feat: purge scan data for deleted images and stacks (#1467)

Vulnerability scan rows were never cleaned up when their image was removed
from Docker or their stack was deleted, so the Security Overview (including
the Top exploit-risk findings card) kept surfacing findings for artifacts that
no longer exist.

Scan results now reflect what is still on the host:

- Deleting a stack immediately purges its stack:<name> compose-config scan.
- A background reconciliation in the monitor janitor removes scans whose image
  is gone from the node, or whose stack folder no longer exists. It is
  fail-safe: a scan is only removed when its artifact is positively known to be
  gone, the Docker image list is read with a timeout (skipped on failure), and
  stack scans are reconciled only when the stack list is non-empty.
- An opt-out "Remove scans for deleted images and stacks" setting (on by
  default, per-node) lets operators retain scan history for removed artifacts.

Scan deletes remove child findings explicitly, since SQLite foreign-key cascade
is not enabled on the connection.

backend/src/services/DatabaseService.ts

@@ -1572,6 +1572,11 @@ export class DatabaseService {
         stmt.run('metrics_retention_hours', '24');
         stmt.run('log_retention_days', '30');
         stmt.run('scan_history_per_image_limit', '50');
+        // Remove scan results when their image is gone from Docker or their
+        // stack folder is deleted, so the Security Overview reflects what still
+        // exists. On by default; operators who keep scan history for deleted
+        // artifacts can turn it off.
+        stmt.run('prune_orphaned_scans', '1');
         stmt.run('trivy_auto_update', '0');
         stmt.run('trivy_last_notified_version', '');
         stmt.run('deploy_block_honor_suppressions', '0');
@@ -4495,6 +4500,55 @@ export class DatabaseService {
         return txn();
     }
 
+    /**
+     * Distinct image_refs that have at least one scan row for a node. Used by
+     * the orphan-scan reconciler to compare stored scans against the artifacts
+     * (images, stacks) that still exist on the host.
+     */
+    public getDistinctScanImageRefs(nodeId: number): string[] {
+        return (
+            this.db
+                .prepare(
+                    'SELECT DISTINCT image_ref FROM vulnerability_scans WHERE node_id = ?',
+                )
+                .all(nodeId) as Array<{ image_ref: string }>
+        ).map((r) => r.image_ref);
+    }
+
+    /**
+     * Delete every scan (and its findings) for one (node_id, image_ref). Used
+     * to purge scans whose artifact is gone. Children are deleted explicitly
+     * because SQLite foreign-key cascade is not enabled at the connection
+     * level (see pruneScanHistoryPerImage). Returns the parent rows removed;
+     * idempotent (0 when nothing matches).
+     */
+    public deleteScansByImageRef(nodeId: number, imageRef: string): number {
+        const idSubquery =
+            'SELECT id FROM vulnerability_scans WHERE node_id = ? AND image_ref = ?';
+        const deleteChild = (table: string) =>
+            this.db
+                .prepare(`DELETE FROM ${table} WHERE scan_id IN (${idSubquery})`)
+                .run(nodeId, imageRef);
+        const deleteParent = this.db.prepare(
+            'DELETE FROM vulnerability_scans WHERE node_id = ? AND image_ref = ?',
+        );
+        const txn = this.db.transaction(() => {
+            deleteChild('vulnerability_details');
+            deleteChild('secret_findings');
+            deleteChild('misconfig_findings');
+            return deleteParent.run(nodeId, imageRef).changes;
+        });
+        return txn();
+    }
+
+    /**
+     * Purge the compose-config (misconfig) scans for a deleted stack, keyed by
+     * the `stack:<name>` image_ref convention used by scanComposeStack.
+     */
+    public deleteStackScans(nodeId: number, stackName: string): number {
+        return this.deleteScansByImageRef(nodeId, `stack:${stackName}`);
+    }
+
     public getLatestScanForImage(
         nodeId: number,
         imageRef: string,

backend/src/services/MonitorService.ts

@@ -3,6 +3,8 @@ import semver from 'semver';
 import DockerController from './DockerController';
 import { DatabaseService, Node, StackAlert } from './DatabaseService';
 import { NodeRegistry } from './NodeRegistry';
+import { FileSystemService } from './FileSystemService';
+import { normalizeImageRef } from './DriftDetectionService';
 import { NotificationService } from './NotificationService';
 import { FleetUpdateTrackerService } from './FleetUpdateTrackerService';
 import { isValidVersion, getSenchoVersion } from './CapabilityRegistry';
@@ -374,6 +376,17 @@ export class MonitorService {
         try {
             const db = DatabaseService.getInstance();
             const settings = db.getGlobalSettings();
+
+            // Prune scans for artifacts that no longer exist (own try/catch so a
+            // reconcile failure never aborts the disk-usage check below). Lives
+            // here because it is Docker-heavy and the 15-min janitor cadence is
+            // the right throttle; it runs even when the disk alert is disabled.
+            try {
+                await this.reconcileOrphanedScans(db, settings);
+            } catch (e) {
+                console.error('[Monitor] Orphaned-scan reconcile failed', e);
+            }
+
             const janitorLimitGb = parseFloat(settings['docker_janitor_gb']);
             if (isNaN(janitorLimitGb) || janitorLimitGb <= 0) return;
 
@@ -445,6 +458,83 @@ export class MonitorService {
         }
     }
 
+    /**
+     * Remove scans whose artifact no longer exists, so the Security Overview
+     * reflects what is still on the host. Per-instance: only the local node's
+     * scans are reconciled against the local Docker image list and stack dirs.
+     *
+     * Fail-safe by construction: a scan is only purged when we positively know
+     * its artifact is gone. If the image list cannot be read, the whole pass is
+     * skipped (we never delete on an unread list). Stack scans are reconciled
+     * only when getStacks() returns a non-empty list, because it returns [] on
+     * both an empty dir and an FS error and we cannot tell them apart.
+     *
+     * Image refs are compared after normalizeImageRef so equivalent forms match
+     * (untagged `alpine` vs Docker's `alpine:latest`, `docker.io/library/`
+     * prefixes), and both RepoTags and RepoDigests feed the live set so a
+     * digest-pinned scan ref still matches a present image. Erring toward
+     * "matches, keep" is the safe direction: the cost of a miss is a stale scan,
+     * not a wrongly deleted one.
+     */
+    private async reconcileOrphanedScans(
+        db: DatabaseService,
+        settings: Readonly<Record<string, string>>,
+    ): Promise<void> {
+        // Destructive cleanup: enabled only on an explicit '1'. A missing key or
+        // a read failure leaves it off, the safe default.
+        if (settings['prune_orphaned_scans'] !== '1') return;
+
+        const nodeId = NodeRegistry.getInstance().getDefaultNodeId();
+
+        let liveImageRefs: Set<string>;
+        try {
+            const images = (await withTimeout(
+                DockerController.getInstance(nodeId).getImages(),
+                JANITOR_TIMEOUT_MS,
+                'docker images (scan reconcile)',
+            )) as Array<{ RepoTags?: string[]; RepoDigests?: string[] }>;
+            liveImageRefs = new Set<string>();
+            for (const img of images) {
+                for (const ref of [...(img.RepoTags ?? []), ...(img.RepoDigests ?? [])]) {
+                    if (ref && !ref.startsWith('<none>')) liveImageRefs.add(normalizeImageRef(ref));
+                }
+            }
+        } catch (e) {
+            if (isDebugEnabled()) {
+                console.debug('[Monitor:diag] Scan reconcile skipped: image list unavailable', e);
+            }
+            return;
+        }
+
+        // getStacks() never throws (it swallows FS errors and returns []), so an
+        // empty list is ambiguous (empty dir vs failed read); reconcileStacks
+        // gates stack purging on a non-empty list to avoid deleting on a failure.
+        const liveStacks = await FileSystemService.getInstance(nodeId).getStacks();
+        const reconcileStacks = liveStacks.length > 0;
+        const liveStackSet = new Set(liveStacks);
+
+        let purgedImageScans = 0;
+        let purgedStackScans = 0;
+        for (const ref of db.getDistinctScanImageRefs(nodeId)) {
+            if (ref.startsWith('stack:')) {
+                if (!reconcileStacks) continue;
+                if (!liveStackSet.has(ref.slice('stack:'.length))) {
+                    purgedStackScans += db.deleteScansByImageRef(nodeId, ref);
+                }
+            } else if (!liveImageRefs.has(normalizeImageRef(ref))) {
+                purgedImageScans += db.deleteScansByImageRef(nodeId, ref);
+            }
+        }
+
+        if (isDebugEnabled() && (purgedImageScans > 0 || purgedStackScans > 0)) {
+            console.debug(
+                `[Monitor:diag] Scan reconcile: purged ${purgedImageScans} image scan(s) `
+                + `(${liveImageRefs.size} live image refs), ${purgedStackScans} stack scan(s)`
+                + (reconcileStacks ? '' : ' (stack reconcile skipped: empty stack list)'),
+            );
+        }
+    }
+
     /**
      * Check GitHub/Docker Hub for a newer Sencho release and dispatch a
      * one-shot notification. Uses getLatestVersion() which wraps CacheService