
process.NewReporter(processCache, hostID),
process.NewReporter(procReader, hostID),
}
taggers = []Tagger{newTopologyTagger(), host.NewTagger(hostID)}
)
defer procReader.Close()
defer endpointReporter.Stop()
if *dockerEnabled {
@@ -134,7 +134,7 @@ func main() {
}
defer dockerRegistry.Stop()
taggers = append(taggers, docker.NewTagger(dockerRegistry, processCache))
taggers = append(taggers, docker.NewTagger(dockerRegistry, procReader))
reporters = append(reporters, docker.NewReporter(dockerRegistry, hostID))
}

30
probe/proc/mocks.go Normal file
View File

@@ -0,0 +1,30 @@
package proc
// note: we must keep this in the "proc" package so it can be used from other packages..
// MockedReader is a mocked "/proc" reader
type MockedReader struct {
Procs []Process
Conns []Connection
}
// Processes walks through the processes provided in the mocked "/proc" reader
func (mw MockedReader) Processes(f func(Process)) error {
for _, p := range mw.Procs {
f(p)
}
return nil
}
// Connections walks through the connections provided in the mocked "/proc" reader
func (mw *MockedReader) Connections(_ bool, f func(Connection)) error {
for _, c := range mw.Conns {
f(c)
}
return nil
}
// Close (mocked version)
func (mw *MockedReader) Close() error {
return nil
}

27
probe/proc/mocks_test.go Normal file
View File

@@ -0,0 +1,27 @@
package proc_test
import (
"bytes"
"github.com/weaveworks/scope/probe/proc"
)
// mockedFile is a mocked file in the "/proc" directory
type mockedFile struct {
Path string
ReadIntoFunc func(buf *bytes.Buffer) error
}
func (mpf mockedFile) ReadInto(buf *bytes.Buffer) error { return mpf.ReadIntoFunc(buf) }
func (mpf mockedFile) Close() error { return nil }
// mockedDir is a mocked "/proc" directory
type mockedDir struct {
Dir string
OpenFunc func(string) (proc.File, error)
ReadDirNamesFunc func(string) ([]string, error)
}
func (p mockedDir) Root() string { return p.Dir }
func (p mockedDir) Open(s string) (proc.File, error) { return p.OpenFunc(s) }
func (p mockedDir) ReadDirNames(s string) ([]string, error) { return p.ReadDirNamesFunc(s) }

43
probe/proc/net.go Normal file
View File

@@ -0,0 +1,43 @@
package proc
import (
"fmt"
"net"
)
const (
tcpEstablished = 1 // according to /include/net/tcp_states.h
)
// Connection is a (TCP) connection. The 'Process' struct might not be filled in.
type Connection struct {
Transport string
LocalAddress net.IP
LocalPort uint16
RemoteAddress net.IP
RemotePort uint16
inode uint64
Process
}
// Copy returns a copy of a connection
func (c Connection) Copy() Connection {
dupIP := func(ip net.IP) net.IP {
dup := make(net.IP, len(ip))
copy(dup, ip)
return dup
}
c.LocalAddress = dupIP(c.LocalAddress)
c.RemoteAddress = dupIP(c.RemoteAddress)
c.Process = c.Process.Copy()
return c
}
// String returns the string repr
func (c Connection) String() string {
return fmt.Sprintf("%s:%d - %s:%d %s#%d",
c.LocalAddress, c.LocalPort,
c.RemoteAddress, c.RemotePort,
c.Transport, c.inode)
}

81
probe/proc/net_darwin.go Normal file
View File

@@ -0,0 +1,81 @@
package proc
import (
"net"
"strconv"
"strings"
)
// parseDarwinNetstat parses netstat output. (Linux has ip:port, darwin
// ip.port. The 'Proto' column value also differs.)
func parseDarwinNetstat(out string) []Connection {
//
// Active Internet connections
// Proto Recv-Q Send-Q Local Address Foreign Address (state)
// tcp4 0 0 10.0.1.6.58287 1.2.3.4.443 ESTABLISHED
//
res := []Connection{}
for i, line := range strings.Split(out, "\n") {
if i == 0 || i == 1 {
// Skip header
continue
}
// Fields are:
fields := strings.Fields(line)
if len(fields) != 6 {
continue
}
if fields[5] != "ESTABLISHED" {
continue
}
t := Connection{
Transport: "tcp",
}
// Format is <ip>.<port>
locals := strings.Split(fields[3], ".")
if len(locals) < 2 {
continue
}
var (
localAddress = strings.Join(locals[:len(locals)-1], ".")
localPort = locals[len(locals)-1]
)
t.LocalAddress = net.ParseIP(localAddress)
p, err := strconv.Atoi(localPort)
if err != nil {
return nil
}
t.LocalPort = uint16(p)
remotes := strings.Split(fields[4], ".")
if len(remotes) < 2 {
continue
}
var (
remoteAddress = strings.Join(remotes[:len(remotes)-1], ".")
remotePort = remotes[len(remotes)-1]
)
t.RemoteAddress = net.ParseIP(remoteAddress)
p, err = strconv.Atoi(remotePort)
if err != nil {
return nil
}
t.RemotePort = uint16(p)
res = append(res, t)
}
return res
}

162
probe/proc/net_linux.go Normal file
View File

@@ -0,0 +1,162 @@
package proc
import (
"bytes"
"net"
)
type netReader struct {
b []byte
c Connection
wantedState uint
bytesLocal, bytesRemote [16]byte
}
func newNetReader(b []byte, wantedState uint) *netReader {
return &netReader{
b: b,
c: Connection{},
wantedState: wantedState,
}
}
// Next returns the next connection. All buffers are re-used, so if you want
// to keep the IPs you have to copy them.
func (p *netReader) Next() *Connection {
again:
if len(p.b) == 0 {
return nil
}
b := p.b
if p.b[2] == 's' {
// Skip header
p.b = nextLine(b)
goto again
}
var (
local, remote, state, inode []byte
)
_, b = nextField(b) // 'sl' column
local, b = nextField(b)
remote, b = nextField(b)
state, b = nextField(b)
if parseHex(state) != p.wantedState {
p.b = nextLine(b)
goto again
}
_, b = nextField(b) // 'tx_queue' column
_, b = nextField(b) // 'rx_queue' column
_, b = nextField(b) // 'tr' column
_, b = nextField(b) // 'uid' column
_, b = nextField(b) // 'timeout' column
inode, b = nextField(b)
p.c.LocalAddress, p.c.LocalPort = scanAddressNA(local, &p.bytesLocal)
p.c.RemoteAddress, p.c.RemotePort = scanAddressNA(remote, &p.bytesRemote)
p.c.inode = parseDec(inode)
p.b = nextLine(b)
return &p.c
}
// scanAddressNA parses 'A12CF62E:00AA' to the address/port. Handles IPv4 and
// IPv6 addresses. The address is a big endian 32 bit ints, hex encoded. We
// just decode the hex and flip the bytes in every group of 4.
func scanAddressNA(in []byte, buf *[16]byte) (net.IP, uint16) {
col := bytes.IndexByte(in, ':')
if col == -1 {
return nil, 0
}
// Network address is big endian. Can be either ipv4 or ipv6.
address := hexDecode32bigNA(in[:col], buf)
return net.IP(address), uint16(parseHex(in[col+1:]))
}
// hexDecode32big decodes sequences of 32bit big endian bytes.
func hexDecode32bigNA(src []byte, buf *[16]byte) []byte {
blocks := len(src) / 8
for block := 0; block < blocks; block++ {
for i := 0; i < 4; i++ {
a := fromHexChar(src[block*8+i*2])
b := fromHexChar(src[block*8+i*2+1])
buf[block*4+3-i] = (a << 4) | b
}
}
return buf[:blocks*4]
}
func nextField(s []byte) ([]byte, []byte) {
// Skip whitespace.
for i, b := range s {
if b != ' ' {
s = s[i:]
break
}
}
// Up until the next whitespace field.
for i, b := range s {
if b == ' ' {
return s[:i], s[i:]
}
}
return nil, nil
}
func nextLine(s []byte) []byte {
i := bytes.IndexByte(s, '\n')
if i == -1 {
return nil
}
return s[i+1:]
}
// Simplified copy of strconv.ParseUint(16).
func parseHex(s []byte) uint {
n := uint(0)
for i := 0; i < len(s); i++ {
n *= 16
n += uint(fromHexChar(s[i]))
}
return n
}
// Simplified copy of strconv.ParseUint(10).
func parseDec(s []byte) uint64 {
n := uint64(0)
for _, c := range s {
n *= 10
n += uint64(c - '0')
}
return n
}
// hexDecode32big decodes sequences of 32bit big endian bytes.
func hexDecode32big(src []byte) []byte {
dst := make([]byte, len(src)/2)
blocks := len(src) / 8
for block := 0; block < blocks; block++ {
for i := 0; i < 4; i++ {
a := fromHexChar(src[block*8+i*2])
b := fromHexChar(src[block*8+i*2+1])
dst[block*4+3-i] = (a << 4) | b
}
}
return dst
}
// fromHexChar converts a hex character into its value.
func fromHexChar(c byte) uint8 {
switch {
case '0' <= c && c <= '9':
return c - '0'
case 'a' <= c && c <= 'f':
return c - 'a' + 10
case 'A' <= c && c <= 'F':
return c - 'A' + 10
}
return 0
}

165
probe/proc/reader.go Normal file
View File

@@ -0,0 +1,165 @@
package proc
import (
"bytes"
"os"
"path"
"sync"
)
// File is a file in the "/proc" directory
type File interface {
ReadInto(buf *bytes.Buffer) error
Close() error
}
// OSFile is a native file
type OSFile struct{ *os.File }
// ReadInto reads the whole file into a buffer
func (of *OSFile) ReadInto(buf *bytes.Buffer) error {
if _, err := of.File.Seek(0, 0); err != nil {
return err
}
_, err := buf.ReadFrom(of.File)
return err
}
// Dir is the '/proc' directory and the associated ops for
// reading subdirs or files.
type Dir interface {
Root() string // the "/proc" directory
Open(s string) (File, error) // open a file in the "/proc" dir
ReadDirNames(string) ([]string, error) // list a subdirectory in the "/proc"
}
// OSDir is a OS "/proc" firectory
type OSDir struct{ Dir string }
// Root returns the "/proc" top dir
func (dp OSDir) Root() string {
return dp.Dir
}
// Open returns a ProcFile
func (dp OSDir) Open(s string) (File, error) {
h, err := os.Open(path.Join(dp.Root(), s))
if err != nil {
return nil, err
}
return &OSFile{h}, nil
}
// ReadDirNames reads all the directory entries
func (dp OSDir) ReadDirNames(s string) ([]string, error) {
f, err := os.Open(s)
if err != nil {
return nil, err
}
list, err := f.Readdirnames(-1)
f.Close()
if err != nil {
return nil, err
}
return list, nil
}
// DefaultProcDir is the default '/proc' directory
var DefaultProcDir = OSDir{Dir: "/proc"}
// Process represents a single process.
type Process struct {
PID, PPID int
Comm string
Cmdline string
Threads int
Inodes []uint64
}
// Copy returns a copy of a process
func (p Process) Copy() Process {
dup := make([]uint64, len(p.Inodes))
copy(dup, p.Inodes)
p.Inodes = dup
return p
}
// Reader is something that reads the /proc directory and
// returns some info like processes and connections
type Reader interface {
// Processes walks through the processes
Processes(func(Process)) error
// Connections walks through the connections
Connections(bool, func(Connection)) error
// Close closes the "/proc" reader
Close() error
}
// CachingProcReader is a '/proc' reader than caches a copy of the output from another
// '/proc' reader, and then allows other concurrent readers to Walk that copy.
type CachingProcReader struct {
procsCache []Process
connsCache []Connection
source Reader
includeProcs bool
sync.RWMutex
}
// NewCachingProcReader returns a new CachingProcReader
func NewCachingProcReader(source Reader, includeProcs bool) *CachingProcReader {
return &CachingProcReader{source: source, includeProcs: includeProcs}
}
// Processes walks a cached copy of process list
func (c *CachingProcReader) Processes(f func(Process)) error {
c.RLock()
defer c.RUnlock()
for _, p := range c.procsCache {
f(p)
}
return nil
}
// Connections walks a cached copy of the connections list
// Note: specifying 'includeProcs' has no effect here, as the cached copy
func (c *CachingProcReader) Connections(_ bool, f func(Connection)) error {
c.RLock()
defer c.RUnlock()
for _, c := range c.connsCache {
f(c)
}
return nil
}
// Close closes the "/proc" reader
func (c *CachingProcReader) Close() error {
return c.source.Close()
}
// Tick updates the cached copy of the processes and connections lists
func (c *CachingProcReader) Tick() error {
newProcsCache := []Process{}
newConnsCache := []Connection{}
if err := c.source.Processes(func(p Process) {
newProcsCache = append(newProcsCache, p)
}); err != nil {
return err
}
if err := c.source.Connections(c.includeProcs, func(conn Connection) {
newConnsCache = append(newConnsCache, conn.Copy())
}); err != nil {
return err
}
c.Lock()
defer c.Unlock()
c.procsCache = newProcsCache
c.connsCache = newConnsCache
return nil
}

View File

@@ -1,27 +1,28 @@
package process
package proc
import (
"fmt"
"net"
"os/exec"
"strconv"
"strings"
)
// NewWalker returns a Darwin (lsof-based) walker.
func NewWalker(_ string) Walker {
return &walker{}
type reader struct{}
// NewReader returns a Darwin (lsof-based) '/proc' reader
func NewReader(proc Dir) Reader {
return &reader{}
}
type walker struct{}
const (
lsofBinary = "lsof"
lsofFields = "cn" // parseLSOF() depends on the order
lsofBinary = "lsof"
lsofFields = "cn" // parseLSOF() depends on the order
netstatBinary = "netstat"
lsofBinary = "lsof"
)
// These functions copied from procspy.
func (walker) Walk(f func(Process)) error {
func (reader) Processes(f func(Process)) error {
output, err := exec.Command(
lsofBinary,
"-i", // only Internet files
@@ -44,6 +45,59 @@ func (walker) Walk(f func(Process)) error {
return nil
}
func (r *reader) Connections(withProcs bool, f func(Connection)) error {
out, err := exec.Command(
netstatBinary,
"-n", // no number resolving
"-W", // Wide output
// "-l", // full IPv6 addresses // What does this do?
"-p", "tcp", // only TCP
).CombinedOutput()
if err != nil {
return err
}
connections := parseDarwinNetstat(string(out))
if withProcs {
out, err := exec.Command(
lsofBinary,
"-i", // only Internet files
"-n", "-P", // no number resolving
"-w", // no warnings
"-F", lsofFields, // \n based output of only the fields we want.
).CombinedOutput()
if err != nil {
return err
}
procs, err := parseLSOF(string(out))
if err != nil {
return err
}
for local, proc := range procs {
for i, c := range connections {
localAddr := net.JoinHostPort(
c.LocalAddress.String(),
strconv.Itoa(int(c.LocalPort)),
)
if localAddr == local {
connections[i].Proc = proc
}
}
}
}
for _, c := range connections {
f(c)
}
return nil
}
// Close closes the Darwin "/proc" reader
func (reader) Close() error {
return nil
}
func parseLSOF(output string) (map[string]Process, error) {
var (
processes = map[string]Process{} // Local addr -> Proc

202
probe/proc/reader_linux.go Normal file
View File

@@ -0,0 +1,202 @@
package proc
import (
"bytes"
"path"
"strconv"
"strings"
"sync"
"syscall"
"time"
"github.com/bluele/gcache"
)
const (
filesCacheLen = 512
filesCacheExpiration = 60 * time.Second
)
var tcpFiles = []string{
"net/tcp",
"net/tcp6",
}
// A cache for files handles
type filesCache struct {
handles gcache.Cache
}
type filesCacheEntry struct {
sync.RWMutex
File
}
func newFilesCache(proc Dir) *filesCache {
loadFunc := func(fileName interface{}) (interface{}, error) {
f, err := proc.Open(fileName.(string))
if err != nil {
return nil, err
}
return filesCacheEntry{File: f}, nil
}
evictionFunc := func(key, value interface{}) {
value.(filesCacheEntry).Close()
}
return &filesCache{
handles: gcache.New(filesCacheLen).LoaderFunc(loadFunc).EvictedFunc(evictionFunc).Expiration(filesCacheExpiration).ARC().Build(),
}
}
// Read a "/proc" file, identified as a file in a subdir (eg "1134/comm"), into a buffer
func (fc *filesCache) ReadInto(filename string, buf *bytes.Buffer) error {
// we could use a lock here, but this is only used from Processes()/Connections(),
// and they are always invoked sequentially...
h, err := fc.handles.Get(filename)
if err != nil {
return err
}
handle := h.(filesCacheEntry)
handle.Lock()
defer handle.Unlock()
return handle.ReadInto(buf)
}
// Close closes all the handles in the cache
func (fc *filesCache) Close() error {
for _, key := range fc.handles.Keys() {
fc.handles.Remove(key)
}
return nil
}
// the Linux "/proc" reader
type reader struct {
proc Dir
handles *filesCache
}
// NewReader creates a new /proc reader.
func NewReader(proc Dir) Reader {
return &reader{
proc: proc,
handles: newFilesCache(proc),
}
}
// Close closes the Linux "/proc" reader
func (r *reader) Close() error {
return r.handles.Close()
}
// Processes walks the /proc directory and marshalls the files into
// instances of Process, which it then passes one-by-one to the
// supplied function. Processes() is only made public so that is
// can be tested.
func (r *reader) Processes(f func(Process)) error {
dirEntries, err := r.proc.ReadDirNames(r.proc.Root())
if err != nil {
return err
}
var fdStat syscall.Stat_t
buf := bytes.Buffer{}
for _, subdir := range dirEntries {
readIntoBuffer := func(filename string) error {
buf.Reset()
return r.handles.ReadInto(path.Join(subdir, filename), &buf)
}
pid, err := strconv.Atoi(subdir)
if err != nil {
continue
}
if readIntoBuffer("stat") != nil {
continue
}
splits := strings.Fields(buf.String())
ppid, err := strconv.Atoi(splits[3])
if err != nil {
return err
}
threads, err := strconv.Atoi(splits[19])
if err != nil {
return err
}
cmdline := ""
if readIntoBuffer("cmdline") == nil {
cmdlineBuf := bytes.Replace(buf.Bytes(), []byte{'\000'}, []byte{' '}, -1)
cmdline = string(cmdlineBuf)
}
comm := "(unknown)"
if readIntoBuffer("comm") == nil {
comm = strings.TrimSpace(buf.String())
}
fdBase := path.Join(r.proc.Root(), strconv.Itoa(pid), "fd")
fdNames, err := r.proc.ReadDirNames(fdBase)
if err != nil {
return err
}
inodes := []uint64{}
for _, fdName := range fdNames {
// Direct use of syscall.Stat() to save garbage.
fdPath := path.Join(fdBase, fdName)
err = syscall.Stat(fdPath, &fdStat)
if err == nil && (fdStat.Mode&syscall.S_IFMT == syscall.S_IFSOCK) { // We want sockets only.
inodes = append(inodes, fdStat.Ino)
}
}
f(Process{
PID: pid,
PPID: ppid,
Comm: comm,
Cmdline: cmdline,
Threads: threads,
Inodes: inodes,
})
}
return nil
}
// Connections walks through all the connections in the "/proc"
func (r *reader) Connections(withProcs bool, f func(Connection)) error {
// create a map of inode->Process
procs := make(map[uint64]Process)
if withProcs {
r.Processes(func(p Process) {
for _, inode := range p.Inodes {
procs[inode] = p
}
})
}
buf := bytes.Buffer{}
for _, tcpFile := range tcpFiles {
err := r.handles.ReadInto(tcpFile, &buf)
if err != nil {
return err
}
}
pn := newNetReader(buf.Bytes(), tcpEstablished)
for {
conn := pn.Next()
if conn == nil {
break // Done!
}
if proc, ok := procs[conn.inode]; ok {
conn.Process = proc
}
f(*conn)
}
return nil
}

View File

@@ -0,0 +1,93 @@
package proc_test
import (
"bytes"
"fmt"
"os"
"reflect"
"strconv"
"strings"
"testing"
"time"
"github.com/weaveworks/scope/probe/proc"
"github.com/weaveworks/scope/test"
)
type mockedProcess struct {
ID, Comm, Cmdline string
}
func (p mockedProcess) Name() string { return p.ID }
func (p mockedProcess) Size() int64 { return 0 }
func (p mockedProcess) Mode() os.FileMode { return 0 }
func (p mockedProcess) ModTime() time.Time { return time.Now() }
func (p mockedProcess) IsDir() bool { return true }
func (p mockedProcess) Sys() interface{} { return nil }
func TestProcReaderProcesses(t *testing.T) {
processes := map[string]mockedProcess{
"3": {ID: "3", Comm: "curl\n", Cmdline: "curl\000google.com"},
"2": {ID: "2", Comm: "bash\n"},
"4": {ID: "4", Comm: "apache\n"},
"notapid": {ID: "notapid"},
"1": {ID: "1", Comm: "init\n"},
}
want := map[int]proc.Process{
3: {PID: 3, PPID: 2, Comm: "curl", Cmdline: "curl google.com", Threads: 1, Inodes: []uint64{}},
2: {PID: 2, PPID: 1, Comm: "bash", Cmdline: "", Threads: 1, Inodes: []uint64{}},
4: {PID: 4, PPID: 3, Comm: "apache", Cmdline: "", Threads: 1, Inodes: []uint64{}},
1: {PID: 1, PPID: 0, Comm: "init", Cmdline: "", Threads: 1, Inodes: []uint64{}},
}
// use a mocked /proc that reads from our mocked processes
procDir := mockedDir{
ReadDirNamesFunc: func(path string) ([]string, error) {
result := []string{}
for k := range processes {
result = append(result, k)
}
return result, nil
},
OpenFunc: func(filename string) (proc.File, error) {
splits := strings.Split(filename, "/")
pid := splits[len(splits)-2]
process, ok := processes[pid]
if !ok {
return nil, fmt.Errorf("not found")
}
file := splits[len(splits)-1]
var content []byte
switch file {
case "comm":
content = []byte(process.Comm)
case "stat":
pid, _ := strconv.Atoi(splits[len(splits)-2])
parent := pid - 1
content = []byte(fmt.Sprintf("%d na R %d 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1", pid, parent))
case "cmdline":
content = []byte(process.Cmdline)
default:
return nil, fmt.Errorf("not found")
}
return mockedFile{
ReadIntoFunc: func(buf *bytes.Buffer) error {
_, err := buf.Write(content)
return err
},
}, nil
},
}
procReader := proc.NewReader(procDir)
have := map[int]proc.Process{}
err := procReader.Processes(func(p proc.Process) {
have[p.PID] = p
})
if err != nil || !reflect.DeepEqual(want, have) {
t.Errorf("%v (%v)", test.Diff(want, have), err)
}
}

69
probe/proc/reader_test.go Normal file
View File

@@ -0,0 +1,69 @@
package proc_test
import (
"reflect"
"testing"
"github.com/weaveworks/scope/probe/proc"
"github.com/weaveworks/scope/test"
)
func TestProcReaderBasic(t *testing.T) {
procFunc := func(proc.Process) {}
nullProcDir := mockedDir{
Dir: "",
OpenFunc: func(string) (proc.File, error) { return &proc.OSFile{File: nil}, nil },
ReadDirNamesFunc: func(string) ([]string, error) { return []string{}, nil },
}
if err := proc.NewReader(nullProcDir).Processes(procFunc); err != nil {
t.Fatal(err)
}
}
func TestCachingProcReader(t *testing.T) {
all := func(w proc.Reader) ([]proc.Process, error) {
all := []proc.Process{}
err := w.Processes(func(p proc.Process) {
all = append(all, p)
})
return all, err
}
processes := []proc.Process{
{PID: 1, PPID: 0, Comm: "init"},
{PID: 2, PPID: 1, Comm: "bash"},
{PID: 3, PPID: 1, Comm: "apache", Threads: 2},
{PID: 4, PPID: 2, Comm: "ping", Cmdline: "ping foo.bar.local"},
}
procReader := &proc.MockedReader{
Procs: processes,
}
cachingProcReader := proc.NewCachingProcReader(procReader, true)
err := cachingProcReader.Tick()
if err != nil {
t.Fatal(err)
}
have, err := all(cachingProcReader)
if err != nil || !reflect.DeepEqual(processes, have) {
t.Errorf("%v (%v)", test.Diff(processes, have), err)
}
procReader.Procs = []proc.Process{}
have, err = all(cachingProcReader)
if err != nil || !reflect.DeepEqual(processes, have) {
t.Errorf("%v (%v)", test.Diff(processes, have), err)
}
err = cachingProcReader.Tick()
if err != nil {
t.Fatal(err)
}
have, err = all(cachingProcReader)
want := []proc.Process{}
if err != nil || !reflect.DeepEqual(want, have) {
t.Errorf("%v (%v)", test.Diff(want, have), err)
}
}

View File

@@ -1,4 +1,4 @@
package process
package proc
import (
"fmt"
@@ -14,9 +14,9 @@ type tree struct {
}
// NewTree returns a new Tree that can be polled.
func NewTree(walker Walker) (Tree, error) {
func NewTree(walker Reader) (Tree, error) {
pt := tree{processes: map[int]Process{}}
err := walker.Walk(func(p Process) {
err := walker.Processes(func(p Process) {
pt.processes[p.PID] = p
})

View File

@@ -1,15 +1,15 @@
package process_test
package proc_test
import (
"reflect"
"testing"
"github.com/weaveworks/scope/probe/process"
"github.com/weaveworks/scope/probe/proc"
)
func TestTree(t *testing.T) {
walker := &mockWalker{
processes: []process.Process{
walker := &proc.MockedReader{
Procs: []proc.Process{
{PID: 1, PPID: 0, Comm: "init"},
{PID: 2, PPID: 1, Comm: "bash"},
{PID: 3, PPID: 1, Comm: "apache", Threads: 2},
@@ -17,7 +17,7 @@ func TestTree(t *testing.T) {
},
}
tree, err := process.NewTree(walker)
tree, err := proc.NewTree(walker)
if err != nil {
t.Fatalf("newProcessTree error: %v", err)
}

View File

@@ -3,6 +3,7 @@ package process
import (
"strconv"
"github.com/weaveworks/scope/probe/proc"
"github.com/weaveworks/scope/report"
)
@@ -18,11 +19,11 @@ const (
// Reporter generates Reports containing the Process topology.
type Reporter struct {
scope string
walker Walker
walker proc.Reader
}
// NewReporter makes a new Reporter.
func NewReporter(walker Walker, scope string) *Reporter {
func NewReporter(walker proc.Reader, scope string) *Reporter {
return &Reporter{
scope: scope,
walker: walker,
@@ -42,7 +43,7 @@ func (r *Reporter) Report() (report.Report, error) {
func (r *Reporter) processTopology() (report.Topology, error) {
t := report.MakeTopology()
err := r.walker.Walk(func(p Process) {
err := r.walker.Processes(func(p proc.Process) {
pidstr := strconv.Itoa(p.PID)
nodeID := report.MakeProcessNodeID(r.scope, pidstr)
t.Nodes[nodeID] = report.MakeNode()

View File

@@ -4,25 +4,15 @@ import (
"reflect"
"testing"
"github.com/weaveworks/scope/probe/proc"
"github.com/weaveworks/scope/probe/process"
"github.com/weaveworks/scope/report"
"github.com/weaveworks/scope/test"
)
type mockWalker struct {
processes []process.Process
}
func (m *mockWalker) Walk(f func(process.Process)) error {
for _, p := range m.processes {
f(p)
}
return nil
}
func TestReporter(t *testing.T) {
walker := &mockWalker{
processes: []process.Process{
procReader := &proc.MockedReader{
Procs: []proc.Process{
{PID: 1, PPID: 0, Comm: "init"},
{PID: 2, PPID: 1, Comm: "bash"},
{PID: 3, PPID: 1, Comm: "apache", Threads: 2},
@@ -31,7 +21,7 @@ func TestReporter(t *testing.T) {
},
}
reporter := process.NewReporter(walker, "")
reporter := process.NewReporter(procReader, "")
want := report.MakeReport()
want.Process = report.Topology{
Nodes: report.Nodes{

View File

@@ -1,56 +0,0 @@
package process
import "sync"
// Process represents a single process.
type Process struct {
PID, PPID int
Comm string
Cmdline string
Threads int
}
// Walker is something that walks the /proc directory
type Walker interface {
Walk(func(Process)) error
}
// CachingWalker is a walker than caches a copy of the output from another
// Walker, and then allows other concurrent readers to Walk that copy.
type CachingWalker struct {
cache []Process
cacheLock sync.RWMutex
source Walker
}
// NewCachingWalker returns a new CachingWalker
func NewCachingWalker(source Walker) *CachingWalker {
return &CachingWalker{source: source}
}
// Walk walks a cached copy of process list
func (c *CachingWalker) Walk(f func(Process)) error {
c.cacheLock.RLock()
defer c.cacheLock.RUnlock()
for _, p := range c.cache {
f(p)
}
return nil
}
// Tick updates cached copy of process list
func (c *CachingWalker) Tick() error {
newCache := []Process{}
err := c.source.Walk(func(p Process) {
newCache = append(newCache, p)
})
if err != nil {
return err
}
c.cacheLock.Lock()
defer c.cacheLock.Unlock()
c.cache = newCache
return nil
}

View File

@@ -1,79 +0,0 @@
package process
import (
"bytes"
"io/ioutil"
"path"
"strconv"
"strings"
)
// Hooks exposed for mocking
var (
ReadDir = ioutil.ReadDir
ReadFile = ioutil.ReadFile
)
type walker struct {
procRoot string
}
// NewWalker creates a new process Walker.
func NewWalker(procRoot string) Walker {
return &walker{procRoot: procRoot}
}
// Walk walks the supplied directory (expecting it to look like /proc)
// and marshalls the files into instances of Process, which it then
// passes one-by-one to the supplied function. Walk is only made public
// so that is can be tested.
func (w *walker) Walk(f func(Process)) error {
dirEntries, err := ReadDir(w.procRoot)
if err != nil {
return err
}
for _, dirEntry := range dirEntries {
filename := dirEntry.Name()
pid, err := strconv.Atoi(filename)
if err != nil {
continue
}
stat, err := ReadFile(path.Join(w.procRoot, filename, "stat"))
if err != nil {
continue
}
splits := strings.Fields(string(stat))
ppid, err := strconv.Atoi(splits[3])
if err != nil {
return err
}
threads, err := strconv.Atoi(splits[19])
if err != nil {
return err
}
cmdline := ""
if cmdlineBuf, err := ReadFile(path.Join(w.procRoot, filename, "cmdline")); err == nil {
cmdlineBuf = bytes.Replace(cmdlineBuf, []byte{'\000'}, []byte{' '}, -1)
cmdline = string(cmdlineBuf)
}
comm := "(unknown)"
if commBuf, err := ReadFile(path.Join(w.procRoot, filename, "comm")); err == nil {
comm = strings.TrimSpace(string(commBuf))
}
f(Process{
PID: pid,
PPID: ppid,
Comm: comm,
Cmdline: cmdline,
Threads: threads,
})
}
return nil
}

View File

@@ -1,90 +0,0 @@
package process_test
import (
"fmt"
"os"
"reflect"
"strconv"
"strings"
"testing"
"time"
"github.com/weaveworks/scope/probe/process"
"github.com/weaveworks/scope/test"
)
type mockProcess struct {
name, comm, cmdline string
}
func (p mockProcess) Name() string { return p.name }
func (p mockProcess) Size() int64 { return 0 }
func (p mockProcess) Mode() os.FileMode { return 0 }
func (p mockProcess) ModTime() time.Time { return time.Now() }
func (p mockProcess) IsDir() bool { return true }
func (p mockProcess) Sys() interface{} { return nil }
func TestWalker(t *testing.T) {
oldReadDir, oldReadFile := process.ReadDir, process.ReadFile
defer func() {
process.ReadDir = oldReadDir
process.ReadFile = oldReadFile
}()
processes := map[string]mockProcess{
"3": {name: "3", comm: "curl\n", cmdline: "curl\000google.com"},
"2": {name: "2", comm: "bash\n"},
"4": {name: "4", comm: "apache\n"},
"notapid": {name: "notapid"},
"1": {name: "1", comm: "init\n"},
}
process.ReadDir = func(path string) ([]os.FileInfo, error) {
result := []os.FileInfo{}
for _, p := range processes {
result = append(result, p)
}
return result, nil
}
process.ReadFile = func(path string) ([]byte, error) {
splits := strings.Split(path, "/")
pid := splits[len(splits)-2]
process, ok := processes[pid]
if !ok {
return nil, fmt.Errorf("not found")
}
file := splits[len(splits)-1]
switch file {
case "comm":
return []byte(process.comm), nil
case "stat":
pid, _ := strconv.Atoi(splits[len(splits)-2])
parent := pid - 1
return []byte(fmt.Sprintf("%d na R %d 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1", pid, parent)), nil
case "cmdline":
return []byte(process.cmdline), nil
}
return nil, fmt.Errorf("not found")
}
want := map[int]process.Process{
3: {PID: 3, PPID: 2, Comm: "curl", Cmdline: "curl google.com", Threads: 1},
2: {PID: 2, PPID: 1, Comm: "bash", Cmdline: "", Threads: 1},
4: {PID: 4, PPID: 3, Comm: "apache", Cmdline: "", Threads: 1},
1: {PID: 1, PPID: 0, Comm: "init", Cmdline: "", Threads: 1},
}
have := map[int]process.Process{}
walker := process.NewWalker("unused")
err := walker.Walk(func(p process.Process) {
have[p.PID] = p
})
if err != nil || !reflect.DeepEqual(want, have) {
t.Errorf("%v (%v)", test.Diff(want, have), err)
}
}

View File

@@ -1,66 +0,0 @@
package process_test
import (
"reflect"
"testing"
"github.com/weaveworks/scope/probe/process"
"github.com/weaveworks/scope/test"
)
func TestBasicWalk(t *testing.T) {
var (
procRoot = "/proc"
procFunc = func(process.Process) {}
)
if err := process.NewWalker(procRoot).Walk(procFunc); err != nil {
t.Fatal(err)
}
}
func TestCache(t *testing.T) {
processes := []process.Process{
{PID: 1, PPID: 0, Comm: "init"},
{PID: 2, PPID: 1, Comm: "bash"},
{PID: 3, PPID: 1, Comm: "apache", Threads: 2},
{PID: 4, PPID: 2, Comm: "ping", Cmdline: "ping foo.bar.local"},
}
walker := &mockWalker{
processes: processes,
}
cachingWalker := process.NewCachingWalker(walker)
err := cachingWalker.Tick()
if err != nil {
t.Fatal(err)
}
have, err := all(cachingWalker)
if err != nil || !reflect.DeepEqual(processes, have) {
t.Errorf("%v (%v)", test.Diff(processes, have), err)
}
walker.processes = []process.Process{}
have, err = all(cachingWalker)
if err != nil || !reflect.DeepEqual(processes, have) {
t.Errorf("%v (%v)", test.Diff(processes, have), err)
}
err = cachingWalker.Tick()
if err != nil {
t.Fatal(err)
}
have, err = all(cachingWalker)
want := []process.Process{}
if err != nil || !reflect.DeepEqual(want, have) {
t.Errorf("%v (%v)", test.Diff(want, have), err)
}
}
func all(w process.Walker) ([]process.Process, error) {
all := []process.Process{}
err := w.Walk(func(p process.Process) {
all = append(all, p)
})
return all, err
}