A script to check the failure domains of OSDs in PGs
The checkPGs script is implemented to check the Object Storage Daemons (OSDs) in Placement Groups (PGs) of ceph pools to make sure OSDs were not allocated from the same failure domain. This script is intended to run from any one of the ceph-mon pods. Invoke the checkPGs script with --help to get the details on how to run it. A Kubernetes cron job is created to schedule the execution of this script at a regular interval. The execution frequency is defined in the ceph-mon/values.yaml file. Change-Id: I5d46bc824e88545cde1cc448ae714d7d3c243817
This commit is contained in:
Al Lau
parent
9f6194ff77
commit
d6cfd78c4d
256
ceph-mon/templates/bin/utils/_checkPGs.py.tpl
Executable file
256
ceph-mon/templates/bin/utils/_checkPGs.py.tpl
Executable file
@ -0,0 +1,256 @@
|
||||
#!/usr/bin/python2
|
||||
|
||||
import subprocess
|
||||
import json
|
||||
import sys
|
||||
from argparse import *
|
||||
|
||||
class cephCRUSH():
|
||||
"""
|
||||
Currently, this script is coded to work with the ceph clusters that have
|
||||
these type-ids -- osd, host, rack, root. To add other type_ids to the
|
||||
CRUSH map, this script needs enhancements to include the new type_ids.
|
||||
|
||||
type_id name
|
||||
------- ----
|
||||
0 osd
|
||||
1 host
|
||||
2 chassis
|
||||
3 rack
|
||||
4 row
|
||||
5 pdu
|
||||
6 pod
|
||||
7 room
|
||||
8 datacenter
|
||||
9 region
|
||||
10 root
|
||||
|
||||
Ceph organizes the CRUSH map in hierarchical topology. At the top, it is
|
||||
the root. The next levels are racks, hosts, and OSDs, respectively. The
|
||||
OSDs are at the leaf level. This script looks at OSDs in each placement
|
||||
group of a ceph pool. For each OSD, starting from the OSD leaf level, this
|
||||
script traverses up to the root. Along the way, the host and rack are
|
||||
recorded and then verified to make sure the paths to the root are in
|
||||
separate failure domains. This script reports the offending PGs to stdout.
|
||||
"""
|
||||
|
||||
"""
|
||||
This list stores the ceph crush hierarchy retrieved from the
|
||||
ceph osd crush tree -f json-pretty
|
||||
"""
|
||||
crushHierarchy = []
|
||||
|
||||
"""
|
||||
Failure Domains - currently our crush map uses these type IDs - osd,
|
||||
host, rack, root
|
||||
If we need to add chassis type (or other types) later on, add the
|
||||
type to the if statement in the crushFD construction section.
|
||||
|
||||
crushFD[0] = {'id': -2, 'name': 'host1', 'type': 'host'}
|
||||
crushFD[23] = {'id': -5, 'name': 'host2', 'type': 'host'}
|
||||
crushFD[68] = {'id': -7, 'name': 'host3', 'type': 'host'}
|
||||
rack_FD[-2] = {'id': -9, 'name': 'rack1', 'type': 'rack' }
|
||||
rack_FD[-15] = {'id': -17, 'name': 'rack2', 'type': 'rack' }
|
||||
root_FD[-17] = {'id': -1, 'name': 'default', 'type': 'root' }}
|
||||
root_FD[-9] = {'id': -1, 'name': 'default', 'type': 'root' }}
|
||||
"""
|
||||
crushFD = {}
|
||||
|
||||
def __init__(self, poolName):
|
||||
if 'all' in poolName or 'All' in poolName:
|
||||
try:
|
||||
poolLs = 'ceph osd pool ls -f json-pretty'
|
||||
poolstr = subprocess.check_output(poolLs, shell=True)
|
||||
self.listPoolName = json.loads(poolstr)
|
||||
except subprocess.CalledProcessError as e:
|
||||
print('{}'.format(e))
|
||||
"""Unable to get all pools - cannot proceed"""
|
||||
sys.exit(2)
|
||||
else:
|
||||
self.listPoolName = poolName
|
||||
|
||||
try:
|
||||
"""Retrieve the crush hierarchies"""
|
||||
crushTree = "ceph osd crush tree -f json-pretty | grep -v '^\[\]'"
|
||||
chstr = subprocess.check_output(crushTree, shell=True)
|
||||
self.crushHierarchy = json.loads(chstr)
|
||||
except subprocess.CalledProcessError as e:
|
||||
print('{}'.format(e))
|
||||
"""Unable to get crush hierarchy - cannot proceed"""
|
||||
sys.exit(2)
|
||||
|
||||
"""
|
||||
Number of racks configured in the ceph cluster. The racks that are
|
||||
present in the crush hierarchy may not be used. The un-used rack
|
||||
would not show up in the crushFD.
|
||||
"""
|
||||
self.count_racks = 0
|
||||
|
||||
"""depth level - 3 is OSD, 2 is host, 1 is rack, 0 is root"""
|
||||
self.osd_depth = 0
|
||||
"""Construct the Failure Domains - OSD -> Host -> Rack -> Root"""
|
||||
for chitem in self.crushHierarchy:
|
||||
if chitem['type'] == 'host' or \
|
||||
chitem['type'] == 'rack' or \
|
||||
chitem['type'] == 'root':
|
||||
for child in chitem['children']:
|
||||
self.crushFD[child] = {'id': chitem['id'], 'name': chitem['name'], 'type': chitem['type']}
|
||||
if chitem['type'] == 'rack' and len(chitem['children']) > 0:
|
||||
self.count_racks += 1
|
||||
elif chitem['type'] == 'osd':
|
||||
if self.osd_depth == 0:
|
||||
self.osd_depth = chitem['depth']
|
||||
|
||||
"""[ { 'pg-name' : [osd.1, osd.2, osd.3] } ... ]"""
|
||||
self.poolPGs = []
|
||||
"""Replica of the pool. Initialize to 0."""
|
||||
self.poolSize = 0
|
||||
|
||||
def getPoolSize(self, poolName):
|
||||
"""
|
||||
size (number of replica) is an attribute of a pool
|
||||
{ "pool": "rbd", "pool_id": 1, "size": 3 }
|
||||
"""
|
||||
pSize = {}
|
||||
"""Get the size attribute of the poolName"""
|
||||
try:
|
||||
poolGet = 'ceph osd pool get ' + poolName + ' size -f json-pretty'
|
||||
szstr = subprocess.check_output(poolGet, shell=True)
|
||||
pSize = json.loads(szstr)
|
||||
self.poolSize = pSize['size']
|
||||
except subprocess.CalledProcessError as e:
|
||||
print('{}'.format(e))
|
||||
self.poolSize = 0
|
||||
"""Continue on"""
|
||||
return
|
||||
|
||||
def checkPGs(self, poolName):
|
||||
if not len(self.poolPGs) > 0:
|
||||
return
|
||||
print('Checking PGs in pool {} ...'.format(poolName)),
|
||||
badPGs = False
|
||||
for pg in self.poolPGs:
|
||||
osdUp = pg['up']
|
||||
"""
|
||||
Construct the OSD path from the leaf to the root. If the
|
||||
replica is set to 3 and there are 3 racks. Each OSD has its
|
||||
own rack (failure domain). If more than one OSD has the
|
||||
same rack, this is a violation. If the number of rack is
|
||||
one, then we need to make sure the hosts for the three OSDs
|
||||
are different.
|
||||
"""
|
||||
check_FD = {}
|
||||
checkFailed = False
|
||||
for osd in osdUp:
|
||||
traverseID = osd
|
||||
"""Start the level with 1 to include the OSD leaf"""
|
||||
traverseLevel = 1
|
||||
while (self.crushFD[traverseID]['type'] != 'root'):
|
||||
crushType = self.crushFD[traverseID]['type']
|
||||
crushName = self.crushFD[traverseID]['name']
|
||||
if crushType in check_FD:
|
||||
check_FD[crushType].append(crushName)
|
||||
else:
|
||||
check_FD[crushType] = [crushName]
|
||||
"""traverse up (to the root) one level"""
|
||||
traverseID = self.crushFD[traverseID]['id']
|
||||
traverseLevel += 1
|
||||
assert (traverseLevel == self.osd_depth), "OSD depth mismatch"
|
||||
"""
|
||||
check_FD should have
|
||||
{
|
||||
'host': ['host1', 'host2', 'host3', 'host4'],
|
||||
'rack': ['rack1', 'rack2', 'rack3']
|
||||
}
|
||||
Not checking for the 'root' as there is only one root.
|
||||
"""
|
||||
for ktype in check_FD:
|
||||
kvalue = check_FD[ktype]
|
||||
if ktype == 'host':
|
||||
"""
|
||||
At the host level, every OSD should come from different
|
||||
host. It is a violation if duplicate hosts are found.
|
||||
"""
|
||||
if len(kvalue) != len(set(kvalue)):
|
||||
if not badPGs:
|
||||
print('Failed')
|
||||
badPGs = True
|
||||
print('OSDs {} in PG {} failed check in host {}'.format(pg['up'], pg['pgid'], kvalue))
|
||||
elif ktype == 'rack':
|
||||
if len(kvalue) == len(set(kvalue)):
|
||||
continue
|
||||
else:
|
||||
"""
|
||||
There are duplicate racks. This could be due to
|
||||
situation like pool's size is 3 and there are only
|
||||
two racks (or one rack). OSDs should come from
|
||||
different hosts as verified in the 'host' section.
|
||||
"""
|
||||
if self.count_racks == len(set(kvalue)):
|
||||
continue
|
||||
elif self.count_racks > len(set(kvalue)):
|
||||
"""Not all the racks were used to allocate OSDs"""
|
||||
if not badPGs:
|
||||
print('Failed')
|
||||
badPGs = True
|
||||
print('OSDs {} in PG {} failed check in rack {}'.format(pg['up'], pg['pgid'], kvalue))
|
||||
check_FD.clear()
|
||||
if not badPGs:
|
||||
print('Passed')
|
||||
return
|
||||
|
||||
def checkPoolPGs(self):
|
||||
for pool in self.listPoolName:
|
||||
self.getPoolSize(pool)
|
||||
if self.poolSize == 1:
|
||||
"""No need to check pool with the size set to 1 copy"""
|
||||
print('Checking PGs in pool {} ... {}'.format(pool, 'Skipped'))
|
||||
continue
|
||||
elif self.poolSize == 0:
|
||||
print('Pool {} was not found.'.format(pool))
|
||||
continue
|
||||
assert (self.poolSize > 1), "Pool size was incorrectly set"
|
||||
|
||||
try:
|
||||
"""Get the list of PGs in the pool"""
|
||||
lsByPool = 'ceph pg ls-by-pool ' + pool + ' -f json-pretty'
|
||||
pgstr = subprocess.check_output(lsByPool, shell=True)
|
||||
self.poolPGs = json.loads(pgstr)
|
||||
"""Check that OSDs in the PG are in separate failure domains"""
|
||||
self.checkPGs(pool)
|
||||
except subprocess.CalledProcessError as e:
|
||||
print('{}'.format(e))
|
||||
"""Continue to the next pool (if any)"""
|
||||
return
|
||||
|
||||
def Main():
|
||||
parser = ArgumentParser(description='''
|
||||
Cross-check the OSDs assigned to the Placement Groups (PGs) of a ceph pool
|
||||
with the CRUSH topology. The cross-check compares the OSDs in a PG and
|
||||
verifies the OSDs reside in separate failure domains. PGs with OSDs in
|
||||
the same failure domain are flagged as violation. The offending PGs are
|
||||
printed to stdout.
|
||||
|
||||
This CLI is executed on-demand on a ceph-mon pod. To invoke the CLI, you
|
||||
can specify one pool or list of pools to check. The special pool name
|
||||
All (or all) checks all the pools in the ceph cluster.
|
||||
''',
|
||||
formatter_class=RawTextHelpFormatter)
|
||||
parser.add_argument('PoolName', type=str, nargs='+',
|
||||
help='List of pools (or All) to validate the PGs and OSDs mapping')
|
||||
args = parser.parse_args()
|
||||
|
||||
if ('all' in args.PoolName or
|
||||
'All' in args.PoolName) and len(args.PoolName) > 1:
|
||||
print('You only need to give one pool with special pool All')
|
||||
sys.exit(1)
|
||||
|
||||
"""
|
||||
Retrieve the crush hierarchies and store it. Cross-check the OSDs
|
||||
in each PG searching for failure domain violation.
|
||||
"""
|
||||
ccm = cephCRUSH(args.PoolName)
|
||||
ccm.checkPoolPGs()
|
||||
|
||||
if __name__ == '__main__':
|
||||
Main()
|
||||
23
ceph-mon/templates/bin/utils/_checkPGs.sh.tpl
Normal file
23
ceph-mon/templates/bin/utils/_checkPGs.sh.tpl
Normal file
@ -0,0 +1,23 @@
|
||||
#!/bin/bash
|
||||
|
||||
{{/*
|
||||
Copyright 2018 The Openstack-Helm Authors.
|
||||
|
||||
Licensed under the Apache License, Version 2.0 (the "License");
|
||||
you may not use this file except in compliance with the License.
|
||||
You may obtain a copy of the License at
|
||||
|
||||
http://www.apache.org/licenses/LICENSE-2.0
|
||||
|
||||
Unless required by applicable law or agreed to in writing, software
|
||||
distributed under the License is distributed on an "AS IS" BASIS,
|
||||
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
See the License for the specific language governing permissions and
|
||||
limitations under the License.
|
||||
*/}}
|
||||
|
||||
set -ex
|
||||
|
||||
monPod=$(kubectl get pods --namespace=${DEPLOYMENT_NAMESPACE} --selector=application=ceph --selector=component=mon --output=jsonpath={.items[0].metadata.name} 2>/dev/null)
|
||||
|
||||
kubectl exec -t ${monPod} --namespace=${DEPLOYMENT_NAMESPACE} -- /tmp/utils-checkPGs.py All 2>/dev/null
|
||||
@ -54,6 +54,12 @@ data:
|
||||
moncheck-reap-zombies.py: |
|
||||
{{ tuple "bin/moncheck/_reap-zombies.py.tpl" . | include "helm-toolkit.utils.template" | indent 4 }}
|
||||
|
||||
utils-checkPGs.py: |
|
||||
{{ tuple "bin/utils/_checkPGs.py.tpl" . | include "helm-toolkit.utils.template" | indent 4 }}
|
||||
|
||||
utils-checkPGs.sh: |
|
||||
{{ tuple "bin/utils/_checkPGs.sh.tpl" . | include "helm-toolkit.utils.template" | indent 4 }}
|
||||
|
||||
{{ if .Values.logging.fluentd }}
|
||||
fluentbit-sidecar.sh: |
|
||||
{{ tuple "bin/mon/_fluentbit-sidecar.sh.tpl" . | include "helm-toolkit.utils.template" | indent 4 }}
|
||||
|
||||
52
ceph-mon/templates/cronjob-checkPGs.yaml
Normal file
52
ceph-mon/templates/cronjob-checkPGs.yaml
Normal file
@ -0,0 +1,52 @@
|
||||
{{/*
|
||||
Copyright 2018 The Openstack-Helm Authors.
|
||||
|
||||
Licensed under the Apache License, Version 2.0 (the "License");
|
||||
you may not use this file except in compliance with the License.
|
||||
You may obtain a copy of the License at
|
||||
|
||||
http://www.apache.org/licenses/LICENSE-2.0
|
||||
|
||||
Unless required by applicable law or agreed to in writing, software
|
||||
distributed under the License is distributed on an "AS IS" BASIS,
|
||||
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
See the License for the specific language governing permissions and
|
||||
limitations under the License.
|
||||
*/}}
|
||||
|
||||
{{- if .Values.manifests.cronjob_checkPGs }}
|
||||
{{- $envAll := . }}
|
||||
|
||||
{{- $serviceAccountName := "ceph-pool-checkpgs" }}
|
||||
{{ tuple $envAll "pool_checkpgs" $serviceAccountName | include "helm-toolkit.snippets.kubernetes_pod_rbac_serviceaccount" }}
|
||||
---
|
||||
apiVersion: batch/v1beta1
|
||||
kind: CronJob
|
||||
metadata:
|
||||
name: {{ $serviceAccountName }}
|
||||
spec:
|
||||
schedule: {{ .Values.jobs.pool_checkPGs.cron | quote }}
|
||||
successfulJobsHistoryLimit: {{ .Values.jobs.pool_checkPGs.history.successJob }}
|
||||
failedJobsHistoryLimit: {{ .Values.jobs.pool_checkPGs.history.failJob }}
|
||||
concurrencyPolicy: {{ .Values.jobs.pool_checkPGs.concurrency.execPolicy }}
|
||||
startingDeadlineSeconds: {{ .Values.jobs.pool_checkPGs.startingDeadlineSecs }}
|
||||
jobTemplate:
|
||||
metadata:
|
||||
labels:
|
||||
{{ tuple $envAll "ceph" "pool-checkpgs" | include "helm-toolkit.snippets.kubernetes_metadata_labels" | indent 8 }}
|
||||
spec:
|
||||
template:
|
||||
spec:
|
||||
containers:
|
||||
- name: {{ $serviceAccountName }}
|
||||
{{ tuple $envAll "ceph_config_helper" | include "helm-toolkit.snippets.image" | indent 12 }}
|
||||
env:
|
||||
- name: DEPLOYMENT_NAMESPACE
|
||||
valueFrom:
|
||||
fieldRef:
|
||||
fieldPath: metadata.namespace
|
||||
command:
|
||||
- /tmp/utils-checkPGs.sh
|
||||
restartPolicy: Never
|
||||
|
||||
{{- end }}
|
||||
@ -156,6 +156,14 @@ spec:
|
||||
mountPath: /tmp/mon-check.sh
|
||||
subPath: mon-check.sh
|
||||
readOnly: true
|
||||
- name: ceph-mon-bin
|
||||
mountPath: /tmp/utils-checkPGs.py
|
||||
subPath: utils-checkPGs.py
|
||||
readOnly: true
|
||||
- name: ceph-mon-bin
|
||||
mountPath: /tmp/utils-checkPGs.sh
|
||||
subPath: utils-checkPGs.sh
|
||||
readOnly: true
|
||||
- name: ceph-mon-etc
|
||||
mountPath: /etc/ceph/ceph.conf
|
||||
subPath: ceph.conf
|
||||
|
||||
@ -113,6 +113,20 @@ network:
|
||||
public: 192.168.0.0/16
|
||||
cluster: 192.168.0.0/16
|
||||
|
||||
jobs:
|
||||
pool_checkPGs:
|
||||
# Execute monthly on the 1st at 00:01 AM
|
||||
cron: "1 0 1 * *"
|
||||
history:
|
||||
# Number of successful job to keep
|
||||
successJob: 1
|
||||
# Number of failed job to keep
|
||||
failJob: 1
|
||||
concurrency:
|
||||
# Skip new job if previous job still active
|
||||
execPolicy: Forbid
|
||||
startingDeadlineSecs: 60
|
||||
|
||||
conf:
|
||||
templates:
|
||||
keyring:
|
||||
@ -319,3 +333,4 @@ manifests:
|
||||
service_mon: true
|
||||
service_mon_discovery: true
|
||||
job_storage_admin_keys: true
|
||||
cronjob_checkPGs: true
|
||||
|
||||
@ -7,6 +7,7 @@ Contents:
|
||||
:maxdepth: 2
|
||||
|
||||
install/index
|
||||
testing/index
|
||||
|
||||
|
||||
Indices and Tables
|
||||
|
||||
21
doc/source/testing/ceph-resiliency/README.rst
Normal file
21
doc/source/testing/ceph-resiliency/README.rst
Normal file
@ -0,0 +1,21 @@
|
||||
==============================================
|
||||
Resiliency Tests for OpenStack-Helm-Infra/Ceph
|
||||
==============================================
|
||||
|
||||
Mission
|
||||
=======
|
||||
|
||||
The goal of our resiliency tests for `OpenStack-Helm-Infra/Ceph
|
||||
<https://github.com/openstack/openstack-helm-infra/tree/master/ceph>`_ is to
|
||||
show symptoms of software/hardware failure and provide the solutions.
|
||||
|
||||
Caveats:
|
||||
- Our focus lies on resiliency for various failure scenarios but
|
||||
not on performance or stress testing.
|
||||
|
||||
Software Failure
|
||||
================
|
||||
* `CRUSH Failure Domain <./failure-domain.html>`_
|
||||
|
||||
Hardware Failure
|
||||
================
|
||||
1234
doc/source/testing/ceph-resiliency/failure-domain.rst
Normal file
1234
doc/source/testing/ceph-resiliency/failure-domain.rst
Normal file
File diff suppressed because it is too large
Load Diff
9
doc/source/testing/ceph-resiliency/index.rst
Normal file
9
doc/source/testing/ceph-resiliency/index.rst
Normal file
@ -0,0 +1,9 @@
|
||||
===============
|
||||
Ceph Resiliency
|
||||
===============
|
||||
|
||||
.. toctree::
|
||||
:maxdepth: 2
|
||||
|
||||
README
|
||||
failure-domain
|
||||
8
doc/source/testing/index.rst
Normal file
8
doc/source/testing/index.rst
Normal file
@ -0,0 +1,8 @@
|
||||
=======
|
||||
Testing
|
||||
=======
|
||||
|
||||
.. toctree::
|
||||
:maxdepth: 2
|
||||
|
||||
ceph-resiliency/index
|
||||
Loading…
Reference in New Issue
Block a user