Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

Commitb68a46a

Browse files
committed
feat: integrate agentexec pkg
- Removes the legacy process priority implementation and replaces current calls to os/exec and pty.Command with agentexec variants.- Adds linting ruleguards to discourage future imports of these packages.
1 parentbbc549d commitb68a46a

File tree

19 files changed

+116
-1173
lines changed

19 files changed

+116
-1173
lines changed

‎agent/agent.go‎

Lines changed: 2 additions & 266 deletions
Original file line numberDiff line numberDiff line change
@@ -12,8 +12,6 @@ import (
1212
"os"
1313
"os/user"
1414
"path/filepath"
15-
"runtime"
16-
"runtime/debug"
1715
"sort"
1816
"strconv"
1917
"strings"
@@ -35,7 +33,6 @@ import (
3533
"tailscale.com/util/clientmetric"
3634

3735
"cdr.dev/slog"
38-
"github.com/coder/coder/v2/agent/agentproc"
3936
"github.com/coder/coder/v2/agent/agentscripts"
4037
"github.com/coder/coder/v2/agent/agentssh"
4138
"github.com/coder/coder/v2/agent/proto"
@@ -82,12 +79,7 @@ type Options struct {
8279
PrometheusRegistry*prometheus.Registry
8380
ReportMetadataInterval time.Duration
8481
ServiceBannerRefreshInterval time.Duration
85-
Syscaller agentproc.Syscaller
86-
// ModifiedProcesses is used for testing process priority management.
87-
ModifiedProcesseschan []*agentproc.Process
88-
// ProcessManagementTick is used for testing process priority management.
89-
ProcessManagementTick<-chan time.Time
90-
BlockFileTransferbool
82+
BlockFileTransferbool
9183
}
9284

9385
typeClientinterface {
@@ -147,10 +139,6 @@ func New(options Options) Agent {
147139
prometheusRegistry=prometheus.NewRegistry()
148140
}
149141

150-
ifoptions.Syscaller==nil {
151-
options.Syscaller=agentproc.NewSyscaller()
152-
}
153-
154142
hardCtx,hardCancel:=context.WithCancel(context.Background())
155143
gracefulCtx,gracefulCancel:=context.WithCancel(hardCtx)
156144
a:=&agent{
@@ -178,9 +166,6 @@ func New(options Options) Agent {
178166
announcementBannersRefreshInterval:options.ServiceBannerRefreshInterval,
179167
sshMaxTimeout:options.SSHMaxTimeout,
180168
subsystems:options.Subsystems,
181-
syscaller:options.Syscaller,
182-
modifiedProcs:options.ModifiedProcesses,
183-
processManagementTick:options.ProcessManagementTick,
184169
logSender:agentsdk.NewLogSender(options.Logger),
185170
blockFileTransfer:options.BlockFileTransfer,
186171

@@ -253,13 +238,7 @@ type agent struct {
253238
prometheusRegistry*prometheus.Registry
254239
// metrics are prometheus registered metrics that will be collected and
255240
// labeled in Coder with the agent + workspace.
256-
metrics*agentMetrics
257-
syscaller agentproc.Syscaller
258-
259-
// modifiedProcs is used for testing process priority management.
260-
modifiedProcschan []*agentproc.Process
261-
// processManagementTick is used for testing process priority management.
262-
processManagementTick<-chan time.Time
241+
metrics*agentMetrics
263242
}
264243

265244
func (a*agent)TailnetConn()*tailnet.Conn {
@@ -308,8 +287,6 @@ func (a *agent) init() {
308287
// may be happening, but regardless after the intermittent
309288
// failure, you'll want the agent to reconnect.
310289
func (a*agent)runLoop() {
311-
goa.manageProcessPriorityUntilGracefulShutdown()
312-
313290
// need to keep retrying up to the hardCtx so that we can send graceful shutdown-related
314291
// messages.
315292
ctx:=a.hardCtx
@@ -1443,162 +1420,6 @@ func (a *agent) Collect(ctx context.Context, networkStats map[netlogtype.Connect
14431420
returnstats
14441421
}
14451422

1446-
varprioritizedProcs= []string{"coder agent"}
1447-
1448-
func (a*agent)manageProcessPriorityUntilGracefulShutdown() {
1449-
// process priority can stop as soon as we are gracefully shutting down
1450-
ctx:=a.gracefulCtx
1451-
deferfunc() {
1452-
ifr:=recover();r!=nil {
1453-
a.logger.Critical(ctx,"recovered from panic",
1454-
slog.F("panic",r),
1455-
slog.F("stack",string(debug.Stack())),
1456-
)
1457-
}
1458-
}()
1459-
1460-
ifval:=a.environmentVariables[EnvProcPrioMgmt];val==""||runtime.GOOS!="linux" {
1461-
a.logger.Debug(ctx,"process priority not enabled, agent will not manage process niceness/oom_score_adj ",
1462-
slog.F("env_var",EnvProcPrioMgmt),
1463-
slog.F("value",val),
1464-
slog.F("goos",runtime.GOOS),
1465-
)
1466-
return
1467-
}
1468-
1469-
ifa.processManagementTick==nil {
1470-
ticker:=time.NewTicker(time.Second)
1471-
deferticker.Stop()
1472-
a.processManagementTick=ticker.C
1473-
}
1474-
1475-
oomScore:=unsetOOMScore
1476-
ifscoreStr,ok:=a.environmentVariables[EnvProcOOMScore];ok {
1477-
score,err:=strconv.Atoi(strings.TrimSpace(scoreStr))
1478-
iferr==nil&&score>=-1000&&score<=1000 {
1479-
oomScore=score
1480-
}else {
1481-
a.logger.Error(ctx,"invalid oom score",
1482-
slog.F("min_value",-1000),
1483-
slog.F("max_value",1000),
1484-
slog.F("value",scoreStr),
1485-
)
1486-
}
1487-
}
1488-
1489-
debouncer:=&logDebouncer{
1490-
logger:a.logger,
1491-
messages:map[string]time.Time{},
1492-
interval:time.Minute,
1493-
}
1494-
1495-
for {
1496-
procs,err:=a.manageProcessPriority(ctx,debouncer,oomScore)
1497-
// Avoid spamming the logs too often.
1498-
iferr!=nil {
1499-
debouncer.Error(ctx,"manage process priority",
1500-
slog.Error(err),
1501-
)
1502-
}
1503-
ifa.modifiedProcs!=nil {
1504-
a.modifiedProcs<-procs
1505-
}
1506-
1507-
select {
1508-
case<-a.processManagementTick:
1509-
case<-ctx.Done():
1510-
return
1511-
}
1512-
}
1513-
}
1514-
1515-
// unsetOOMScore is set to an invalid OOM score to imply an unset value.
1516-
constunsetOOMScore=1001
1517-
1518-
func (a*agent)manageProcessPriority(ctx context.Context,debouncer*logDebouncer,oomScoreint) ([]*agentproc.Process,error) {
1519-
const (
1520-
niceness=10
1521-
)
1522-
1523-
// We fetch the agent score each time because it's possible someone updates the
1524-
// value after it is started.
1525-
agentScore,err:=a.getAgentOOMScore()
1526-
iferr!=nil {
1527-
agentScore=unsetOOMScore
1528-
}
1529-
ifoomScore==unsetOOMScore&&agentScore!=unsetOOMScore {
1530-
// If the child score has not been explicitly specified we should
1531-
// set it to a score relative to the agent score.
1532-
oomScore=childOOMScore(agentScore)
1533-
}
1534-
1535-
procs,err:=agentproc.List(a.filesystem,a.syscaller)
1536-
iferr!=nil {
1537-
returnnil,xerrors.Errorf("list: %w",err)
1538-
}
1539-
1540-
modProcs:= []*agentproc.Process{}
1541-
1542-
for_,proc:=rangeprocs {
1543-
containsFn:=func(estring)bool {
1544-
contains:=strings.Contains(proc.Cmd(),e)
1545-
returncontains
1546-
}
1547-
1548-
// If the process is prioritized we should adjust
1549-
// it's oom_score_adj and avoid lowering its niceness.
1550-
ifslices.ContainsFunc(prioritizedProcs,containsFn) {
1551-
continue
1552-
}
1553-
1554-
score,niceErr:=proc.Niceness(a.syscaller)
1555-
ifniceErr!=nil&&!isBenignProcessErr(niceErr) {
1556-
debouncer.Warn(ctx,"unable to get proc niceness",
1557-
slog.F("cmd",proc.Cmd()),
1558-
slog.F("pid",proc.PID),
1559-
slog.Error(niceErr),
1560-
)
1561-
}
1562-
1563-
// We only want processes that don't have a nice value set
1564-
// so we don't override user nice values.
1565-
// Getpriority actually returns priority for the nice value
1566-
// which is niceness + 20, so here 20 = a niceness of 0 (aka unset).
1567-
ifscore!=20 {
1568-
// We don't log here since it can get spammy
1569-
continue
1570-
}
1571-
1572-
ifniceErr==nil {
1573-
err:=proc.SetNiceness(a.syscaller,niceness)
1574-
iferr!=nil&&!isBenignProcessErr(err) {
1575-
debouncer.Warn(ctx,"unable to set proc niceness",
1576-
slog.F("cmd",proc.Cmd()),
1577-
slog.F("pid",proc.PID),
1578-
slog.F("niceness",niceness),
1579-
slog.Error(err),
1580-
)
1581-
}
1582-
}
1583-
1584-
// If the oom score is valid and it's not already set and isn't a custom value set by another process then it's ok to update it.
1585-
ifoomScore!=unsetOOMScore&&oomScore!=proc.OOMScoreAdj&&!isCustomOOMScore(agentScore,proc) {
1586-
oomScoreStr:=strconv.Itoa(oomScore)
1587-
err:=afero.WriteFile(a.filesystem,fmt.Sprintf("/proc/%d/oom_score_adj",proc.PID), []byte(oomScoreStr),0o644)
1588-
iferr!=nil&&!isBenignProcessErr(err) {
1589-
debouncer.Warn(ctx,"unable to set oom_score_adj",
1590-
slog.F("cmd",proc.Cmd()),
1591-
slog.F("pid",proc.PID),
1592-
slog.F("score",oomScoreStr),
1593-
slog.Error(err),
1594-
)
1595-
}
1596-
}
1597-
modProcs=append(modProcs,proc)
1598-
}
1599-
returnmodProcs,nil
1600-
}
1601-
16021423
// isClosed returns whether the API is closed or not.
16031424
func (a*agent)isClosed()bool {
16041425
returna.hardCtx.Err()!=nil
@@ -1992,88 +1813,3 @@ func PrometheusMetricsHandler(prometheusRegistry *prometheus.Registry, logger sl
19921813
}
19931814
})
19941815
}
1995-
1996-
// childOOMScore returns the oom_score_adj for a child process. It is based
1997-
// on the oom_score_adj of the agent process.
1998-
funcchildOOMScore(agentScoreint)int {
1999-
// If the agent has a negative oom_score_adj, we set the child to 0
2000-
// so it's treated like every other process.
2001-
ifagentScore<0 {
2002-
return0
2003-
}
2004-
2005-
// If the agent is already almost at the maximum then set it to the max.
2006-
ifagentScore>=998 {
2007-
return1000
2008-
}
2009-
2010-
// If the agent oom_score_adj is >=0, we set the child to slightly
2011-
// less than the maximum. If users want a different score they set it
2012-
// directly.
2013-
return998
2014-
}
2015-
2016-
func (a*agent)getAgentOOMScore() (int,error) {
2017-
scoreStr,err:=afero.ReadFile(a.filesystem,"/proc/self/oom_score_adj")
2018-
iferr!=nil {
2019-
return0,xerrors.Errorf("read file: %w",err)
2020-
}
2021-
2022-
score,err:=strconv.Atoi(strings.TrimSpace(string(scoreStr)))
2023-
iferr!=nil {
2024-
return0,xerrors.Errorf("parse int: %w",err)
2025-
}
2026-
2027-
returnscore,nil
2028-
}
2029-
2030-
// isCustomOOMScore checks to see if the oom_score_adj is not a value that would
2031-
// originate from an agent-spawned process.
2032-
funcisCustomOOMScore(agentScoreint,process*agentproc.Process)bool {
2033-
score:=process.OOMScoreAdj
2034-
returnagentScore!=score&&score!=1000&&score!=0&&score!=998
2035-
}
2036-
2037-
// logDebouncer skips writing a log for a particular message if
2038-
// it's been emitted within the given interval duration.
2039-
// It's a shoddy implementation used in one spot that should be replaced at
2040-
// some point.
2041-
typelogDebouncerstruct {
2042-
logger slog.Logger
2043-
messagesmap[string]time.Time
2044-
interval time.Duration
2045-
}
2046-
2047-
func (l*logDebouncer)Warn(ctx context.Context,msgstring,fields...any) {
2048-
l.log(ctx,slog.LevelWarn,msg,fields...)
2049-
}
2050-
2051-
func (l*logDebouncer)Error(ctx context.Context,msgstring,fields...any) {
2052-
l.log(ctx,slog.LevelError,msg,fields...)
2053-
}
2054-
2055-
func (l*logDebouncer)log(ctx context.Context,level slog.Level,msgstring,fields...any) {
2056-
// This (bad) implementation assumes you wouldn't reuse the same msg
2057-
// for different levels.
2058-
iflast,ok:=l.messages[msg];ok&&time.Since(last)<l.interval {
2059-
return
2060-
}
2061-
switchlevel {
2062-
caseslog.LevelWarn:
2063-
l.logger.Warn(ctx,msg,fields...)
2064-
caseslog.LevelError:
2065-
l.logger.Error(ctx,msg,fields...)
2066-
}
2067-
l.messages[msg]=time.Now()
2068-
}
2069-
2070-
funcisBenignProcessErr(errerror)bool {
2071-
returnerr!=nil&&
2072-
(xerrors.Is(err,os.ErrNotExist)||
2073-
xerrors.Is(err,os.ErrPermission)||
2074-
isNoSuchProcessErr(err))
2075-
}
2076-
2077-
funcisNoSuchProcessErr(errerror)bool {
2078-
returnerr!=nil&&strings.Contains(err.Error(),"no such process")
2079-
}

0 commit comments

Comments
 (0)
[8]ページ先頭
©2009-2025 Movatter.jp