add repair functionality

server/handler/snapshotRepair.go

@@ -0,0 +1,219 @@
+package handler
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"strconv"
+	"strings"
+	"time"
+	"vctp/db"
+	"vctp/internal/report"
+)
+
+// SnapshotRepair scans existing daily summaries and backfills missing SnapshotTime and lifecycle fields.
+// @Summary Repair daily summaries
+// @Description Backfills SnapshotTime and lifecycle info for existing daily summary tables and reruns monthly lifecycle refinement using hourly data.
+// @Tags snapshots
+// @Produce json
+// @Success 200 {object} map[string]string
+// @Router /api/snapshots/repair [post]
+func (h *Handler) SnapshotRepair(w http.ResponseWriter, r *http.Request) {
+	if r.Method != http.MethodPost {
+		http.Error(w, "method not allowed", http.StatusMethodNotAllowed)
+		return
+	}
+	h.Logger.Info("snapshot repair started", "scope", "daily")
+	repaired, failed := h.repairDailySummaries(r.Context(), time.Now())
+	h.Logger.Info("snapshot repair finished", "daily_repaired", repaired, "daily_failed", failed)
+
+	resp := map[string]string{
+		"status":      "ok",
+		"repaired":    strconv.Itoa(repaired),
+		"failed":      strconv.Itoa(failed),
+    }
+    w.Header().Set("Content-Type", "application/json")
+    _ = json.NewEncoder(w).Encode(resp)
+}
+
+func (h *Handler) repairDailySummaries(ctx context.Context, now time.Time) (repaired int, failed int) {
+	dbConn := h.Database.DB()
+
+	dailyRecs, err := report.SnapshotRecordsWithFallback(ctx, h.Database, "daily", "inventory_daily_summary_", "20060102", time.Time{}, now)
+	if err != nil {
+        h.Logger.Warn("failed to list daily summaries", "error", err)
+        return 0, 1
+	}
+
+	for _, rec := range dailyRecs {
+		h.Logger.Debug("repair daily summary table", "table", rec.TableName, "snapshot_time", rec.SnapshotTime)
+		dayStart := rec.SnapshotTime
+		dayEnd := dayStart.Add(24 * time.Hour)
+
+		if err := db.EnsureSummaryTable(ctx, dbConn, rec.TableName); err != nil {
+			h.Logger.Warn("ensure summary table failed", "table", rec.TableName, "error", err)
+            failed++
+            continue
+        }
+
+        hourlyRecs, err := report.SnapshotRecordsWithFallback(ctx, h.Database, "hourly", "inventory_hourly_", "epoch", dayStart, dayEnd)
+        if err != nil || len(hourlyRecs) == 0 {
+            h.Logger.Warn("no hourly snapshots for repair window", "table", rec.TableName, "error", err)
+            failed++
+            continue
+        }
+
+        cols := []string{
+            `"InventoryId"`, `"Name"`, `"Vcenter"`, `"VmId"`, `"EventKey"`, `"CloudId"`, `"CreationTime"`,
+            `"DeletionTime"`, `"ResourcePool"`, `"Datacenter"`, `"Cluster"`, `"Folder"`,
+            `"ProvisionedDisk"`, `"VcpuCount"`, `"RamGB"`, `"IsTemplate"`, `"PoweredOn"`,
+            `"SrmPlaceholder"`, `"VmUuid"`, `"SnapshotTime"`,
+        }
+        union, err := buildUnionFromRecords(hourlyRecs, cols, `COALESCE(CAST("IsTemplate" AS TEXT), '') NOT IN ('TRUE','true','1')`)
+        if err != nil {
+            h.Logger.Warn("failed to build union for repair", "table", rec.TableName, "error", err)
+            failed++
+			continue
+		}
+
+		h.Logger.Debug("built hourly union for repair", "table", rec.TableName, "hourly_tables", len(hourlyRecs))
+		if err := db.BackfillSnapshotTimeFromUnion(ctx, dbConn, rec.TableName, union); err != nil {
+			h.Logger.Warn("failed to backfill snapshot time", "table", rec.TableName, "error", err)
+			failed++
+			continue
+		}
+		h.Logger.Debug("snapshot time backfill complete", "table", rec.TableName)
+		if err := db.RefineCreationDeletionFromUnion(ctx, dbConn, rec.TableName, union); err != nil {
+			h.Logger.Warn("failed to refine lifecycle during repair", "table", rec.TableName, "error", err)
+			failed++
+			continue
+		}
+		h.Logger.Debug("lifecycle refinement complete", "table", rec.TableName)
+		h.Logger.Info("repair applied", "table", rec.TableName, "actions", "snapshot_time+lifecycle")
+		repaired++
+	}
+	return repaired, failed
+}
+
+// SnapshotRepairSuite runs a sequence of repair routines to fix older deployments in one call.
+// It rebuilds the snapshot registry, syncs vcenter totals, repairs daily summaries, and refines monthly lifecycle data.
+// @Summary Run full snapshot repair suite
+// @Description Rebuilds snapshot registry, backfills per-vCenter totals, repairs daily summaries (SnapshotTime/lifecycle), and refines monthly lifecycle.
+// @Tags snapshots
+// @Produce json
+// @Success 200 {object} map[string]string
+// @Router /api/snapshots/repair/all [post]
+func (h *Handler) SnapshotRepairSuite(w http.ResponseWriter, r *http.Request) {
+	if r.Method != http.MethodPost {
+		http.Error(w, "method not allowed", http.StatusMethodNotAllowed)
+		return
+	}
+
+	ctx := r.Context()
+	dbConn := h.Database.DB()
+
+	// Step 1: rebuild snapshot registry from existing tables.
+	h.Logger.Info("repair suite step", "step", "snapshot_registry")
+	if stats, err := report.MigrateSnapshotRegistry(ctx, h.Database); err != nil {
+		h.Logger.Warn("snapshot registry migration failed", "error", err)
+	} else {
+		h.Logger.Info("snapshot registry migration complete", "hourly_renamed", stats.HourlyRenamed, "daily_registered", stats.DailyRegistered, "monthly_registered", stats.MonthlyRegistered, "errors", stats.Errors)
+	}
+
+	// Step 2: backfill vcenter_totals from registry hourly tables.
+	h.Logger.Info("repair suite step", "step", "vcenter_totals")
+	if err := db.SyncVcenterTotalsFromSnapshots(ctx, dbConn); err != nil {
+		h.Logger.Warn("sync vcenter totals failed", "error", err)
+	}
+
+	// Step 3: repair daily summaries (snapshot time + lifecycle).
+	h.Logger.Info("repair suite step", "step", "daily_summaries")
+	dailyRepaired, dailyFailed := h.repairDailySummaries(ctx, time.Now())
+
+	// Step 4: refine monthly lifecycle using daily summaries (requires SnapshotTime now present after step 3).
+	h.Logger.Info("repair suite step", "step", "monthly_refine")
+	monthlyRefined, monthlyFailed := h.refineMonthlyFromDaily(ctx, time.Now())
+
+	resp := map[string]string{
+		"status":          "ok",
+		"daily_repaired":  strconv.Itoa(dailyRepaired),
+        "daily_failed":    strconv.Itoa(dailyFailed),
+        "monthly_refined": strconv.Itoa(monthlyRefined),
+        "monthly_failed":  strconv.Itoa(monthlyFailed),
+    }
+    w.Header().Set("Content-Type", "application/json")
+    _ = json.NewEncoder(w).Encode(resp)
+}
+
+func (h *Handler) refineMonthlyFromDaily(ctx context.Context, now time.Time) (refined int, failed int) {
+	dbConn := h.Database.DB()
+
+	dailyRecs, err := report.SnapshotRecordsWithFallback(ctx, h.Database, "daily", "inventory_daily_summary_", "20060102", time.Time{}, now)
+	if err != nil {
+		h.Logger.Warn("failed to list daily summaries for monthly refine", "error", err)
+		return 0, 1
+	}
+
+	// Group daily tables by month (YYYYMM).
+	grouped := make(map[string][]report.SnapshotRecord)
+	for _, rec := range dailyRecs {
+		key := rec.SnapshotTime.Format("200601")
+		grouped[key] = append(grouped[key], rec)
+	}
+
+	cols := []string{
+		`"InventoryId"`, `"Name"`, `"Vcenter"`, `"VmId"`, `"EventKey"`, `"CloudId"`, `"CreationTime"`,
+		`"DeletionTime"`, `"ResourcePool"`, `"Datacenter"`, `"Cluster"`, `"Folder"`,
+		`"ProvisionedDisk"`, `"VcpuCount"`, `"RamGB"`, `"IsTemplate"`, `"PoweredOn"`,
+		`"SrmPlaceholder"`, `"VmUuid"`, `"SnapshotTime"`,
+	}
+
+	for monthKey, recs := range grouped {
+		summaryTable := fmt.Sprintf("inventory_monthly_summary_%s", monthKey)
+		h.Logger.Debug("monthly refine", "table", summaryTable, "daily_tables", len(recs))
+		if err := db.EnsureSummaryTable(ctx, dbConn, summaryTable); err != nil {
+			h.Logger.Warn("ensure monthly summary failed", "table", summaryTable, "error", err)
+			failed++
+			continue
+		}
+
+        union, err := buildUnionFromRecords(recs, cols, `COALESCE(CAST("IsTemplate" AS TEXT), '') NOT IN ('TRUE','true','1')`)
+        if err != nil {
+            h.Logger.Warn("failed to build union for monthly refine", "table", summaryTable, "error", err)
+            failed++
+            continue
+		}
+
+		if err := db.RefineCreationDeletionFromUnion(ctx, dbConn, summaryTable, union); err != nil {
+			h.Logger.Warn("failed to refine monthly lifecycle", "table", summaryTable, "error", err)
+			failed++
+			continue
+		}
+		h.Logger.Debug("monthly refine applied", "table", summaryTable)
+		refined++
+	}
+	return refined, failed
+}
+
+func buildUnionFromRecords(recs []report.SnapshotRecord, columns []string, where string) (string, error) {
+	if len(recs) == 0 {
+		return "", fmt.Errorf("no tables provided for union")
+	}
+	colList := strings.Join(columns, ", ")
+	parts := make([]string, 0, len(recs))
+	for _, rec := range recs {
+		if err := db.ValidateTableName(rec.TableName); err != nil {
+			continue
+		}
+		q := fmt.Sprintf(`SELECT %s FROM %s`, colList, rec.TableName)
+		if where != "" {
+			q = q + " WHERE " + where
+		}
+		parts = append(parts, q)
+	}
+	if len(parts) == 0 {
+		return "", fmt.Errorf("no valid tables for union")
+	}
+	return strings.Join(parts, "\nUNION ALL\n"), nil
+}