# Wetech Cloud maximum TiKV node memory alert

Apply this topic only when the typed ticket has a non-empty `orgId` and the
question contains the exact Wetech-managed `MaxTiKVMemoryUsageTooHigh` name or
maximum-TiKV-node alert text. That alert identity is sufficient when Jira does
not provide Account Alias. Never apply it to Wetech OP, an APID ticket, or an
ordinary memory question without an alert identity.

`MaxTiKVMemoryUsageTooHigh` evaluates the largest host-memory utilization among
TiKV nodes:

```promql
1 - min by (tidb_cluster_id) (
  avg by (instance, tidb_cluster_id) (
    (
      node_memory_MemFree_bytes{component="tikv"} +
      node_memory_Buffers_bytes{component="tikv"} +
      node_memory_Cached_bytes{component="tikv"}
    ) / node_memory_MemTotal_bytes{component="tikv"}
  )
)
```

- Compare the alert with the same per-instance host-memory calculation. A
  cluster average below the threshold does not contradict one TiKV node
  crossing it.
- The threshold and duration in the actual alert configuration are
  authoritative. The runbooks template currently shows `> 0.7` for 10 minutes.
- TiKV process RSS divided by the advertised node size is a different metric
  and cannot reproduce or negate this alert.
- The rule explains why the alert fired. It does not establish which allocator,
  workload, cache, backlog, or leak caused the host-memory utilization.

Source: `tidbcloud/runbooks/rules/customer/resources.rules.yaml`, alert
`MaxTiKVMemoryUsageTooHigh`, observed at runbooks commit `b2c38f9c` on
2026-08-20.