fix(log-analyzer): track storage probe failures (#7434)

* fix(log-analyzer): track storage probe failures

* fix(error): merge equivalent api message branches

Co-Authored-By: heihutu <heihutu@gmail.com>

Co-Authored-By: zhi22915 <qiuzgang@gmail.com>

---------

Co-authored-by: houseme <housemecn@gmail.com>
Co-authored-by: zhi22915 <qiuzgang@gmail.com>

.config/e2e-repl-nightly-selection.txt

@@ -1 +1 @@
-sha256=0fe8408874ccec3620262a9812d67920ddd72dc9edf0e36e0d0aed3f8bad026e
+sha256=95c8adc016bbc0df9fb2afa24a108bcdf6567ec4d0518725a6cae301593ab556

crates/log-analyzer/src/rules/seed/disk.rs

@@ -68,14 +68,17 @@ pub(super) fn rules() -> Vec<Rule> {
             )
         },
         Rule {
-            anchors: strings(["reporting peer disks offline after consecutive storage_info failures"]),
+            anchors: strings(["Storage inventory probe failed; current drive health is unknown"]),
             ..base(
                 "peer-disks-offline",
                 P2Degraded,
                 "disk",
-                "peer 磁盘被整体判定离线",
-                contains("reporting peer disks offline after consecutive storage_info failures"),
-                "对某 peer 连续 storage_info 失败,判定其磁盘整体离线。",
+                "peer 存储清单探测失败",
+                any([
+                    contains("Storage inventory probe failed; current drive health is unknown"),
+                    contains("reporting peer disks offline after consecutive storage_info failures"),
+                ]),
+                "某 peer 的 storage_info 探测失败,当前磁盘健康状态未知。",
                 "检查该 peer 节点存活与 RPC 端口可达。",
             )
         },

crates/log-analyzer/src/rules/seed/tests.rs

@@ -110,7 +110,7 @@ fn every_rule_has_a_positive_sample() {
         ("remote-peer-faulty", msg("Remote peer health check failed for node2: marking as faulty")),
         (
             "peer-disks-offline",
-            msg("reporting peer disks offline after consecutive storage_info failures"),
+            msg("Storage inventory probe failed; current drive health is unknown"),
         ),
         ("drive-faulty-error", msg("remote drive is faulty")),
         (
@@ -318,6 +318,10 @@ fn smoke_samples_hit_exact_rule_sets() {
         &["disk-marked-faulty"],
     );
     exact(&msg("erasure write quorum (required=8, achieved=5)"), &["ec-write-quorum"]);
+    exact(
+        &msg("reporting peer disks offline after consecutive storage_info failures"),
+        &["peer-disks-offline"],
+    );
     exact(
         &Sample {
             message: "Metacache listing quorum failed",