Skip to content

Commit 1f9cd24

Browse files
diamonwigginsdiamonwiggins
committed
add embedded cluster spec
1 parent 8ea335c commit 1f9cd24

File tree

1 file changed

+316
-0
lines changed

1 file changed

+316
-0
lines changed

‎host/embedded-cluster.yaml

Lines changed: 316 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,316 @@
1+
apiVersion: troubleshoot.sh/v1beta2
2+
kind: SupportBundle
3+
metadata:
4+
name: default
5+
spec:
6+
hostCollectors:
7+
- cpu: {}
8+
- hostOS: {}
9+
- memory: {}
10+
- blockDevices: {}
11+
- hostServices: {}
12+
- ipv4Interfaces: {}
13+
- time: {}
14+
- diskUsage:
15+
collectorName: root-disk-usage
16+
path: /
17+
- diskUsage:
18+
collectorName: openebs-disk-usage
19+
path: /var/openebs/local
20+
- run:
21+
collectorName: k0s-status
22+
command: k0s
23+
args: [ "status" ]
24+
- run:
25+
collectorName: k0s-issue-template
26+
command: sh
27+
args: [ "-c", "uname -srvmo; cat /etc/os-release || lsb_release -a" ]
28+
- run:
29+
collectorName: k0s-sysinfo
30+
command: k0s
31+
args: [ "sysinfo" ]
32+
- copy:
33+
collectorName: installer-logs
34+
path: /var/lib/embedded-cluster/logs/*.log
35+
- run:
36+
collectorName: k8s-api-healthz-6443
37+
command: "curl"
38+
args: ["-k", "https://localhost:6443/healthz?verbose"]
39+
- run:
40+
collectorName: "free"
41+
command: "free"
42+
args: ["-m"]
43+
- run:
44+
collectorName: "top"
45+
command: "top"
46+
args: ["-b", "-n", "1"]
47+
- run:
48+
collectorName: "uptime"
49+
command: "uptime"
50+
args: []
51+
- run:
52+
collectorName: "uname"
53+
command: "uname"
54+
args: ["-a"]
55+
- run:
56+
collectorName: "df"
57+
command: "df"
58+
args: ["-h"]
59+
- run:
60+
collectorName: "iostat"
61+
command: "iostat"
62+
args: ["-x"]
63+
- run:
64+
collectorName: "sestatus"
65+
command: "sestatus"
66+
args: []
67+
- run:
68+
collectorName: "apparmor-status"
69+
command: "apparmor_status"
70+
args: []
71+
- run:
72+
collectorName: "iptables"
73+
command: "iptables"
74+
args: ["-L", "-v"]
75+
- run:
76+
collectorName: "iptables-version"
77+
command: "iptables"
78+
args: ["-V"]
79+
- run:
80+
collectorName: "nftables-list"
81+
command: "nft"
82+
args: ["list", "table", "filter"]
83+
- run:
84+
collectorName: "ipvsadm"
85+
command: "ipvsadm"
86+
args: ["-l", "-n"]
87+
- run:
88+
collectorName: "lsblk"
89+
command: "lsblk"
90+
args: ["--fs"]
91+
- run:
92+
collectorName: "netstat-ports"
93+
command: "netstat"
94+
args: ["-t", "-u", "-l", "-p", "-n"]
95+
- run:
96+
collectorName: "netstat-route-table"
97+
command: "netstat"
98+
args: ["-r", "-n"]
99+
- run:
100+
collectorName: "resolvectl-status"
101+
command: "resolvectl"
102+
args: ["status"]
103+
- run:
104+
collectorName: "resolv-conf"
105+
command: "cat"
106+
args: ["/etc/resolv.conf"]
107+
- run:
108+
collectorName: "systemd-resolved-conf"
109+
command: "cat"
110+
args: ["/etc/systemd/resolved.conf"]
111+
- run:
112+
collectorName: "nsswitch-conf"
113+
command: "cat"
114+
args: ["/etc/nsswitch.conf"]
115+
- run:
116+
collectorName: "hosts"
117+
command: "cat"
118+
args: ["/etc/hosts"]
119+
- run:
120+
collectorName: "ip-route-table"
121+
command: "ip"
122+
args: ["route"]
123+
- run:
124+
collectorName: "sysctl"
125+
command: "sysctl"
126+
args: ["-a"]
127+
# Gathering hostname info to help troubleshoot scenarios where the hostname mismatch
128+
- run:
129+
collectorName: "hostnames"
130+
command: "sh"
131+
args:
132+
- -c
133+
- |
134+
echo "hostname = $(hostname)"
135+
echo "/proc/sys/kernel/hostname = $(cat /proc/sys/kernel/hostname)"
136+
echo "uname -n = $(uname -n)"
137+
- http:
138+
collectorName: curl-api-replicated-com
139+
get:
140+
url: https://api.replicated.com/healthz
141+
- http:
142+
collectorName: curl-get-replicated-com
143+
get:
144+
url: https://get.replicated.com/healthz
145+
- http:
146+
collectorName: curl-registry-replicated-com
147+
get:
148+
url: https://registry.replicated.com/healthz
149+
- http:
150+
collectorName: curl-proxy-replicated-com
151+
get:
152+
url: https://proxy.replicated.com/healthz
153+
- http:
154+
collectorName: curl-replicated-app
155+
get:
156+
url: https://replicated.app/healthz
157+
- run:
158+
collectorName: "du-root"
159+
command: "sh"
160+
args: ["-c", "du -Shax / --exclude /proc | sort -rh | head -20"]
161+
- run:
162+
collectorName: "mount"
163+
command: "mount"
164+
args: ["-l"]
165+
- run:
166+
collectorName: "vmstat"
167+
command: "vmstat"
168+
args: ["-w"]
169+
- run:
170+
collectorName: "ps-high-load"
171+
command: "sh"
172+
args: ["-c", "ps -eo s,user,cmd | grep ^[RD] | sort | uniq -c | sort -nbr | head -20"]
173+
- run:
174+
collectorName: "ps-detect-antivirus-and-security-tools"
175+
command: "sh"
176+
args: [-c, "ps -ef | grep -E 'clamav|sophos|esets_daemon|fsav|symantec|mfend|ds_agent|kav|bdagent|s1agent|falcon|illumio|xagt' | grep -v grep"]
177+
hostAnalyzers:
178+
- memory:
179+
checkName: Amount of Memory
180+
outcomes:
181+
- warn:
182+
when: "< 2G"
183+
message: At least 2G of memory is recommended
184+
- pass:
185+
message: The system has at least 2G of memory
186+
- diskUsage:
187+
checkName: Root disk usage
188+
collectorName: root-disk-usage
189+
outcomes:
190+
- fail:
191+
when: "total < 40Gi"
192+
message: The disk containing directory / has less than 40Gi of total space
193+
- warn:
194+
when: "used/total > 80%"
195+
message: The disk containing directory / is more than 80% full
196+
- warn:
197+
when: "available < 10Gi"
198+
message: The disk containing directory / has less than 10Gi of disk space available
199+
- pass:
200+
message: The disk containing directory / has sufficient space
201+
- diskUsage:
202+
checkName: OpenEBS disk usage
203+
collectorName: openebs-disk-usage
204+
outcomes:
205+
- fail:
206+
when: "total < 40Gi"
207+
message: The disk containing OpenEBS volumes has less than 40Gi of space
208+
- warn:
209+
when: "used/total > 80%"
210+
message: The disk containing OpenEBS volumes is more than 80% full
211+
- warn:
212+
when: "available < 10Gi"
213+
message: The disk containing OpenEBS volumes has less than 10Gi of disk space available
214+
- pass:
215+
message: The disk containing directory OpenEBS volumes has sufficient space
216+
- textAnalyze:
217+
checkName: Kubernetes API probing
218+
fileName: host-collectors/run-host/k0s-status.txt
219+
regex: 'Kube-api probing successful: true'
220+
outcomes:
221+
- fail:
222+
when: "false"
223+
message: Kubernetes API probing is reporting a failure
224+
- pass:
225+
when: "true"
226+
message: Kubernetes API probing is reporting success
227+
- time:
228+
checkName: "ntp-status"
229+
outcomes:
230+
- fail:
231+
when: "ntp == unsynchronized+inactive"
232+
message: "System clock is not synchronized"
233+
- warn:
234+
when: "ntp == unsynchronized+active"
235+
message: System clock not yet synchronized
236+
- pass:
237+
when: "ntp == synchronized+active"
238+
message: "System clock is synchronized"
239+
- warn:
240+
when: "timezone != UTC"
241+
message: "Non UTC timezone can interfere with system function"
242+
- pass:
243+
when: "timezone == UTC"
244+
message: "Timezone is set to UTC"
245+
- http:
246+
checkName: curl-k8s-api-6443
247+
collectorName: curl-k8s-api-6443
248+
outcomes:
249+
- warn:
250+
when: "error"
251+
message: Unable to curl https://localhost:6443/healthz. Please, run `curl -k https://localhost:6443/healthz` to check further information.
252+
- pass:
253+
when: "statusCode == 200"
254+
message: curl -k https://localhost:6443/healthz returned HTTP CODE response 200.
255+
- warn:
256+
message: "Unexpected response. HTTP CODE response is not 200. Please, run `curl -ki https://localhost:6443/healthz` to check further information."
257+
- http:
258+
checkName: curl-api-replicated-com
259+
collectorName: curl-api-replicated-com
260+
outcomes:
261+
- warn:
262+
when: "error"
263+
message: Error connecting to https://api.replicated.com/healthz
264+
- pass:
265+
when: "statusCode == 200"
266+
message: Connected to https://api.replicated.com/healthz
267+
- warn:
268+
message: "Unexpected response"
269+
- http:
270+
checkName: curl-get-replicated-com
271+
collectorName: curl-get-replicated-com
272+
outcomes:
273+
- warn:
274+
when: "error"
275+
message: Error connecting to https://get.replicated.com/healthz
276+
- pass:
277+
when: "statusCode == 200"
278+
message: Connected to https://get.replicated.com/healthz
279+
- warn:
280+
message: "Unexpected response"
281+
- http:
282+
checkName: curl-registry-replicated-com
283+
collectorName: curl-registry-replicated-com
284+
outcomes:
285+
- warn:
286+
when: "error"
287+
message: Error connecting to https://registry.replicated.com/healthz
288+
- pass:
289+
when: "statusCode == 200"
290+
message: Connected to https://registry.replicated.com/healthz
291+
- warn:
292+
message: "Unexpected response"
293+
- http:
294+
checkName: curl-proxy-replicated-com
295+
collectorName: curl-proxy-replicated-com
296+
outcomes:
297+
- warn:
298+
when: "error"
299+
message: Error connecting to https://proxy.replicated.com/healthz
300+
- pass:
301+
when: "statusCode == 200"
302+
message: Connected to https://proxy.replicated.com/healthz
303+
- warn:
304+
message: "Unexpected response"
305+
- http:
306+
checkName: curl-replicated-app
307+
collectorName: curl-replicated-app
308+
outcomes:
309+
- warn:
310+
when: "error"
311+
message: Error connecting to https://replicated.app/healthz
312+
- pass:
313+
when: "statusCode == 200"
314+
message: Connected to https://replicated.app/healthz
315+
- warn:
316+
message: "Unexpected response"

0 commit comments

Comments
 (0)