fix(log-analyzer): track storage probe failures

(cherry picked from commit f5b6cbd5d3)
This commit is contained in:
overtrue
2026-09-08 01:34:29 +08:00
parent 7c759a72d3
commit d4f2467958
2 changed files with 12 additions and 5 deletions
+7 -4
View File
@@ -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。",
)
},
+5 -1
View File
@@ -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",