fix: preserve original JSON bytes in normalizeCacheControlTTL when no TTL change needed

normalizeCacheControlTTL unconditionally re-serializes the entire request
body through json.Unmarshal/json.Marshal even when no TTL normalization
is needed. Go's json.Marshal randomizes map key order and HTML-escapes
<, >, & characters (to \u003c, \u003e, \u0026), producing different raw
bytes on every call.

Anthropic's prompt caching uses byte-prefix matching, so any byte-level
difference causes a cache miss. This means the ~119K system prompt and
tools are re-processed on every request when routed through CPA.

The fix adds a bool return to normalizeTTLForBlock to indicate whether
it actually modified anything, and skips the marshal step in
normalizeCacheControlTTL when no blocks were changed.
This commit is contained in:
Kirill Turanskiy
2026-03-05 22:28:01 +03:00
parent 9397f7049f
commit 97fdd2e088
2 changed files with 32 additions and 7 deletions

View File

@@ -369,6 +369,19 @@ func TestNormalizeCacheControlTTL_DowngradesLaterOneHourBlocks(t *testing.T) {
}
}
func TestNormalizeCacheControlTTL_PreservesOriginalBytesWhenNoChange(t *testing.T) {
// Payload where no TTL normalization is needed (all blocks use 1h with no
// preceding 5m block). The text intentionally contains HTML chars (<, >, &)
// that json.Marshal would escape to \u003c etc., altering byte identity.
payload := []byte(`{"tools":[{"name":"t1","cache_control":{"type":"ephemeral","ttl":"1h"}}],"system":[{"type":"text","text":"<system-reminder>foo & bar</system-reminder>","cache_control":{"type":"ephemeral","ttl":"1h"}}],"messages":[{"role":"user","content":[{"type":"text","text":"hello"}]}]}`)
out := normalizeCacheControlTTL(payload)
if !bytes.Equal(out, payload) {
t.Fatalf("normalizeCacheControlTTL altered bytes when no change was needed.\noriginal: %s\ngot: %s", payload, out)
}
}
func TestEnforceCacheControlLimit_StripsNonLastToolBeforeMessages(t *testing.T) {
payload := []byte(`{
"tools": [