Antalya 26.8: Do not read empty namespaces from catalog

[zvonand]

Changelog category (leave one):

• Bug Fix (user-visible misbehavior in an official stable release)

Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):

Fix infinite recursion with empty namespace in Iceberg catalog (#1762 by @ianton-ru) (#2124 by @zvonand).

CI/CD Options

Exclude tests:

• Fast test
• Integration Tests
• Stateless tests
• Stateful tests
• Performance tests
• All with ASAN
• All with TSAN
• All with MSAN
• All with UBSAN
• All with Coverage
• All with Aarch64
• All Regression
• Disable CI Cache

Regression jobs to run:

• Fast suites (mostly <1h)
• Aggregate Functions (2h)
• Alter (1.5h)
• Benchmark (30m)
• ClickHouse Keeper (1h)
• Iceberg (2h)
• LDAP (1h)
• Parquet (1.5h)
• RBAC (1.5h)
• SSL Server (1h)
• S3 (2h)
• S3 Export (2h)
• Swarms (30m)
• Tiered Storage (2h)

Cherry-picked from #2124.

---

Documentation entry for user-facing changes

Solved #1382

[github-actions]

Workflow [PR], commit [f5ce352]

[zvonand]

@blau-ai

[blau-ai]

CI triage — 1 failing check, not caused by this PR

Verdict: 1 failure — 0 PR-caused, 1 flaky/infra. Everything else is green (all builds, Fast test, Stateless amd_debug/arm_binary + s3/distributed-plan, Integration tests arm_binary 1–4, AST fuzzer, BuzzHouse, SQLLogic, SQLStorm, Compatibility, Install packages, Parser memory check).

❌ Stress test (arm_release) — flaky infra, re-run

• Reported failures: Cannot start clickhouse-server → Check failed with exit code 1 (job).
• What actually happened: the stress harness's server-startup health check timed out on the aarch64 runner. There is no crash and no application error — fatal.log and application_errors.txt are both 0 bytes, and clickhouse-server.err.log contains only warnings (Listen [::]:… DNS error: EAI: Address family for hostname not supported — IPv6 disabled on the runner, plus CgroupsReader: Cannot find 'kernel' and shutdown-time MergeMutateSelectedEntry was not finalized warnings).
• Server did come up: on the restart the server (pid 1215) started cleanly — loaded storage policies, metadata and all 16 system tables and began listening. So this is a slow-startup/timeout on a loaded ARM release runner (ThreadFuzzer enabled), not a startup regression.

Why this is not caused by the PR

• The diff is a single 7-line guard in src/Databases/DataLake/RestCatalog.cpp::getNamespacesRecursive (skip subnamespaces with empty names in the Iceberg REST catalog). That code path only runs when querying a DataLake/Iceberg REST catalog database — it is not on the server-startup path and is not exercised by the stress test.
• The exact same binary/commit passed Stress test (amd_debug) and Stress test (arm_debug) (0 failures each). Only the arm_release variant tripped, on a startup-timeout — the signature of a flaky/infra failure, not a code defect.
• No stack trace, sanitizer report, or error referencing catalog/namespace code anywhere in the logs.

Suggested next step

• Re-run "Stress test (arm_release)" — no code change needed. If it reproduces on a clean re-run, it would point at runner/config (IPv6 <listen_host>::</listen_host> binding on a v4-only host, or an undersized/loaded aarch64 runner), still independent of this PR.

The PR itself looks healthy: minimal, well-scoped defensive change with green functional/integration/stress(debug) coverage.

_{Head SHA f5ce352. Analysis from the praktika S3 report + server logs; I can't build or run ClickHouse in this container, so final confirmation is the CI re-run.}