olivier/claude-failover
1
0
Fork 0
Code Issues Pull requests 3 Projects Releases Packages Wiki Activity Actions

fix(quota): veto 5xx errors + tighten patterns to stop false-positive swaps

Browse source 
v0.2.2's 2-poll confirmation was insufficient because Anthropic 500/503
errors are printed into Claude Code's conversation transcript and stay
visible in every tmux capture until the user scrolls. A persistent
server error would confirm on the second poll and still trigger a swap.

Root cause: the pattern "rate limit" (bare substring) matched any 500
payload that happened to mention rate limits in its error text. Real
HTTP 429s from Anthropic are typed as "rate_limit_error" in the error
payload — and that's the signature we should actually key on.

- Remove "rate limit" from quotaPatterns (too generic — matches transcripts).
- Add "rate_limit_error" (Anthropic's typed 429 error) and "5-hour limit".
- Add serverErrorPatterns veto: "api_error", "overloaded_error",
  "internal server error", "api error: 5". When any is present in the
  pane, isQuotaExhausted returns false even if a quota pattern matched.
- 4 new subtests covering the veto paths + sanity that real 429s pass.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
Ubuntu 2026-04-15 19:26:00 +00:00
parent 7c5f8384fa
commit 62e98cb9e7
3 changed files with 102 additions and 23 deletions
Download patch file Download diff file Expand all files Collapse all files

46
internal/quota/monitor.go
View file

@ -23,13 +23,26 @@ type SwitchRequest struct {
}
// quotaPatterns are substrings that indicate quota exhaustion in a pane.
// Must be specific enough to avoid matching Anthropic server errors (500/503)
// that happen to include generic words like "rate limit" in their payload.
var quotaPatterns = []string{
"you've hit your limit",
"rate limit",
"quota exceeded",
"usage limit reached",
"claude pro usage",
"too many requests",
"you've hit your limit", // Claude Code TUI friendly message
"rate_limit_error", // Anthropic typed error (real HTTP 429)
"quota exceeded", // generic, but specific enough
"usage limit reached", // Claude Pro phrasing
"claude pro usage", // Claude Pro dashboard phrasing
"too many requests", // HTTP 429 status text
"5-hour limit", // Claude Code 5h window phrasing
}
// serverErrorPatterns indicate a transient upstream server error
// (Anthropic 500/503), NOT a quota hit. When any of these is present in
// the pane, we treat the "hit" as a false positive and do NOT swap.
var serverErrorPatterns = []string{
"api_error", // Anthropic typed error (HTTP 500)
"overloaded_error", // Anthropic typed error (HTTP 503)
"internal server error", // HTTP 500 status text
"api error: 5", // Claude TUI rendering of 5xx
}
// resetTimeRe extracts reset times like "resets 8pm", "resets at 11:30pm",
@ -210,9 +223,14 @@ func (m *Monitor) isQuotaPaused() bool {
return len(m.switchCh) > 0
}
// isQuotaExhausted returns true if the pane content indicates quota exhaustion.
// isQuotaExhausted returns true if the pane content indicates quota exhaustion
// AND does not simultaneously show a transient server error (which would make
// the apparent quota match a false positive).
func isQuotaExhausted(paneContent string) bool {
return firstMatchingPattern(paneContent) != ""
if firstMatchingPattern(paneContent) == "" {
return false
}
return !hasServerError(paneContent)
}
// firstMatchingPattern returns the first quota pattern found in paneContent,
@ -227,6 +245,18 @@ func firstMatchingPattern(paneContent string) string {
return ""
}
// hasServerError reports whether the pane content shows an Anthropic 5xx /
// transient server error. Used to veto a quota match: a 500 is not a quota.
func hasServerError(paneContent string) bool {
lower := strings.ToLower(paneContent)
for _, p := range serverErrorPatterns {
if strings.Contains(lower, p) {
return true
}
}
return false
}
// snippet returns a 120-char single-line excerpt of pane content for logging.
func snippet(s string) string {
s = strings.ReplaceAll(s, "\n", " ")