feat(coderd): add overload protection with rate limiting and concurrency control (#21161)

## Summary

This adds configurable overload protection to the AI Bridge daemon to
prevent the server from being overwhelmed during periods of high load.

Partially addresses coder/internal#1153 (rate limits and concurrency
control; circuit breakers are deferred to a follow-up).

## New Configuration Options

| Option | Environment Variable | Description | Default |
|--------|---------------------|-------------|---------|
| `--aibridge-max-concurrency` | `CODER_AIBRIDGE_MAX_CONCURRENCY` |
Maximum number of concurrent AI Bridge requests. Set to 0 to disable
(unlimited). | `0` |
| `--aibridge-rate-limit` | `CODER_AIBRIDGE_RATE_LIMIT` | Maximum number
of AI Bridge requests per second. Set to 0 to disable rate limiting. |
`0` |

## Behavior

When limits are exceeded:
- **Concurrency limit**: Returns HTTP `503 Service Unavailable` with
message "AI Bridge is currently at capacity. Please try again later."
- **Rate limit**: Returns HTTP `429 Too Many Requests` with
`Retry-After` header.

Both protections are optional and disabled by default (0 values).

## Implementation

The overload protection is implemented as reusable middleware in
`coderd/httpmw/ratelimit.go`:

1. **`RateLimitByAuthToken`**: Per-user rate limiting that uses
`APITokenFromRequest` to extract the authentication token, with fallback
to `X-Api-Key` header for AI provider compatibility (e.g., Anthropic).
Falls back to IP-based rate limiting if no token is present. Includes
`Retry-After` header for backpressure signaling.
2. **`ConcurrencyLimit`**: Uses an atomic counter to track in-flight
requests and reject when at capacity.

The middleware is applied in `enterprise/coderd/aibridge.go` via
`r.Group` in the following order:
1. Concurrency check (faster rejection for load shedding)
2. Rate limit check

**Note**: Rate limiting currently applies to all AI Bridge requests,
including pass-through requests. Ideally only actual interceptions
should count, but this would require changes in the aibridge library.

## Testing

Added comprehensive tests for:
- Rate limiting by auth token (Bearer token, X-Api-Key, no token
fallback to IP)
- Different tokens not rate limited against each other
- Disabled when limit is zero
- Retry-After header is set on 429 responses
- Concurrency limiting (allows within limit, rejects over limit,
disabled when zero)
This commit is contained in:
Kacper Sawicki
2025-12-11 16:38:54 +01:00
committed by GitHub
parent 8ead6f795d
commit 6f86f67754
17 changed files with 568 additions and 37 deletions
+33 -18
View File
@@ -5,6 +5,7 @@ import (
"fmt"
"net/http"
"strings"
"time"
"github.com/go-chi/chi/v5"
"github.com/google/uuid"
@@ -23,10 +24,19 @@ import (
const (
maxListInterceptionsLimit = 1000
defaultListInterceptionsLimit = 100
// aiBridgeRateLimitWindow is the fixed duration for rate limiting AI Bridge
// requests. This is hardcoded to keep configuration simple.
aiBridgeRateLimitWindow = time.Second
)
// aibridgeHandler handles all aibridged-related endpoints.
func aibridgeHandler(api *API, middlewares ...func(http.Handler) http.Handler) func(r chi.Router) {
// Build the overload protection middleware chain for the aibridged handler.
// These limits are applied per-replica.
bridgeCfg := api.DeploymentValues.AI.BridgeConfig
concurrencyLimiter := httpmw.ConcurrencyLimit(bridgeCfg.MaxConcurrency.Value(), "AI Bridge")
rateLimiter := httpmw.RateLimitByAuthToken(int(bridgeCfg.RateLimit.Value()), aiBridgeRateLimitWindow)
return func(r chi.Router) {
r.Use(api.RequireFeatureMW(codersdk.FeatureAIBridge))
r.Group(func(r chi.Router) {
@@ -34,25 +44,30 @@ func aibridgeHandler(api *API, middlewares ...func(http.Handler) http.Handler) f
r.Get("/interceptions", api.aiBridgeListInterceptions)
})
// This is a bit funky but since aibridge only exposes a HTTP
// handler, this is how it has to be.
r.HandleFunc("/*", func(rw http.ResponseWriter, r *http.Request) {
if api.aibridgedHandler == nil {
httpapi.Write(r.Context(), rw, http.StatusNotFound, codersdk.Response{
Message: "aibridged handler not mounted",
})
return
}
// Strip either the experimental or stable prefix.
// TODO: experimental route is deprecated and must be removed with Beta.
prefixes := []string{"/api/experimental/aibridge", "/api/v2/aibridge"}
for _, prefix := range prefixes {
if strings.Contains(r.URL.String(), prefix) {
http.StripPrefix(prefix, api.aibridgedHandler).ServeHTTP(rw, r)
break
// Apply overload protection middleware to the aibridged handler.
// Concurrency limit is checked first for faster rejection under load.
r.Group(func(r chi.Router) {
r.Use(concurrencyLimiter, rateLimiter)
// This is a bit funky but since aibridge only exposes a HTTP
// handler, this is how it has to be.
r.HandleFunc("/*", func(rw http.ResponseWriter, r *http.Request) {
if api.aibridgedHandler == nil {
httpapi.Write(r.Context(), rw, http.StatusNotFound, codersdk.Response{
Message: "aibridged handler not mounted",
})
return
}
}
// Strip either the experimental or stable prefix.
// TODO: experimental route is deprecated and must be removed with Beta.
prefixes := []string{"/api/experimental/aibridge", "/api/v2/aibridge"}
for _, prefix := range prefixes {
if strings.Contains(r.URL.String(), prefix) {
http.StripPrefix(prefix, api.aibridgedHandler).ServeHTTP(rw, r)
break
}
}
})
})
}
}
+132
View File
@@ -703,3 +703,135 @@ func TestAIBridgeRouting(t *testing.T) {
})
}
}
func TestAIBridgeRateLimiting(t *testing.T) {
t.Parallel()
dv := coderdtest.DeploymentValues(t)
// Set a low rate limit for testing.
dv.AI.BridgeConfig.RateLimit = 2
client, closer, api, _ := coderdenttest.NewWithAPI(t, &coderdenttest.Options{
Options: &coderdtest.Options{
DeploymentValues: dv,
},
LicenseOptions: &coderdenttest.LicenseOptions{
Features: license.Features{
codersdk.FeatureAIBridge: 1,
},
},
})
t.Cleanup(func() {
_ = closer.Close()
})
// Register a simple test handler.
testHandler := http.HandlerFunc(func(rw http.ResponseWriter, r *http.Request) {
rw.WriteHeader(http.StatusOK)
})
api.RegisterInMemoryAIBridgedHTTPHandler(testHandler)
ctx := testutil.Context(t, testutil.WaitLong)
httpClient := &http.Client{}
url := client.URL.String() + "/api/v2/aibridge/test"
// Make requests up to the limit - should succeed.
for range 2 {
req, err := http.NewRequestWithContext(ctx, http.MethodPost, url, nil)
require.NoError(t, err)
req.Header.Set(codersdk.SessionTokenHeader, client.SessionToken())
resp, err := httpClient.Do(req)
require.NoError(t, err)
_ = resp.Body.Close()
require.Equal(t, http.StatusOK, resp.StatusCode)
}
// Next request should be rate limited.
req, err := http.NewRequestWithContext(ctx, http.MethodPost, url, nil)
require.NoError(t, err)
req.Header.Set(codersdk.SessionTokenHeader, client.SessionToken())
resp, err := httpClient.Do(req)
require.NoError(t, err)
defer resp.Body.Close()
require.Equal(t, http.StatusTooManyRequests, resp.StatusCode)
require.NotEmpty(t, resp.Header.Get("Retry-After"))
}
func TestAIBridgeConcurrencyLimiting(t *testing.T) {
t.Parallel()
dv := coderdtest.DeploymentValues(t)
// Set a low concurrency limit for testing.
dv.AI.BridgeConfig.MaxConcurrency = 1
client, closer, api, _ := coderdenttest.NewWithAPI(t, &coderdenttest.Options{
Options: &coderdtest.Options{
DeploymentValues: dv,
},
LicenseOptions: &coderdenttest.LicenseOptions{
Features: license.Features{
codersdk.FeatureAIBridge: 1,
},
},
})
t.Cleanup(func() {
_ = closer.Close()
})
// Register a handler that blocks until signaled.
started := make(chan struct{})
unblock := make(chan struct{})
testHandler := http.HandlerFunc(func(rw http.ResponseWriter, r *http.Request) {
started <- struct{}{}
<-unblock
rw.WriteHeader(http.StatusOK)
})
api.RegisterInMemoryAIBridgedHTTPHandler(testHandler)
ctx := testutil.Context(t, testutil.WaitLong)
httpClient := &http.Client{}
url := client.URL.String() + "/api/v2/aibridge/test"
// Start a request that will block.
done := make(chan struct{})
go func() {
defer close(done)
req, err := http.NewRequestWithContext(ctx, http.MethodPost, url, nil)
if err != nil {
return
}
req.Header.Set(codersdk.SessionTokenHeader, client.SessionToken())
resp, err := httpClient.Do(req)
if err == nil {
_ = resp.Body.Close()
}
}()
// Wait for the first request to start processing.
select {
case <-started:
case <-ctx.Done():
t.Fatal("timed out waiting for first request to start")
}
// Second request should be rejected with 503.
req, err := http.NewRequestWithContext(ctx, http.MethodPost, url, nil)
require.NoError(t, err)
req.Header.Set(codersdk.SessionTokenHeader, client.SessionToken())
resp, err := httpClient.Do(req)
require.NoError(t, err)
defer resp.Body.Close()
require.Equal(t, http.StatusServiceUnavailable, resp.StatusCode)
// Unblock the first request and wait for it to complete.
close(unblock)
select {
case <-done:
case <-ctx.Done():
t.Fatal("timed out waiting for first request to complete")
}
}