@celilo/e2e 0.19.2 → 0.20.0

package/src/container-manager.test.ts

@@ -0,0 +1,84 @@
+import { describe, expect, test } from 'bun:test';
+import { removeAlienInterface } from './container-manager';
+import type { ExecResult } from './types';
+
+// A faithful `ip -o addr show scope global` line: "N: name<tab>inet addr/prefix
+// scope global name\..." — the same shape the real container reports, so the
+// orphan regex under test matches it the way it matches the real thing.
+function tableLine(iface: string, cidr: string): string {
+  return `5: ${iface}\t inet ${cidr} scope global ${iface}\\       valid_lft forever preferred_lft forever`;
+}
+
+/** Fakes exec. Interface-table reads return `tables` in sequence, holding the
+ * last one once exhausted; `ip link del` returns `delExitCode`. Every command
+ * is recorded for reach assertions. */
+function fakeExec(tables: string[], delExitCode = 0) {
+  const cmds: string[] = [];
+  let read = 0;
+  const exec = (cmd: string): ExecResult => {
+    cmds.push(cmd);
+    if (cmd.includes('ip -o addr show scope global')) {
+      const stdout = tables[Math.min(read, tables.length - 1)];
+      read++;
+      return { stdout, stderr: '', exitCode: 0 };
+    }
+    if (cmd.startsWith('ip link del ')) {
+      return {
+        stdout: '',
+        stderr: delExitCode === 0 ? '' : 'RTNETLINK answers: Operation not permitted',
+        exitCode: delExitCode,
+      };
+    }
+    throw new Error(`fakeExec: unexpected command: ${cmd}`);
+  };
+  return { exec, cmds };
+}
+
+const SUBNET = '172.31.99.0/24';
+const ORPHAN_TABLE = `${tableLine('eth0', '10.226.20.5/24')}\n${tableLine('eth4', '172.31.99.2/24')}`;
+const CLEAN_TABLE = tableLine('eth0', '10.226.20.5/24');
+
+describe('removeAlienInterface', () => {
+  test('deletes the orphan interface carrying the alien subnet', () => {
+    const { exec, cmds } = fakeExec([ORPHAN_TABLE, CLEAN_TABLE]);
+    removeAlienInterface({ service: 'fw-main', subnet: SUBNET, exec });
+    expect(cmds).toContain('ip link del eth4');
+  });
+
+  test('does nothing when no interface carries the alien subnet', () => {
+    const { exec, cmds } = fakeExec([CLEAN_TABLE]);
+    removeAlienInterface({ service: 'fw-main', subnet: SUBNET, exec });
+    expect(cmds).toEqual(['ip -o addr show scope global']);
+  });
+
+  test('throws naming the surviving interface and container when the repair does not hold', () => {
+    // Both reads report eth4: the delete claimed to run, the interface is
+    // still there. This is exactly the state that made stage 3 of
+    // firewall-interface-classification fail twelve seconds and one converge
+    // away from the real event (celilo#1261).
+    const { exec } = fakeExec([ORPHAN_TABLE, ORPHAN_TABLE]);
+    expect(() => removeAlienInterface({ service: 'fw-main', subnet: SUBNET, exec })).toThrow(
+      /eth4.*fw-main|fw-main.*eth4/s,
+    );
+  });
+
+  test('the throw carries the disconnect failure when docker refused to detach', () => {
+    const { exec } = fakeExec([ORPHAN_TABLE, ORPHAN_TABLE]);
+    expect(() =>
+      removeAlienInterface({
+        service: 'fw-main',
+        subnet: SUBNET,
+        exec,
+        disconnectError: 'Error response from daemon: endpoint not found',
+      }),
+    ).toThrow(/endpoint not found/);
+  });
+
+  test('does not throw when the interface is gone after the repair, even if the delete reported failure', () => {
+    // docker's exit code is not to be believed for veth surgery (see
+    // attachAlienSegment), so a failed delete over a table that is actually
+    // clean is surfaced, not fatal.
+    const { exec } = fakeExec([ORPHAN_TABLE, CLEAN_TABLE], 1);
+    expect(() => removeAlienInterface({ service: 'fw-main', subnet: SUBNET, exec })).not.toThrow();
+  });
+});