feat(logs): configurable request-log retention (from #336)

deploy/docker-compose.yml

@@ -17,6 +17,10 @@ services:
       UPDATE_AGENT_URL: "${UPDATE_AGENT_URL:-}"
       UPDATE_AGENT_TOKEN: "${UPDATE_AGENT_TOKEN:-}"
       UPDATE_GITHUB_TOKEN: "${UPDATE_GITHUB_TOKEN:-}"
+      # Request-log retention: default 7 days / 20k rows. Raise both when the
+      # usage/cost panel needs a longer history (~0.7 KB/row; ~150 MB at 90d @ 1.5k rows/day).
+      CLI2API_LOG_RETENTION_DAYS: "${CLI2API_LOG_RETENTION_DAYS:-7}"
+      CLI2API_LOG_MAX_ROWS: "${CLI2API_LOG_MAX_ROWS:-20000}"
     volumes:
       - qoder-data:/data
       - "${CLI2API_UPDATER_SOCKET_DIR:-/tmp/cli2api-updater}:/run/cli2api-updater:ro"
@@ -33,4 +37,4 @@ services:
     restart: unless-stopped
 
 volumes:
-  qoder-data:
+  qoder-data:

internal/logs/recorder.go

@@ -4,13 +4,47 @@ import (
 	"context"
 	"fmt"
 	"log"
+	"os"
+	"strconv"
 	"strings"
 	"sync"
 	"time"
 
 	"github.com/caigee-cmd/cli2api/internal/accounts"
 )
 
+// Request-log retention. Defaults preserve the historical behaviour (7 days /
+// 20k rows); deployments that want a longer usage history can raise the
+// retention window and the row cap together.
+const (
+	defaultRetentionDays = 7
+	defaultMaxLogRows    = 20_000
+	retentionDaysEnv     = "CLI2API_LOG_RETENTION_DAYS"
+	maxLogRowsEnv        = "CLI2API_LOG_MAX_ROWS"
+)
+
+// RetentionDays is how long request logs are kept; 0 or an invalid value falls
+// back to the default.
+func RetentionDays() int {
+	if v := strings.TrimSpace(os.Getenv(retentionDaysEnv)); v != "" {
+		if n, err := strconv.Atoi(v); err == nil && n > 0 {
+			return n
+		}
+	}
+	return defaultRetentionDays
+}
+
+// MaxLogRows caps the retained rows independently of the age window, so a burst
+// of traffic cannot grow the SQLite file without bound.
+func MaxLogRows() int {
+	if v := strings.TrimSpace(os.Getenv(maxLogRowsEnv)); v != "" {
+		if n, err := strconv.Atoi(v); err == nil && n > 0 {
+			return n
+		}
+	}
+	return defaultMaxLogRows
+}
+
 // RequestPersister is the write surface the recorder actually uses.
 // Query methods stay off this interface so a later store package can
 // satisfy logs without logs depending on the concrete SQLite type.
@@ -200,7 +234,8 @@ func (r *RequestRecorder) PurgeLoop(stop <-chan struct{}, every time.Duration) {
 
 func (r *RequestRecorder) purgeOnce() {
 	r.enqueue(func() {
-		if _, err := r.store.PurgeRequestLogs(context.Background(), 7*24*time.Hour, 20_000); err != nil {
+		keep := time.Duration(RetentionDays()) * 24 * time.Hour
+		if _, err := r.store.PurgeRequestLogs(context.Background(), keep, MaxLogRows()); err != nil {
 			logf("purge request logs: %v", err)
 		}
 	})

internal/logs/retention_test.go

@@ -0,0 +1,44 @@
+package logs
+
+import (
+	"testing"
+)
+
+func TestRetentionDaysDefaultsAndOverride(t *testing.T) {
+	t.Setenv(retentionDaysEnv, "")
+	if got := RetentionDays(); got != defaultRetentionDays {
+		t.Fatalf("default retention = %d, want %d", got, defaultRetentionDays)
+	}
+
+	t.Setenv(retentionDaysEnv, "90")
+	if got := RetentionDays(); got != 90 {
+		t.Fatalf("override retention = %d, want 90", got)
+	}
+
+	// Invalid values must not silently disable pruning.
+	for _, bad := range []string{"0", "-3", "abc", "  "} {
+		t.Setenv(retentionDaysEnv, bad)
+		if got := RetentionDays(); got != defaultRetentionDays {
+			t.Fatalf("retention for %q = %d, want default %d", bad, got, defaultRetentionDays)
+		}
+	}
+}
+
+func TestMaxLogRowsDefaultsAndOverride(t *testing.T) {
+	t.Setenv(maxLogRowsEnv, "")
+	if got := MaxLogRows(); got != defaultMaxLogRows {
+		t.Fatalf("default max rows = %d, want %d", got, defaultMaxLogRows)
+	}
+
+	t.Setenv(maxLogRowsEnv, "200000")
+	if got := MaxLogRows(); got != 200000 {
+		t.Fatalf("override max rows = %d, want 200000", got)
+	}
+
+	for _, bad := range []string{"0", "-1", "lots"} {
+		t.Setenv(maxLogRowsEnv, bad)
+		if got := MaxLogRows(); got != defaultMaxLogRows {
+			t.Fatalf("max rows for %q = %d, want default %d", bad, got, defaultMaxLogRows)
+		}
+	}
+}