openstack/openstack-helm-infra
Code Issues Proposed changes

ceph-osd: Simplify failure domain config using map

Browse Source 
Using a subset of the characters in the hostname to determine the
failure domain is not always possible, and using overrides based on
hostnames is in some ways overkill.

This change provides a simple way to map hostnames to failure domains.
It is used only when 'failure_domain' is set other than 'host', and when
'failure_domain_by_hostname' is 'false'. Any hosts not referenced in the
map will be given the default treatment (root=default host=hostname)

Example usage:

    conf:
      storage:
        failure_domain: rack
        failure_domain_by_hostname_map:
          hostfoo: rack1
          hostbar: rack1
          hostbaz: rack2
          hostqux: rack2

Change-Id: Ia98fec8c623486f80054877e40e0753e4b939e8e
This commit is contained in:
Sphicas, Phil (ps3910) 2020-06-05 07:29:36 +00:00 committed by Phil Sphicas
parent 01705232be
commit 577dcd5d19
3 changed files with 14 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
ceph-osd/templates/bin/osd/ceph-disk/_common.sh.tpl
View File

@ -27,6 +27,7 @@ set -ex
eval CRUSH_FAILURE_DOMAIN_TYPE=$(cat /etc/ceph/storage.json | python -c 'import sys, json; data = json.load(sys.stdin); print(json.dumps(data["failure_domain"]))') eval CRUSH_FAILURE_DOMAIN_TYPE=$(cat /etc/ceph/storage.json | python -c 'import sys, json; data = json.load(sys.stdin); print(json.dumps(data["failure_domain"]))')
eval CRUSH_FAILURE_DOMAIN_NAME=$(cat /etc/ceph/storage.json | python -c 'import sys, json; data = json.load(sys.stdin); print(json.dumps(data["failure_domain_name"]))') eval CRUSH_FAILURE_DOMAIN_NAME=$(cat /etc/ceph/storage.json | python -c 'import sys, json; data = json.load(sys.stdin); print(json.dumps(data["failure_domain_name"]))')
eval CRUSH_FAILURE_DOMAIN_BY_HOSTNAME=$(cat /etc/ceph/storage.json | python -c 'import sys, json; data = json.load(sys.stdin); print(json.dumps(data["failure_domain_by_hostname"]))') eval CRUSH_FAILURE_DOMAIN_BY_HOSTNAME=$(cat /etc/ceph/storage.json | python -c 'import sys, json; data = json.load(sys.stdin); print(json.dumps(data["failure_domain_by_hostname"]))')
eval CRUSH_FAILURE_DOMAIN_FROM_HOSTNAME_MAP=$(cat /etc/ceph/storage.json | jq '.failure_domain_by_hostname_map."'$HOSTNAME'"')
eval DEVICE_CLASS=$(cat /etc/ceph/storage.json | python -c 'import sys, json; data = json.load(sys.stdin); print(json.dumps(data["device_class"]))') eval DEVICE_CLASS=$(cat /etc/ceph/storage.json | python -c 'import sys, json; data = json.load(sys.stdin); print(json.dumps(data["device_class"]))')
if [[ $(ceph -v | egrep -q "nautilus|mimic|luminous"; echo $?) -ne 0 ]]; then if [[ $(ceph -v | egrep -q "nautilus|mimic|luminous"; echo $?) -ne 0 ]]; then
@ -102,6 +103,8 @@ function crush_location {
crush_add_and_move "${CRUSH_FAILURE_DOMAIN_TYPE}" "${CRUSH_FAILURE_DOMAIN_NAME}" crush_add_and_move "${CRUSH_FAILURE_DOMAIN_TYPE}" "${CRUSH_FAILURE_DOMAIN_NAME}"
elif [ "x${CRUSH_FAILURE_DOMAIN_BY_HOSTNAME}" != "xfalse" ]; then elif [ "x${CRUSH_FAILURE_DOMAIN_BY_HOSTNAME}" != "xfalse" ]; then
crush_add_and_move "${CRUSH_FAILURE_DOMAIN_TYPE}" "$(echo ${CRUSH_FAILURE_DOMAIN_TYPE}_$(echo ${HOSTNAME} | cut -c ${CRUSH_FAILURE_DOMAIN_BY_HOSTNAME}))" crush_add_and_move "${CRUSH_FAILURE_DOMAIN_TYPE}" "$(echo ${CRUSH_FAILURE_DOMAIN_TYPE}_$(echo ${HOSTNAME} | cut -c ${CRUSH_FAILURE_DOMAIN_BY_HOSTNAME}))"
elif [ "x${CRUSH_FAILURE_DOMAIN_FROM_HOSTNAME_MAP}" != "xnull" ]; then
crush_add_and_move "${CRUSH_FAILURE_DOMAIN_TYPE}" "${CRUSH_FAILURE_DOMAIN_FROM_HOSTNAME_MAP}"
else else
# NOTE(supamatt): neither variables are defined then we fall back to default behavior # NOTE(supamatt): neither variables are defined then we fall back to default behavior
crush_create_or_move "${CRUSH_LOCATION}" crush_create_or_move "${CRUSH_LOCATION}"

3
ceph-osd/templates/bin/osd/ceph-volume/_common.sh.tpl
View File

@ -27,6 +27,7 @@ set -ex
eval CRUSH_FAILURE_DOMAIN_TYPE=$(cat /etc/ceph/storage.json | python -c 'import sys, json; data = json.load(sys.stdin); print(json.dumps(data["failure_domain"]))') eval CRUSH_FAILURE_DOMAIN_TYPE=$(cat /etc/ceph/storage.json | python -c 'import sys, json; data = json.load(sys.stdin); print(json.dumps(data["failure_domain"]))')
eval CRUSH_FAILURE_DOMAIN_NAME=$(cat /etc/ceph/storage.json | python -c 'import sys, json; data = json.load(sys.stdin); print(json.dumps(data["failure_domain_name"]))') eval CRUSH_FAILURE_DOMAIN_NAME=$(cat /etc/ceph/storage.json | python -c 'import sys, json; data = json.load(sys.stdin); print(json.dumps(data["failure_domain_name"]))')
eval CRUSH_FAILURE_DOMAIN_BY_HOSTNAME=$(cat /etc/ceph/storage.json | python -c 'import sys, json; data = json.load(sys.stdin); print(json.dumps(data["failure_domain_by_hostname"]))') eval CRUSH_FAILURE_DOMAIN_BY_HOSTNAME=$(cat /etc/ceph/storage.json | python -c 'import sys, json; data = json.load(sys.stdin); print(json.dumps(data["failure_domain_by_hostname"]))')
eval CRUSH_FAILURE_DOMAIN_FROM_HOSTNAME_MAP=$(cat /etc/ceph/storage.json | jq '.failure_domain_by_hostname_map."'$HOSTNAME'"')
eval DEVICE_CLASS=$(cat /etc/ceph/storage.json | python -c 'import sys, json; data = json.load(sys.stdin); print(json.dumps(data["device_class"]))') eval DEVICE_CLASS=$(cat /etc/ceph/storage.json | python -c 'import sys, json; data = json.load(sys.stdin); print(json.dumps(data["device_class"]))')
if [[ $(ceph -v | egrep -q "nautilus|mimic|luminous"; echo $?) -ne 0 ]]; then if [[ $(ceph -v | egrep -q "nautilus|mimic|luminous"; echo $?) -ne 0 ]]; then
@ -102,6 +103,8 @@ function crush_location {
crush_add_and_move "${CRUSH_FAILURE_DOMAIN_TYPE}" "${CRUSH_FAILURE_DOMAIN_NAME}" crush_add_and_move "${CRUSH_FAILURE_DOMAIN_TYPE}" "${CRUSH_FAILURE_DOMAIN_NAME}"
elif [ "x${CRUSH_FAILURE_DOMAIN_BY_HOSTNAME}" != "xfalse" ]; then elif [ "x${CRUSH_FAILURE_DOMAIN_BY_HOSTNAME}" != "xfalse" ]; then
crush_add_and_move "${CRUSH_FAILURE_DOMAIN_TYPE}" "$(echo ${CRUSH_FAILURE_DOMAIN_TYPE}_$(echo ${HOSTNAME} | cut -c ${CRUSH_FAILURE_DOMAIN_BY_HOSTNAME}))" crush_add_and_move "${CRUSH_FAILURE_DOMAIN_TYPE}" "$(echo ${CRUSH_FAILURE_DOMAIN_TYPE}_$(echo ${HOSTNAME} | cut -c ${CRUSH_FAILURE_DOMAIN_BY_HOSTNAME}))"
elif [ "x${CRUSH_FAILURE_DOMAIN_FROM_HOSTNAME_MAP}" != "xnull" ]; then
crush_add_and_move "${CRUSH_FAILURE_DOMAIN_TYPE}" "${CRUSH_FAILURE_DOMAIN_FROM_HOSTNAME_MAP}"
else else
# NOTE(supamatt): neither variables are defined then we fall back to default behavior # NOTE(supamatt): neither variables are defined then we fall back to default behavior
crush_create_or_move "${CRUSH_LOCATION}" crush_create_or_move "${CRUSH_LOCATION}"

8
ceph-osd/values.yaml
View File

@ -203,10 +203,18 @@ conf:
# `failure_domain`: Set the CRUSH bucket type for your OSD to reside in. See the supported CRUSH configuration # `failure_domain`: Set the CRUSH bucket type for your OSD to reside in. See the supported CRUSH configuration
# as listed here: Supported CRUSH configuration is listed here: http://docs.ceph.com/docs/nautilus/rados/operations/crush-map/ # as listed here: Supported CRUSH configuration is listed here: http://docs.ceph.com/docs/nautilus/rados/operations/crush-map/
# `failure_domain_by_hostname`: Specify the portion of the hostname to use for your failure domain bucket name. # `failure_domain_by_hostname`: Specify the portion of the hostname to use for your failure domain bucket name.
# `failure_domain_by_hostname_map`: Explicit mapping of hostname to failure domain, as a simpler alternative to overrides.
# `failure_domain_name`: Manually name the failure domain bucket name. This configuration option should only be used # `failure_domain_name`: Manually name the failure domain bucket name. This configuration option should only be used
# when using host based overrides. # when using host based overrides.
failure_domain: "host" failure_domain: "host"
failure_domain_by_hostname: "false" failure_domain_by_hostname: "false"
failure_domain_by_hostname_map: {}
# Example:
# failure_domain_map_hostname_map:
# hostfoo: rack1
# hostbar: rack1
# hostbaz: rack2
# hostqux: rack2
failure_domain_name: "false" failure_domain_name: "false"
# Note: You can override the device class by adding the value (e.g., hdd, ssd or nvme). # Note: You can override the device class by adding the value (e.g., hdd, ssd or nvme).