feat: classify provider_disabled 503 as non-retryable (#25800)

Builds on top of https://github.com/coder/coder/pull/25794

Adds a new `provider_disabled` error classification in `chatd` with the
corresponding plumbing to classify it as non-retryable. Also adds a
story for how this particular error kind is displayed in the UI.
This commit is contained in:
Cian Johnston
2026-05-29 13:14:04 +01:00
committed by GitHub
parent 4144eb3c4f
commit d0a51da0a9
13 changed files with 200 additions and 64 deletions
+4 -2
View File
@@ -16498,7 +16498,8 @@ const docTemplate = `{
"auth",
"config",
"usage_limit",
"missing_key"
"missing_key",
"provider_disabled"
],
"x-enum-varnames": [
"ChatErrorKindGeneric",
@@ -16509,7 +16510,8 @@ const docTemplate = `{
"ChatErrorKindAuth",
"ChatErrorKindConfig",
"ChatErrorKindUsageLimit",
"ChatErrorKindMissingKey"
"ChatErrorKindMissingKey",
"ChatErrorKindProviderDisabled"
]
},
"codersdk.ChatFileMetadata": {
+4 -2
View File
@@ -14848,7 +14848,8 @@
"auth",
"config",
"usage_limit",
"missing_key"
"missing_key",
"provider_disabled"
],
"x-enum-varnames": [
"ChatErrorKindGeneric",
@@ -14859,7 +14860,8 @@
"ChatErrorKindAuth",
"ChatErrorKindConfig",
"ChatErrorKindUsageLimit",
"ChatErrorKindMissingKey"
"ChatErrorKindMissingKey",
"ChatErrorKindProviderDisabled"
]
},
"codersdk.ChatFileMetadata": {
+8
View File
@@ -195,6 +195,7 @@ func Classify(err error) ClassifiedError {
}
retryableHTTP2StreamReset, hasHTTP2StreamReset := classifyHTTP2StreamReset(err)
providerDisabledMatch := containsAny(lower, providerDisabledPatterns...)
deadline := errors.Is(err, context.DeadlineExceeded) || strings.Contains(lower, "context deadline exceeded")
overloadedMatch := statusCode == 529 || containsAny(lower, overloadedPatterns...)
usageLimitMatch := containsAny(lower, usageLimitPatterns...)
@@ -221,6 +222,8 @@ func Classify(err error) ClassifiedError {
// over whatever HTTP status code the provider happened to use.
// Strong auth still stays above config because bad credentials are
// the root cause when both signals appear.
// Provider-disabled must precede timeout because disabled providers
// return 503, which matches the timeout rule.
rules := []struct {
match bool
kind codersdk.ChatErrorKind
@@ -251,6 +254,11 @@ func Classify(err error) ClassifiedError {
kind: codersdk.ChatErrorKindRateLimit,
retryable: true,
},
{
match: providerDisabledMatch,
kind: codersdk.ChatErrorKindProviderDisabled,
retryable: false,
},
{
match: timeoutMatch && !configMatch,
kind: codersdk.ChatErrorKindTimeout,
+81
View File
@@ -2,6 +2,7 @@ package chaterror_test
import (
"context"
"fmt"
"io"
"net/http"
"strings"
@@ -218,6 +219,85 @@ func TestClassify(t *testing.T) {
StatusCode: 0,
},
},
// The next cases model the error that fantasy produces
// when aibridge's disabledProviderHandler returns a 503
// plain-text sentinel. Fantasy sets Title from the HTTP
// status text and Message from the response body (including
// the trailing newline written by http.Error).
{
name: "ProviderDisabled503ClassifiesAsProviderDisabled",
err: &fantasy.ProviderError{
Title: fantasy.ErrorTitleForStatusCode(http.StatusServiceUnavailable),
Message: fmt.Sprintf("%s: AI provider %q is disabled\n", codersdk.ChatErrorKindProviderDisabled, "openai"),
StatusCode: http.StatusServiceUnavailable,
},
want: chaterror.ClassifiedError{
Message: "The OpenAI provider has been disabled. Contact your Coder administrator.",
Detail: fmt.Sprintf("%s: AI provider %q is disabled", codersdk.ChatErrorKindProviderDisabled, "openai"),
Kind: codersdk.ChatErrorKindProviderDisabled,
Provider: "openai",
Retryable: false,
StatusCode: 503,
},
},
{
name: "ProviderDisabled503UnknownProvider",
err: &fantasy.ProviderError{
Title: fantasy.ErrorTitleForStatusCode(http.StatusServiceUnavailable),
Message: fmt.Sprintf("%s: AI provider %q is disabled\n", codersdk.ChatErrorKindProviderDisabled, "mycustomprovider"),
StatusCode: http.StatusServiceUnavailable,
},
want: chaterror.ClassifiedError{
Message: "The AI provider has been disabled. Contact your Coder administrator.",
Detail: fmt.Sprintf("%s: AI provider %q is disabled", codersdk.ChatErrorKindProviderDisabled, "mycustomprovider"),
Kind: codersdk.ChatErrorKindProviderDisabled,
Provider: "",
Retryable: false,
StatusCode: 503,
},
},
{
name: "ProviderDisabledPlainErrorString",
err: xerrors.New(fmt.Sprintf("%s: AI provider %q is disabled", codersdk.ChatErrorKindProviderDisabled, "anthropic")),
want: chaterror.ClassifiedError{
Message: "The Anthropic provider has been disabled. Contact your Coder administrator.",
Kind: codersdk.ChatErrorKindProviderDisabled,
Provider: "anthropic",
Retryable: false,
StatusCode: 0,
},
},
{
name: "ProviderDisabledBeatsTimeout503",
err: &fantasy.ProviderError{
Title: fantasy.ErrorTitleForStatusCode(http.StatusServiceUnavailable),
Message: fmt.Sprintf("%s: AI provider %q is disabled\n", codersdk.ChatErrorKindProviderDisabled, "google"),
StatusCode: http.StatusServiceUnavailable,
},
want: chaterror.ClassifiedError{
Message: "The Google provider has been disabled. Contact your Coder administrator.",
Detail: fmt.Sprintf("%s: AI provider %q is disabled", codersdk.ChatErrorKindProviderDisabled, "google"),
Kind: codersdk.ChatErrorKindProviderDisabled,
Provider: "google",
Retryable: false,
StatusCode: 503,
},
},
{
name: "Generic503StillClassifiesAsTimeout",
err: &fantasy.ProviderError{
Message: "service unavailable",
StatusCode: 503,
},
want: chaterror.ClassifiedError{
Message: "The AI provider is temporarily unavailable.",
Detail: "service unavailable",
Kind: codersdk.ChatErrorKindTimeout,
Provider: "",
Retryable: true,
StatusCode: 503,
},
},
}
for _, tt := range tests {
@@ -363,6 +443,7 @@ func TestClassify_PatternCoverage(t *testing.T) {
{name: "OperationInterruptedLiteral", err: "operation interrupted", wantKind: codersdk.ChatErrorKindGeneric, wantRetry: false},
{name: "Status408", err: "status 408", wantKind: codersdk.ChatErrorKindTimeout, wantRetry: true},
{name: "Status500", err: "status 500", wantKind: codersdk.ChatErrorKindGeneric, wantRetry: true},
{name: "ProviderDisabledLiteral", err: "provider_disabled", wantKind: codersdk.ChatErrorKindProviderDisabled, wantRetry: false},
}
for _, tt := range tests {
+39 -38
View File
@@ -4,6 +4,7 @@ import (
"fmt"
"strings"
stringutil "github.com/coder/coder/v2/coderd/util/strings"
"github.com/coder/coder/v2/codersdk"
)
@@ -16,60 +17,58 @@ func terminalMessage(classified ClassifiedError) string {
subject := providerSubject(classified.Provider)
switch classified.Kind {
case codersdk.ChatErrorKindOverloaded:
return fmt.Sprintf("%s is temporarily overloaded.", subject)
return stringutil.Capitalize(fmt.Sprintf("%s is temporarily overloaded.", subject))
case codersdk.ChatErrorKindRateLimit:
return fmt.Sprintf("%s is rate limiting requests.", subject)
return stringutil.Capitalize(fmt.Sprintf("%s is rate limiting requests.", subject))
case codersdk.ChatErrorKindTimeout:
if !classified.Retryable && classified.StatusCode == 0 {
return "The request timed out before it completed."
}
return fmt.Sprintf("%s is temporarily unavailable.", subject)
return stringutil.Capitalize(fmt.Sprintf("%s is temporarily unavailable.", subject))
case codersdk.ChatErrorKindStartupTimeout:
return fmt.Sprintf(
return stringutil.Capitalize(fmt.Sprintf(
"%s did not start responding in time.", subject,
)
))
case codersdk.ChatErrorKindUsageLimit:
displayName := providerDisplayName(classified.Provider)
if displayName == "" {
displayName = "the AI provider"
}
return fmt.Sprintf(
return stringutil.Capitalize(fmt.Sprintf(
"The usage quota for %s has been exceeded."+
" Check the billing and quota settings for the provider account.",
displayName,
)
subject,
))
case codersdk.ChatErrorKindAuth:
displayName := providerDisplayName(classified.Provider)
if displayName == "" {
displayName = "the AI provider"
}
return fmt.Sprintf(
"Authentication with %s failed."+
" Check the API key and permissions.",
displayName,
subject,
)
case codersdk.ChatErrorKindConfig:
return fmt.Sprintf(
return stringutil.Capitalize(fmt.Sprintf(
"%s rejected the model configuration."+
" Check the selected model and provider settings.",
subject,
)
))
case codersdk.ChatErrorKindMissingKey:
return "This conversation was started with an API key that is no longer available." +
" Send your message again to continue."
case codersdk.ChatErrorKindProviderDisabled:
displayName := providerDisplayName(classified.Provider)
return fmt.Sprintf(
"The %s provider has been disabled."+
" Contact your Coder administrator.",
displayName,
)
default:
if !classified.Retryable && classified.StatusCode == 0 {
return "The chat request failed unexpectedly."
}
return fmt.Sprintf("%s returned an unexpected error.", subject)
return stringutil.Capitalize(fmt.Sprintf("%s returned an unexpected error.", subject))
}
}
@@ -85,41 +84,43 @@ func retryMessage(classified ClassifiedError) string {
subject := providerSubject(classified.Provider)
switch classified.Kind {
case codersdk.ChatErrorKindOverloaded:
return fmt.Sprintf("%s is temporarily overloaded.", subject)
return stringutil.Capitalize(fmt.Sprintf("%s is temporarily overloaded.", subject))
case codersdk.ChatErrorKindRateLimit:
return fmt.Sprintf("%s is rate limiting requests.", subject)
return stringutil.Capitalize(fmt.Sprintf("%s is rate limiting requests.", subject))
case codersdk.ChatErrorKindTimeout:
return fmt.Sprintf("%s is temporarily unavailable.", subject)
return stringutil.Capitalize(fmt.Sprintf("%s is temporarily unavailable.", subject))
case codersdk.ChatErrorKindStartupTimeout:
return fmt.Sprintf(
return stringutil.Capitalize(fmt.Sprintf(
"%s did not start responding in time.", subject,
)
))
case codersdk.ChatErrorKindAuth:
displayName := providerDisplayName(classified.Provider)
if displayName == "" {
displayName = "the AI provider"
}
return fmt.Sprintf(
"Authentication with %s failed.", displayName,
"Authentication with %s failed.", subject,
)
case codersdk.ChatErrorKindConfig:
return fmt.Sprintf(
return stringutil.Capitalize(fmt.Sprintf(
"%s rejected the model configuration.", subject,
)
))
case codersdk.ChatErrorKindMissingKey:
return "The API key for this conversation is no longer available."
default:
case codersdk.ChatErrorKindProviderDisabled:
displayName := providerDisplayName(classified.Provider)
return fmt.Sprintf(
"%s returned an unexpected error.", subject,
"The %s provider has been disabled by an administrator.",
displayName,
)
default:
return stringutil.Capitalize(fmt.Sprintf(
"%s returned an unexpected error.", subject,
))
}
}
func providerSubject(provider string) string {
if displayName := providerDisplayName(provider); displayName != "" {
if displayName := providerDisplayName(provider); displayName != "AI" && displayName != "" {
return displayName
}
return "The AI provider"
return "the AI provider"
}
func providerDisplayName(provider string) string {
@@ -141,7 +142,7 @@ func providerDisplayName(provider string) string {
case "vercel":
return "Vercel AI Gateway"
default:
return ""
return "AI"
}
}
+3
View File
@@ -4,6 +4,8 @@ import (
"regexp"
"strconv"
"strings"
"github.com/coder/coder/v2/aibridge"
)
type providerHint struct {
@@ -83,6 +85,7 @@ var (
}
genericRetryablePatterns = []string{"server error", "internal server error"}
interruptedPatterns = []string{"chat interrupted", "request interrupted", "operation interrupted"}
providerDisabledPatterns = []string{aibridge.ErrorCodeProviderDisabled}
)
func extractStatusCode(lower string) int {