mirror of
https://github.com/mudler/LocalAI.git
synced 2026-09-29 09:35:02 -04:00
Ettore Di Giacinto
e61bd2d78e
ResourceExhausted is retried and trips the target, which is right for a rate limit or an out-of-memory backend. A payload over the gRPC message cap is ResourceExhausted too, but every target rejects it the same way, so it tripped the whole chain. Classify it as a request error. Assisted-by: Claude:claude-opus-5-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
59 lines
3.3 KiB
Go
59 lines
3.3 KiB
Go
package failover
|
|
|
|
import (
|
|
"context"
|
|
"errors"
|
|
"fmt"
|
|
"net/http"
|
|
|
|
"github.com/labstack/echo/v4"
|
|
. "github.com/onsi/ginkgo/v2"
|
|
. "github.com/onsi/gomega"
|
|
"google.golang.org/grpc/codes"
|
|
grpcstatus "google.golang.org/grpc/status"
|
|
)
|
|
|
|
var _ = DescribeTable("IsRetryable",
|
|
func(err error, status int, want bool) {
|
|
Expect(IsRetryable(err, status)).To(Equal(want))
|
|
},
|
|
Entry("nil error, no status", nil, 0, false),
|
|
Entry("held 503", nil, http.StatusServiceUnavailable, true),
|
|
Entry("held 500", nil, http.StatusInternalServerError, true),
|
|
Entry("held 501", nil, http.StatusNotImplemented, false),
|
|
Entry("client cancel", context.Canceled, 0, false),
|
|
Entry("wrapped client cancel", fmt.Errorf("predict: %w", context.Canceled), 0, false),
|
|
Entry("deadline", context.DeadlineExceeded, 0, true),
|
|
Entry("echo 502", echo.NewHTTPError(http.StatusBadGateway, "x"), 0, true),
|
|
Entry("echo 400", echo.NewHTTPError(http.StatusBadRequest, "x"), 0, false),
|
|
Entry("echo 404", echo.NewHTTPError(http.StatusNotFound, "x"), 0, false),
|
|
Entry("grpc unavailable", grpcstatus.Error(codes.Unavailable, "x"), 0, true),
|
|
Entry("grpc internal", grpcstatus.Error(codes.Internal, "x"), 0, true),
|
|
Entry("grpc deadline", grpcstatus.Error(codes.DeadlineExceeded, "x"), 0, true),
|
|
Entry("grpc unknown", grpcstatus.Error(codes.Unknown, "x"), 0, true),
|
|
Entry("grpc invalid argument", grpcstatus.Error(codes.InvalidArgument, "x"), 0, false),
|
|
Entry("grpc resource exhausted (rate limit, OOM)", grpcstatus.Error(codes.ResourceExhausted, "x"), 0, true),
|
|
Entry("cloud-proxy upstream 503", errors.New("cloud-proxy: upstream 503: no healthy nodes"), 0, true),
|
|
Entry("cloud-proxy upstream 429 stays 4xx", errors.New("cloud-proxy: upstream 429: slow down"), 0, false),
|
|
Entry("context overflow", errors.New("the request exceeds the available context size"), 0, false),
|
|
Entry("grpc received message over the size limit", grpcstatus.Error(codes.ResourceExhausted, "grpc: received message larger than max (60000000 vs. 52428800)"), 0, false),
|
|
Entry("grpc sent message over the size limit", grpcstatus.Error(codes.ResourceExhausted, "grpc: trying to send message larger than max (60000000 vs. 52428800)"), 0, false),
|
|
Entry("embeddings over the gRPC message limit", fmt.Errorf("conversation too long for per-token embeddings (gRPC message limit exceeded): %w",
|
|
grpcstatus.Error(codes.ResourceExhausted, "grpc: received message larger than max (1 vs. 0)")), 0, false),
|
|
Entry("dial error", errors.New("dial tcp 10.0.0.1:8080: connect: connection refused"), 0, true),
|
|
)
|
|
|
|
var _ = DescribeTable("IsCapabilityGap",
|
|
func(err error, want bool) {
|
|
Expect(IsCapabilityGap(err)).To(Equal(want))
|
|
},
|
|
Entry("nil error", nil, false),
|
|
Entry("grpc unimplemented", grpcstatus.Error(codes.Unimplemented, "x"), true),
|
|
Entry("wrapped grpc unimplemented", fmt.Errorf("call: %w", grpcstatus.Error(codes.Unimplemented, "x")), true),
|
|
Entry("echo 501 (a handler that mapped Unimplemented)", echo.NewHTTPError(http.StatusNotImplemented, "x"), true),
|
|
Entry("echo 500 is a failure, not a gap", echo.NewHTTPError(http.StatusInternalServerError, "x"), false),
|
|
Entry("grpc resource exhausted is a failure, not a gap", grpcstatus.Error(codes.ResourceExhausted, "x"), false),
|
|
Entry("grpc unavailable", grpcstatus.Error(codes.Unavailable, "x"), false),
|
|
Entry("plain error", errors.New("boom"), false),
|
|
)
|