feature(nemesis): Migrate extracted nemesis precheck guards

[pehala]

Migrate extracted nemesis into using precheck.

Testing

• unittests

PR pre-checks (self review)

• I added the relevant backport labels
• I didn't leave commented-out/debugging code

[scylladb-promoter]

✅ Test Summary: PASSED

✅ Precommit: PASSED

Total	Passed	Failed	Skipped
67	19	0	48

✅ Tests: PASSED

Total	Passed	Failed	Errors	Skipped
4523	4492	0	0	31

Full build log

[pehala]

@claude review

[claude]

Claude finished @pehala's task in 2m 48s —— View job

---

Code Review

• Gather context: read diff and changed files
• Review nemesis modules (add_remove_dc, network, sla)
• Review unit tests
• Post findings

---

The PR is a clean, well-scoped migration. The core logic is correct across all three modules. A few observations below — two worth addressing, one nit.

sdcm/nemesis/monkey/sla.py

Behavior change in enterprise check (sla.py:32): The old validate_sla_preconditions() always used self.runner.cluster.nodes[0].is_enterprise. The new precheck(node) uses the node parameter passed by the framework. This is the right change — the representative node API is the correct abstraction — but worth noting as a deliberate semantic shift, since the old hard-code to nodes[0] is gone.

requires_prefilled_cs_data class attribute (sla.py:19,94): Clean design. RemoveServiceLevelMonkey sets it to False with a clear comment. The inheritance chain is easy to follow.

get_stress_params() safety-net raise (sla.py:68-80): The comment and the preserved raise are good — they document that the precheck() prune is the primary gate but the in-method guard still defends against any bypass path.

sdcm/nemesis/monkey/network.py

extra_network_interface_precheck() helper (network.py:220-224): Good extraction. Three consumers, no duplication.

Kubernetes bypass (network.py:220,380): RandomInterruptionNetworkMonkey and BlockNetworkMonkey both have kubernetes = True and their own _disrupt_k8s() paths, so skipping the interface check on k8s is correct. StopStartInterfacesNetworkMonkey lacks kubernetes = True and has no k8s code path in disrupt(), so its precheck() correctly always enforces the interface check.

RejectInterNodeNetworkMonkey.precheck() (network.py:280-285): Minor — there is no unit test for the "issue is active → return URL" path. The fixture patches SkipPerIssues.get_issue_details to None globally, so every test in the module exercises only the "issue closed → return None" branch. A test flipping that mock to a real-looking value would close the gap. Fix this →

sdcm/nemesis/monkey/add_remove_dc.py

precheck() (add_remove_dc.py:192-195): Straightforward. The MULTI_REGION check is a config-time constant, correctly classified as a static guard. The prepare_phase_active check stays in disrupt() because it reads event state at execution time — this split is correct.

unit_tests/unit/nemesis/monkey/test_add_remove_dc.py

Stale assertions in test_precheck_keeps_for_single_region (test_add_remove_dc.py:123-125): The test asserts runner.run_repair.assert_not_called(), runner.decommission_nodes.assert_not_called(), and not runner.executed. Since disrupt() is never called, these are vacuously true — they pass unconditionally and give no signal. They're carryover from the replaced test_disrupt_raises_unsupported_for_multi_region test. They don't hurt correctness but make the test misleading. Fix this →

unit_tests/unit/nemesis/monkey/test_sla.py

Parametrized coverage over all 7 SLA monkeys: Good approach — the _ALL_SLA_MONKEYS list covers every concrete subclass and the RemoveServiceLevelMonkey-specific exclusion in test_precheck_skips_when_no_prepare_write_cmd is explicit and clearly explained.

Docs

The plan updates accurately reflect what was done vs. what remains. Phase 3 DoD checkboxes are all ticked with matching deliverables. The phase description change (category-based → location-based split) is a cleaner framing of the remaining work.

---

Summary: Two items worth addressing — a missing negative-path test for RejectInterNodeNetworkMonkey.precheck() and stale assertions in one test function. Production code has no bugs. CI is green.