grpc resource_error ai_generated true

INTERNAL: grpc: max concurrent streams exceeded on server

ID: grpc/grpc-max-concurrent-streams-exceeded

Also available as: JSON · Markdown · 中文

80%Fix Rate

88%Confidence

1Evidence

2023-11-20First Seen

Version Compatibility

Version	Status	Introduced	Deprecated	Notes
grpc-go 1.61.0	active	—	—	—
grpc-java 1.60.0	active	—	—	—
envoy 1.27.0	active	—	—	—
nginx 1.25.0	active	—	—	—

Root Cause

Server's HTTP/2 max concurrent streams limit (default 100) is reached, and new streams are rejected.

generic

中文

服务器的 HTTP/2 最大并发流限制（默认 100）已达到，新流被拒绝。

Official Documentation

https://grpc.io/docs/guides/performance/#flow-control

Workarounds

85% success Increase the server's max concurrent streams limit. In gRPC-Go, use: import "google.golang.org/grpc" s := grpc.NewServer(grpc.MaxConcurrentStreams(200)) Or set environment variable GRPC_GO_MAX_CONCURRENT_STREAMS=200 before starting the server.
```
Increase the server's max concurrent streams limit. In gRPC-Go, use:
  import "google.golang.org/grpc"
  s := grpc.NewServer(grpc.MaxConcurrentStreams(200))
  Or set environment variable GRPC_GO_MAX_CONCURRENT_STREAMS=200 before starting the server.
```
75% success Reduce the number of long-lived streams by batching requests or using unary RPCs instead of streaming. Alternatively, add more server instances behind a load balancer to distribute streams.
```
Reduce the number of long-lived streams by batching requests or using unary RPCs instead of streaming. Alternatively, add more server instances behind a load balancer to distribute streams.
```

中文步骤

Increase the server's max concurrent streams limit. In gRPC-Go, use:
  import "google.golang.org/grpc"
  s := grpc.NewServer(grpc.MaxConcurrentStreams(200))
  Or set environment variable GRPC_GO_MAX_CONCURRENT_STREAMS=200 before starting the server.

Reduce the number of long-lived streams by batching requests or using unary RPCs instead of streaming. Alternatively, add more server instances behind a load balancer to distribute streams.

Dead Ends

Common approaches that don't work:

85% fail
The limit is on the server side per connection; client pool size doesn't affect server stream limits.
90% fail
Stream count is unrelated to message size; this does not free up stream slots.