mirror of
https://github.com/tiennm99/goclaw.git
synced 2026-10-12 14:15:00 +00:00
Clark Cant
5ca433b8ac
The v3 pipeline compacts session history mid-loop (prune_stage + final-request guard) but only mutates the run's message buffer, never the session store. Each turn reloads full history and re-compacts from scratch: message_tokens climb 129k->156k across turns while every turn compacts back down to ~60k. The lossy compaction differs per run, degrading the agent. The same missing persistence stalls episodic memory: the cumulative compaction count never advances, so the episodic worker's idempotency key (sessionKey:count) is pinned and every cycle after the first is skipped. Observed on live traffic: 8 run.completed since deploy, 0 new episodic. Fixes, all reusing existing machinery (no new store methods, no migrations): - Bug A: emitSessionCompleted reads cumulative GetCompactionCount (matching the legacy v2 path) instead of the per-run counter that resets to 0. - Bug B/anti-loop: finalize passes state.Prune.MidLoopCompacted into maybeSummarize; under pressure it lowers the trigger to a unit-aligned threshold (compactionInputCap - overhead, same MaxRequestShare the guard uses) so the compaction is PERSISTED via the existing TruncateHistory + IncrementCompaction path. Defensive floor prevents over-compaction on pathological config; tool-result-only bloat still skips (history-only). - Bug C: SourceID embeds the count (sessionKey:count) so the eventbus dedup key advances per compaction cycle instead of swallowing rapid same-session turns within the 5m TTL. Tests: episodic compaction, maybe_summarize pressure, request budget. go build (PG + sqliteonly), go vet, go test -race all green.
79 lines
2.7 KiB
Go
79 lines
2.7 KiB
Go
package agent
|
|
|
|
import (
|
|
"context"
|
|
"testing"
|
|
|
|
"github.com/nextlevelbuilder/goclaw/internal/providers"
|
|
)
|
|
|
|
// capturingProvider records every ChatRequest passed to Chat.
|
|
// Distinct from stubProvider in intent_classify_test.go (that one ignores the request).
|
|
type capturingProvider struct {
|
|
captured []providers.ChatRequest
|
|
response string
|
|
}
|
|
|
|
func (c *capturingProvider) Chat(_ context.Context, req providers.ChatRequest) (*providers.ChatResponse, error) {
|
|
c.captured = append(c.captured, req)
|
|
return &providers.ChatResponse{Content: c.response}, nil
|
|
}
|
|
func (c *capturingProvider) ChatStream(_ context.Context, req providers.ChatRequest, _ func(providers.StreamChunk)) (*providers.ChatResponse, error) {
|
|
c.captured = append(c.captured, req)
|
|
return &providers.ChatResponse{Content: c.response}, nil
|
|
}
|
|
func (c *capturingProvider) DefaultModel() string { return "capturing-model" }
|
|
func (c *capturingProvider) Name() string { return "capturing" }
|
|
|
|
// TestCompactMessagesInPlace_MaxTokensDynamic verifies that compactMessagesInPlace
|
|
// passes max_tokens == dynamicSummaryMax(estimatedInputTokens) to the provider.
|
|
func TestCompactMessagesInPlace_MaxTokensDynamic(t *testing.T) {
|
|
cap := &capturingProvider{response: "Summary of conversation."}
|
|
|
|
loop := &Loop{
|
|
provider: cap,
|
|
model: "claude-3-5-sonnet",
|
|
contextWindow: 200_000,
|
|
// tokenCounter nil → estimateSummaryInputTokens uses rune/3 fallback
|
|
}
|
|
|
|
// Build 10 dummy messages (>= 6 required by compactMessagesInPlace).
|
|
msgs := make([]providers.Message, 10)
|
|
for i := range msgs {
|
|
if i%2 == 0 {
|
|
msgs[i] = providers.Message{Role: "user", Content: "user message"}
|
|
} else {
|
|
msgs[i] = providers.Message{Role: "assistant", Content: "assistant reply"}
|
|
}
|
|
}
|
|
|
|
result := loop.compactMessagesInPlace(context.Background(), msgs)
|
|
if result == nil {
|
|
t.Fatal("compactMessagesInPlace returned nil; expected compaction to succeed")
|
|
}
|
|
|
|
if len(cap.captured) != 1 {
|
|
t.Fatalf("provider.Chat called %d time(s), want 1", len(cap.captured))
|
|
}
|
|
|
|
req := cap.captured[0]
|
|
maxTokensRaw, ok := req.Options["max_tokens"]
|
|
if !ok {
|
|
t.Fatal("Options[\"max_tokens\"] not set in ChatRequest")
|
|
}
|
|
|
|
maxTokens, ok := maxTokensRaw.(int)
|
|
if !ok {
|
|
t.Fatalf("Options[\"max_tokens\"] type = %T, want int", maxTokensRaw)
|
|
}
|
|
|
|
// Compute expected using the same formula the implementation uses.
|
|
// With keepCount=4 and 10 messages, splitIdx=6 (first 6 messages summarised).
|
|
// tokenCounter nil → rune/3 fallback.
|
|
expectedIn := loop.estimateSummaryInputTokens(msgs[:6])
|
|
wantMax := dynamicSummaryMax(expectedIn)
|
|
if maxTokens != wantMax {
|
|
t.Errorf("max_tokens = %d, want %d (dynamicSummaryMax(%d))", maxTokens, wantMax, expectedIn)
|
|
}
|
|
}
|