hectorzhao/lislgosms
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

fix: deduplicate filesystem mounts in monitoring

Browse Source
This commit is contained in:
hectorzhao
2026-08-31 23:39:24 +08:00
parent 12668cee87
commit 1f2dfb5caf
14 changed files with 246 additions and 51 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tools/monitoring/cmpp-alerts.yml
+8 -8
View File
@@ -108,50 +108,50 @@ groups:
threshold: "95%"
- alert: HostRootDiskUsageWarning
expr: ((1 - node_filesystem_avail_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"} / node_filesystem_size_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"}) * 100 > 80) and ((1 - node_filesystem_avail_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"} / node_filesystem_size_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"}) * 100 <= 90)
expr: (max by (instance, device, fstype) ((1 - node_filesystem_avail_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"} / node_filesystem_size_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"}) * 100) > 80) and (max by (instance, device, fstype) ((1 - node_filesystem_avail_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"} / node_filesystem_size_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"}) * 100) <= 90)
for: 15m
labels:
severity: warning
service: host
annotations:
summary: 磁盘文件系统空间不足
description: "磁盘文件系统使用率连续15分钟高于80%。 挂载点：{{ $labels.mountpoint }}；设备：{{ $labels.device }}。"
description: "磁盘文件系统使用率连续15分钟高于80%。 设备：{{ $labels.device }}；文件系统：{{ $labels.fstype }}（绑定挂载已合并）。"
currentValue: "{{ printf \"%.1f\" $value }}%"
threshold: "80%"
- alert: HostRootDiskUsageCritical
expr: (1 - node_filesystem_avail_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"} / node_filesystem_size_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"}) * 100 > 90
expr: max by (instance, device, fstype) ((1 - node_filesystem_avail_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"} / node_filesystem_size_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"}) * 100) > 90
for: 5m
labels:
severity: critical
service: host
annotations:
summary: 磁盘文件系统空间严重不足
description: "磁盘文件系统使用率连续5分钟高于90%。 挂载点：{{ $labels.mountpoint }}；设备：{{ $labels.device }}。"
description: "磁盘文件系统使用率连续5分钟高于90%。 设备：{{ $labels.device }}；文件系统：{{ $labels.fstype }}（绑定挂载已合并）。"
currentValue: "{{ printf \"%.1f\" $value }}%"
threshold: "90%"
- alert: HostRootInodeUsageWarning
expr: ((1 - node_filesystem_files_free{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"} / node_filesystem_files{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"}) * 100 > 80) and ((1 - node_filesystem_files_free{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"} / node_filesystem_files{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"}) * 100 <= 90)
expr: (max by (instance, device, fstype) ((1 - node_filesystem_files_free{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"} / node_filesystem_files{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"}) * 100) > 80) and (max by (instance, device, fstype) ((1 - node_filesystem_files_free{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"} / node_filesystem_files{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"}) * 100) <= 90)
for: 15m
labels:
severity: warning
service: host
annotations:
summary: 磁盘文件系统inode余量偏低
description: "磁盘文件系统inode使用率连续15分钟高于80%。 挂载点：{{ $labels.mountpoint }}；设备：{{ $labels.device }}。"
description: "磁盘文件系统inode使用率连续15分钟高于80%。 设备：{{ $labels.device }}；文件系统：{{ $labels.fstype }}（绑定挂载已合并）。"
currentValue: "{{ printf \"%.1f\" $value }}%"
threshold: "80%"
- alert: HostRootInodeUsageCritical
expr: (1 - node_filesystem_files_free{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"} / node_filesystem_files{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"}) * 100 > 90
expr: max by (instance, device, fstype) ((1 - node_filesystem_files_free{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"} / node_filesystem_files{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"}) * 100) > 90
for: 5m
labels:
severity: critical
service: host
annotations:
summary: 磁盘文件系统inode严重不足
description: "磁盘文件系统inode使用率连续5分钟高于90%。 挂载点：{{ $labels.mountpoint }}；设备：{{ $labels.device }}。"
description: "磁盘文件系统inode使用率连续5分钟高于90%。 设备：{{ $labels.device }}；文件系统：{{ $labels.fstype }}（绑定挂载已合并）。"
currentValue: "{{ printf \"%.1f\" $value }}%"
threshold: "90%"
tools/monitoring/cmpp-managed-alerts.yml
+4 -4
View File
@@ -22,15 +22,15 @@ groups:
labels: { severity: critical, service: host }
annotations: { summary: "主机内存使用率达到严重阈值", description: "主机内存使用率持续超过95%。", currentValue: "{{ $value }}", threshold: "95%" }
- alert: HostRootDiskUsageWarning
expr: ((1 - node_filesystem_avail_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"} / node_filesystem_size_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"}) * 100 > 80) and ((1 - node_filesystem_avail_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"} / node_filesystem_size_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"}) * 100 <= 90)
expr: (max by (instance, device, fstype) ((1 - node_filesystem_avail_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"} / node_filesystem_size_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"}) * 100) > 80) and (max by (instance, device, fstype) ((1 - node_filesystem_avail_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"} / node_filesystem_size_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"}) * 100) <= 90)
for: 15m
labels: { severity: warning, service: host }
annotations: { summary: "磁盘（所有挂载点）使用率达到警告阈值", description: "磁盘（所有挂载点）使用率持续超过80%。", currentValue: "{{ $value }}", threshold: "80%" }
annotations: { summary: "磁盘（独立文件系统）使用率达到警告阈值", description: "磁盘（独立文件系统）使用率持续超过80%。 设备：{{ $labels.device }}；文件系统：{{ $labels.fstype }}（绑定挂载已合并）。", currentValue: "{{ $value }}", threshold: "80%" }
- alert: HostRootDiskUsageCritical
expr: (1 - node_filesystem_avail_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"} / node_filesystem_size_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"}) * 100 > 90
expr: max by (instance, device, fstype) ((1 - node_filesystem_avail_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"} / node_filesystem_size_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"}) * 100) > 90
for: 5m
labels: { severity: critical, service: host }
annotations: { summary: "磁盘（所有挂载点）使用率达到严重阈值", description: "磁盘（所有挂载点）使用率持续超过90%。", currentValue: "{{ $value }}", threshold: "90%" }
annotations: { summary: "磁盘（独立文件系统）使用率达到严重阈值", description: "磁盘（独立文件系统）使用率持续超过90%。 设备：{{ $labels.device }}；文件系统：{{ $labels.fstype }}（绑定挂载已合并）。", currentValue: "{{ $value }}", threshold: "90%" }
- alert: CmppApiHttpErrorRateWarning
expr: (100 * sum(rate(cmpp_api_http_requests_total{status=~"5.."}[5m])) / clamp_min(sum(rate(cmpp_api_http_requests_total[5m])), 0.001) > 1) and (100 * sum(rate(cmpp_api_http_requests_total{status=~"5.."}[5m])) / clamp_min(sum(rate(cmpp_api_http_requests_total[5m])), 0.001) <= 5) and (sum(increase(cmpp_api_http_requests_total{status=~"5.."}[5m])) >= 5)
for: 5m