…k #15215)

Replace the InjectProblemPattern no-op stub: store injected patterns
(mutex-guarded, name-keyed/idempotent) and feed them into
detectCommonCauseCorrelation alongside built-ins (deduped by name) via
the same findNodesWithAllProblems/MinNodesForCorrelation/confidence
logic. Signature now returns error (no existing callers): empty name/
problems rejected. Add correlator_test.go (first tests for the package):
detection thresholds, confidence, GetStats, injection validation, and
injected-pattern participation. Race-clean.

Note: task referenced GetCorrelationSummary/AddPattern which don't exist;
real equivalents are GetStats/GetActiveCorrelations.

…k #15216)

MaxRemediationsPerMinute was parsed/defaulted(2)/validated but never
enforced. Add a perMinuteBucket *rate.Limiter (golang.org/x/time/rate)
on RemediatorRegistry via SetMaxRemediationsPerMinute (matching the
Set* idiom; 0 = unlimited), wired in main from config. Enforced in
Remediate AFTER the non-consuming per-hour window check so a per-hour
rejection never burns a per-minute token and a rejected call isn't
counted as executed. Updated the stale config comment. Tests: N succeed
/ N+1 rejected within a minute, and unlimited at 0. go mod tidy promoted
x/time (+modernc/sqlite) to direct, dropped unused gogo/protobuf.

New pkg/logger.Init(cfg) builds a slog JSON/text handler at the configured
level writing to stdout/stderr/file (consuming LogOutput/LogFile, which
were validated but unused), sets slog default, and bridges the stdlib log
package via an io.Writer that strips [LEVEL] prefixes -> slog levels — so
all 374 existing log.Printf sites honor the configured format/destination
with zero churn. Wired in main after config validation (warn-and-continue
on error). Converted 7 detector hot paths to native structured slog with
attributes. Updated stale config comment. Tests: JSON parseable, text,
file output, prefix-bridge level mapping, level filtering.

Also poll-for-stat fix to 3 pre-existing flaky remediation tests exposed
by the slog timing change (production logic unchanged).

…nings (Task #15218)

Add lease_client_test.go covering RequestLease: reachable+grants,
reachable+denies, unreachable+fallback=true (FAKE approval — test
documents the cluster-wide remediation storm this enables), and
unreachable+fallback=false (error). Add prominent storm-risk WARNING
above the FallbackOnUnreachable config field and in the helm values
comment (value stays false). Committed only my hunks (pre-existing image.tag bump excluded).

…219)

Add a test-ci target that runs the fast unit suite (-short) AND the
integration suite (go test ./test/integration/... no -short, 10m timeout)
so coverage gaps like TestLeaseCoordinationFlow and
TestCorrelationDetectionFlow (test/integration/controller/) are exercised
in one gate. Build-tagged integration tests (kind dual-stack) still need
-tags=integration and are not run here. Added to .PHONY.

Apply PRAGMA journal_mode=WAL and synchronous=NORMAL in SQLiteStorage.
Initialize (after ping) so concurrent RequestLease reads don't block the
single writer and writes are faster while staying crash-safe in WAL mode.
Connection pool was already pinned (SetMaxOpenConns/IdleConns(1),
ConnMaxLifetime(0)). Add TestSQLiteStorage_WALAndSynchronous (file-backed
temp DB) asserting journal_mode=wal, synchronous=1 (NORMAL), and
MaxOpenConnections=1.

… #19263 phase 2)

Add RegisterClusterRemediators(registry,cfg,client,nodeName,self...) wiring
the two destructive strategies, fail-closed: registers node-reboot +
pod-delete ONLY when a k8s client and nodeName are present (else logs and
registers nothing). main builds an in-cluster clientset (non-fatal on
error) + passes downward-API POD_NAME/POD_NAMESPACE for self-skip.

NodeRebootRemediator: cordon -> drain (evict, skipping DaemonSet/mirror/
self pods) -> reboot via injected runner (default systemctl reboot);
dry-run returns before ANY mutation; cordon failure aborts reboot; bounded
per-phase timeouts; 30m cooldown. PodDeleteRemediator: deletes only the
pod named in Problem.Metadata (namespace/pod), refuses mirror/self, dry-run
log-only. Tested with fake clientset incl. dry-run-does-not-execute,
drain ordering/skips, and fail-closed registration.

…222 (Task #19263 phase 3)

Add flush-dns/restart-interface/reset-routing/flush-ipv6-route to
validRemediationStrategies (+ restart-interface requires an interface) and
register all four in RegisterBuiltinRemediators as NetworkRemediator
closures (per-op). flush-ipv6-route is now registered + reachable, closing
#17222. NetworkRemediator resolves the target interface from
Problem.Metadata["interface"] (new MonitorRemediationConfig.Interface
field threaded by the detector) with config fallback; AllowMetadataInterface
lets the registered restart-interface singleton construct without a fixed
NIC. Tests: 6 safe strategies registered, dry-run + executor dispatch of
flush-ipv6-route, interface-from-metadata, missing-interface error.

Remove the freePort helper (bind :0 -> read port -> close -> rebind race).
The exporter now captures its real bound address in Start (from the
eagerly-bound listener) and exposes BoundAddr()/BoundPort(); tests
construct an ephemeral-port exporter (Port 0, 9100 default applied only on
the production path) and read the actual port back instead of
pre-allocating. Migrated all freePort call sites. Verified: package
-count=20 and -race -count=5 pass with zero bind flakes; full -short
suite stable.