[ceph-osd, ceph-client] Weight OSDs as they are added

Currently OSDs are added by the ceph-osd chart with zero weight
and they get reweighted to proper weights in the ceph-client chart
after all OSDs have been deployed. This causes a problem when a
deployment is partially completed and additional OSDs are added
later. In this case the ceph-client chart has already run and the
new OSDs don't ever get weighted correctly. This change weights
OSDs properly as they are deployed instead. As noted in the
script, the noin flag may be set during the deployment to prevent
rebalancing as OSDs are added if necessary.

Added the ability to set and unset Ceph cluster flags in the
ceph-client chart.

Change-Id: Iac50352c857d874f3956776c733d09e0034a0285

ceph-client/templates/bin/pool/_init.sh.tpl

@@ -67,13 +67,6 @@ create_crushrule {{ .name }} {{ .crush_rule }} {{ .failure_domain }} {{ .device_
 {{- end }}
 {{- end }}
 
-function reweight_osds () {
-  for OSD_ID in $(ceph --cluster "${CLUSTER}" osd df | awk '$3 == "0" {print $1}'); do
-    OSD_WEIGHT=$(ceph --cluster "${CLUSTER}" osd df --format json-pretty| grep -A7 "\bosd.${OSD_ID}\b" | awk '/"kb"/{ gsub(",",""); d= $2/1073741824 ; r = sprintf("%.2f", d); print r }');
-    ceph --cluster "${CLUSTER}" osd crush reweight osd.${OSD_ID} ${OSD_WEIGHT};
-  done
-}
-
 function enable_autoscaling () {
   if [[ "${ENABLE_AUTOSCALER}" == "true" ]]; then
     ceph mgr module enable pg_autoscaler
@@ -81,6 +74,22 @@ function enable_autoscaling () {
   fi
 }
 
+function set_cluster_flags () {
+  if [[ ! -z "${CLUSTER_SET_FLAGS}" ]]; then
+    for flag in ${CLUSTER_SET_FLAGS}; do
+      ceph osd set ${flag}
+    done
+  fi
+}
+
+function unset_cluster_flags () {
+  if [[ ! -z "${CLUSTER_UNSET_FLAGS}" ]]; then
+    for flag in ${CLUSTER_UNSET_FLAGS}; do
+      ceph osd unset ${flag}
+    done
+  fi
+}
+
 function create_pool () {
   POOL_APPLICATION=$1
   POOL_NAME=$2
@@ -162,8 +171,6 @@ function manage_pool () {
   ceph --cluster "${CLUSTER}" osd pool set-quota "${POOL_NAME}" max_bytes $POOL_QUOTA
 }
 
-reweight_osds
-
 {{ $targetPGperOSD := .Values.conf.pool.target.pg_per_osd }}
 {{ $crushRuleDefault := .Values.conf.pool.default.crush_rule }}
 {{ $targetQuota := .Values.conf.pool.target.quota | default 100 }}
@@ -175,6 +182,8 @@ if [[ -z "$(ceph osd versions | grep ceph\ version | grep -v nautilus)" ]]; then
 else
   cluster_capacity=$(ceph --cluster "${CLUSTER}" df | head -n3 | tail -n1 | awk '{print $1 substr($2, 1, 1)}' | numfmt --from=iec)
 fi
+set_cluster_flags
+unset_cluster_flags
 {{- range $pool := .Values.conf.pool.spec -}}
 {{- with $pool }}
 {{- if .crush_rule }}

ceph-client/templates/job-rbd-pool.yaml

@@ -52,6 +52,10 @@ spec:
               value: "ceph"
             - name: ENABLE_AUTOSCALER
               value: {{ .Values.conf.features.pg_autoscaler | quote }}
+            - name: CLUSTER_SET_FLAGS
+              value: {{ .Values.conf.features.cluster_flags.set | quote }}
+            - name: CLUSTER_UNSET_FLAGS
+              value: {{ .Values.conf.features.cluster_flags.unset | quote }}
           command:
             - /tmp/pool-init.sh
           volumeMounts:

ceph-client/values.yaml

@@ -254,6 +254,10 @@ conf:
     mds: true
     mgr: true
     pg_autoscaler: true
+    cluster_flags:
+      # List of flags to set or unset separated by spaces
+      set: ""
+      unset: ""
   pool:
   #NOTE(portdirect): this drives a simple approximation of
   # https://ceph.com/pgcalc/, the `target.osd` key should be set to match the

ceph-osd/templates/bin/osd/ceph-volume/_block.sh.tpl

@@ -83,8 +83,8 @@ else
     --no-systemd ${OSD_ID} ${OSD_FSID}
 fi
 
-# NOTE(supamatt): set the initial crush weight of the OSD to 0 to prevent automatic rebalancing
-OSD_WEIGHT=0
+# NOTE(stevetaylor): Set the OSD's crush weight (use noin flag to prevent rebalancing if necessary)
+OSD_WEIGHT=$(get_osd_crush_weight_from_device ${OSD_DEVICE})
 # NOTE(supamatt): add or move the OSD's CRUSH location
 crush_location
 

ceph-osd/templates/bin/osd/ceph-volume/_bluestore.sh.tpl

@@ -89,8 +89,8 @@ else
   fi
 fi
 
-# NOTE(supamatt): set the initial crush weight of the OSD to 0 to prevent automatic rebalancing
-OSD_WEIGHT=0
+# NOTE(stevetaylor): Set the OSD's crush weight (use noin flag to prevent rebalancing if necessary)
+OSD_WEIGHT=$(get_osd_crush_weight_from_device ${OSD_DEVICE})
 # NOTE(supamatt): add or move the OSD's CRUSH location
 crush_location
 

ceph-osd/templates/bin/osd/ceph-volume/_common.sh.tpl

@@ -276,6 +276,27 @@ function get_lvm_tag_from_device {
   get_lvm_tag_from_volume ${logical_volume} ${tag}
 }
 
+# Helper function to get the size of a logical volume
+function get_lv_size_from_device {
+  device="$1"
+  logical_volume="$(get_lv_from_device ${device})"
+
+  lvs ${logical_volume} -o LV_SIZE --noheadings --units k --nosuffix | xargs | cut -d'.' -f1
+}
+
+# Helper function to get the crush weight for an osd device
+function get_osd_crush_weight_from_device {
+  device="$1"
+  lv_size="$(get_lv_size_from_device ${device})" # KiB
+
+  if [[ ! -z "${BLOCK_DB_SIZE}" ]]; then
+    db_size=$(echo "${BLOCK_DB_SIZE}" | cut -d'B' -f1 | numfmt --from=iec | awk '{print $1/1024}') # KiB
+    lv_size=$((lv_size+db_size)) # KiB
+  fi
+
+  echo ${lv_size} | awk '{printf("%.2f\n", $1/1073741824)}' # KiB to TiB
+}
+
 # Helper function to get a cluster FSID from a physical device
 function get_cluster_fsid_from_device {
   device="$1"