[scotwells]

Summary

Since #1372, a shard whose SID matches only a default route counts as unreachable, so a node that has lost every shard of a group now treats all of them as down at once. This adds two cases to the standby table test to pin what happens then: the group publishes no active shard, holds no NAT64 standby, and leaves every Maglev slot empty, so tenants hit group_empty instead of a dead shard. It covers a NAT64 group with a standby and a plain NAT66 group. The review on #1372 also asked for a log line, but none is added, because the existing "egress shard unreachable" warning already prints the default-route reason.

Test plan

• The standby table test passes with the new cases
• Lint passes for the egress route package
• CI unit and root jobs pass
• CI E2E passes

---

---

Generated by Claude Code

[scotwells]

ci / Build and test failed in TestRetirementBacksOffWhileGatewayNeverAcknowledges (internal/serviceaddress) with first wait = 1.000112143s, want 1s. That package isn't part of this diff, which only adds cases to an egress route test.

The likely cause is in the test itself: the retirement backoff is time.Since(LastTransitionTime), and the fake client stores condition times at second precision. A transition late in a second is read back up to a second earlier, so the first wait can come out just over the 1s minimum. It passed 50 times in a row locally under -race. I've re-run the job once.

No fix exists on main yet. A suggested patch, kept out of this PR: have the test assert the first wait is between minRetirementBackoff and minRetirementBackoff + time.Second, or have the reconciler return the minimum for the first wait.

---

Generated by Claude Code

[ecv]

LGTM. Both cases pin the all-unreachable outcome #1372 created: empty group, no NAT64 standby held, so tenants hit group_empty rather than a dead shard. Unit job green at this head.