fix: Wrap init segcore tracing with golang timeout (#33494)

See also #33483

Wrap `C.InitTrace` & `C.SetTrace` with timeout preventing otlp
initializtion hangs forever when endpoint is not set correctly

---------

Signed-off-by: Congqi Xia <congqi.xia@zilliz.com>
pull/33570/head
congqixia 2024-06-03 19:25:51 +08:00 committed by GitHub
parent 34c6a989ab
commit 2b285e5573
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 66 additions and 9 deletions

View File

@ -55,13 +55,13 @@ initTelemetry(const TraceConfig& cfg) {
opts.transport_format = jaeger::TransportFormat::kThriftHttp;
opts.endpoint = cfg.jaegerURL;
exporter = jaeger::JaegerExporterFactory::Create(opts);
LOG_INFO("init jaeger exporter, endpoint:", opts.endpoint);
LOG_INFO("init jaeger exporter, endpoint: {}", opts.endpoint);
} else if (cfg.exporter == "otlp") {
auto opts = otlp::OtlpGrpcExporterOptions{};
opts.endpoint = cfg.otlpEndpoint;
opts.use_ssl_credentials = cfg.oltpSecure;
exporter = otlp::OtlpGrpcExporterFactory::Create(opts);
LOG_INFO("init otlp exporter, endpoint:", opts.endpoint);
LOG_INFO("init otlp exporter, endpoint: {}", opts.endpoint);
} else {
LOG_INFO("Empty Trace");
enable_trace = false;

View File

@ -29,6 +29,7 @@ import "C"
import (
"fmt"
"time"
"unsafe"
"github.com/cockroachdb/errors"
@ -61,7 +62,13 @@ func InitTraceConfig(params *paramtable.ComponentParam) {
otlpEndpoint: endpoint,
nodeID: nodeID,
}
C.InitTrace(&config)
// oltp grpc may hangs forever, add timeout logic at go side
timeout := params.TraceCfg.InitTimeoutSeconds.GetAsDuration(time.Second)
callWithTimeout(func() {
C.InitTrace(&config)
}, func() {
panic("init segcore tracing timeout, See issue #33483")
}, timeout)
}
func ResetTraceConfig(params *paramtable.ComponentParam) {
@ -81,7 +88,31 @@ func ResetTraceConfig(params *paramtable.ComponentParam) {
otlpEndpoint: endpoint,
nodeID: nodeID,
}
C.SetTrace(&config)
// oltp grpc may hangs forever, add timeout logic at go side
timeout := params.TraceCfg.InitTimeoutSeconds.GetAsDuration(time.Second)
callWithTimeout(func() {
C.SetTrace(&config)
}, func() {
panic("set segcore tracing timeout, See issue #33483")
}, timeout)
}
func callWithTimeout(fn func(), timeoutHandler func(), timeout time.Duration) {
if timeout > 0 {
ch := make(chan struct{})
go func() {
defer close(ch)
fn()
}()
select {
case <-ch:
case <-time.After(timeout):
timeoutHandler()
}
} else {
fn()
}
}
func InitRemoteChunkManager(params *paramtable.ComponentParam) error {

View File

@ -19,6 +19,8 @@ package initcore
import (
"testing"
"github.com/stretchr/testify/assert"
"github.com/milvus-io/milvus/pkg/util/paramtable"
)
@ -29,3 +31,17 @@ func TestTracer(t *testing.T) {
paramtable.Get().Save(paramtable.Get().TraceCfg.Exporter.Key, "stdout")
ResetTraceConfig(paramtable.Get())
}
func TestOtlpHang(t *testing.T) {
paramtable.Init()
InitTraceConfig(paramtable.Get())
paramtable.Get().Save(paramtable.Get().TraceCfg.Exporter.Key, "otlp")
paramtable.Get().Save(paramtable.Get().TraceCfg.InitTimeoutSeconds.Key, "1")
defer paramtable.Get().Reset(paramtable.Get().TraceCfg.Exporter.Key)
defer paramtable.Get().Reset(paramtable.Get().TraceCfg.InitTimeoutSeconds.Key)
assert.Panics(t, func() {
ResetTraceConfig(paramtable.Get())
})
}

View File

@ -787,11 +787,12 @@ func (t *gpuConfig) init(base *BaseTable) {
}
type traceConfig struct {
Exporter ParamItem `refreshable:"false"`
SampleFraction ParamItem `refreshable:"false"`
JaegerURL ParamItem `refreshable:"false"`
OtlpEndpoint ParamItem `refreshable:"false"`
OtlpSecure ParamItem `refreshable:"false"`
Exporter ParamItem `refreshable:"false"`
SampleFraction ParamItem `refreshable:"false"`
JaegerURL ParamItem `refreshable:"false"`
OtlpEndpoint ParamItem `refreshable:"false"`
OtlpSecure ParamItem `refreshable:"false"`
InitTimeoutSeconds ParamItem `refreshable:"false"`
}
func (t *traceConfig) init(base *BaseTable) {
@ -839,6 +840,15 @@ Fractions >= 1 will always sample. Fractions < 0 are treated as zero.`,
Export: true,
}
t.OtlpSecure.Init(base.mgr)
t.InitTimeoutSeconds = ParamItem{
Key: "trace.initTimeoutSeconds",
Version: "2.4.4",
DefaultValue: "10",
Export: true,
Doc: "segcore initialization timeout in seconds, preventing otlp grpc hangs forever",
}
t.InitTimeoutSeconds.Init(base.mgr)
}
type logConfig struct {