Bash script to show k8s PVC usage

kubedf

#!/usr/bin/env bash

NODESAPI=/api/v1/nodes

function getNodes() {
  kubectl get --raw $NODESAPI | jq -r '.items[].metadata.name'
}

function getPVCs() {
  jq -s '[flatten | .[].pods[].volume[]? | select(has("pvcRef")) | '\
'{name: .pvcRef.name, capacityBytes, usedBytes, availableBytes, '\
'percentageUsed: (.usedBytes / .capacityBytes * 100)}] | sort_by(.name)'
}

function column() {
  awk '{ for (i = 1; i <= NF; i++) { d[NR, i] = $i; w[i] = length($i) > w[i] ? length($i) : w[i] } } '\
'END { for (i = 1; i <= NR; i++) { printf("%-*s", w[1], d[i, 1]); for (j = 2; j <= NF; j++ ) { printf("%*s", w[j] + 1, d[i, j]) } print "" } }'
}

function defaultFormat() {
  awk 'BEGIN { print "PVC 1K-blocks Used Available Use%" } '\
'{$2 = $2/1024; $3 = $3/1024; $4 = $4/1024; $5 = sprintf("%.0f%%",$5); print $0}'
}

function humanFormat() {
  awk 'BEGIN { print "PVC Size Used Avail Use%" } '\
'{$5 = sprintf("%.0f%%",$5); printf("%s ", $1); system(sprintf("numfmt --to=iec %s %s %s | sed '\''N;N;s/\\n/ /g'\'' | tr -d \\\\n", $2, $3, $4)); print " " $5 }'
}

function format() {
  jq -r '.[] | "\(.name) \(.capacityBytes) \(.usedBytes) \(.availableBytes) \(.percentageUsed)"' |
    $format | column
}

if [ "$1" == "-h" ]; then
  format=humanFormat
else
  format=defaultFormat
fi

for node in $(getNodes); do
  kubectl get --raw $NODESAPI/$node/proxy/stats/summary
done | getPVCs | format

If there's no output at all from the curl statement, you might want to try:
curl -i $KUBEAPI/$node/proxy/stats/summary

to include the response headers, or even:
curl -v $KUBEAPI/$node/proxy/stats/summary

for additional debug. This might help identify why you're not getting any output.

Oh, there is output from the stats/summary curl. No pvcRef information.

…

On Thu, Feb 27, 2020 at 4:46 PM Brendan McGrath ***@***.***> wrote: If there's no output at all from the curl statement, you might want to try: curl -i $KUBEAPI/$node/proxy/stats/summary to include the response headers, or even: curl -v $KUBEAPI/$node/proxy/stats/summary for additional debug. This might help identify why you're not getting any output. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <https://gist.github.com/60cfff7bca6f32969188008ad4a44c9a?email_source=notifications&email_token=AB3BGXIRNUAR4DUHBEBFQIDRFAYETA5CNFSM4HLOQVQKYY3PNVWWK3TUL52HS4DFVNDWS43UINXW23LFNZ2KUY3PNVWWK3TUL5UWJTQAGC242#gistcomment-3192269>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB3BGXM5M2YZSYQLNM3IJTLRFAYETANCNFSM4HLOQVQA> .

same as @tomsherrod except I use Rook Ceph with FlexVolumes.
I do not have any PvcRef with : curl -i $KUBEAPI/$node/proxy/stats/summary
Is it the driver that does not pass PvcRef along?
Can't test with CSI drivers for the moment, I'll let you know

So I adjusted your script to work for my Microk8s env and double checked the results. It is missing pvcref as well. But I think I know why...hostpath does not display them. Each volume manager has to implement metrics. Hostpath doesn't even show limits within the container themselves let alone so its all kind of moot for me I guess.

https://stackoverflow.com/questions/44718268/how-to-monitor-disk-usage-of-kubernetes-persistent-volumes

Very useful, ty.

For those who don't want to use kubectl proxy

#!/usr/bin/env bash

function getNodes() {
  kubectl get --raw=/api/v1/nodes | jq -r '.items[].metadata.name'
}

function getPVCs() {
  jq -s '[flatten | .[].pods[].volume[]? | select(has("pvcRef")) | '\
'{name: .pvcRef.name, capacityBytes, usedBytes, availableBytes, '\
'percentageUsed: (.usedBytes / .capacityBytes * 100)}] | sort_by(.name)'
}

function column() {
  awk '{ for (i = 1; i <= NF; i++) { d[NR, i] = $i; w[i] = length($i) > w[i] ? length($i) : w[i] } } '\
'END { for (i = 1; i <= NR; i++) { printf("%-*s", w[1], d[i, 1]); for (j = 2; j <= NF; j++ ) { printf("%*s", w[j] + 1, d[i, j]) } print "" } }'
}

function defaultFormat() {
  awk 'BEGIN { print "PVC 1K-blocks Used Available Use%" } '\
'{$2 = $2/1024; $3 = $3/1024; $4 = $4/1024; $5 = sprintf("%.0f%%",$5); print $0}'
}

function humanFormat() {
  awk 'BEGIN { print "PVC Size Used Avail Use%" } '\
'{$5 = sprintf("%.0f%%",$5); printf("%s ", $1); system(sprintf("numfmt --to=iec %s %s %s | sed '\''N;N;s/\\n/ /g'\'' | tr -d \\\\n", $2, $3, $4)); print " " $5 }'
}

function format() {
  jq '.[] | "\(.name) \(.capacityBytes) \(.usedBytes) \(.availableBytes) \(.percentageUsed)"' |
    sed 's/^"\|"$//g' |
    $format | column
}

if [ "$1" == "-h" ]; then
  format=humanFormat
else
  format=defaultFormat
fi

for node in $(getNodes); do
  kubectl get --raw=/api/v1/nodes/$node/proxy/stats/summary
done | getPVCs | format

@brun0queiroz That's much better. Thanks for sharing that. I've updated the gist

I am getting this error: parse error:

parse error: Invalid numeric literal at line 1, column 10
PVC Size Used Avail Use%

That's an error from jq. To find out why, you might want to view the individual responses. You can try something similar to what's described here:
https://gist.github.com/redmcg/60cfff7bca6f32969188008ad4a44c9a#gistcomment-3191345

Linode LKE returns nothing. What could be the reason?

It will likely be a result of the csi driver in use:
https://kubernetes-csi.github.io/docs/drivers.html

The driver may not be providing k8s their metrics.

You could follow the instructions here to get a better picture:
https://gist.github.com/redmcg/60cfff7bca6f32969188008ad4a44c9a#gistcomment-3191345

Thank you!

That's awesome, sir!

Works like a charm, thank you very much :)

Awesome, love this script
my customer requested me exactly this output, thank you very much for sharing

Could you please help me to add "NAMESPACE" column as well along with "PVC Capacity Used Available Used%" in your script.

Thank you once again!

Thanks for the kind words.

I'd love to help, but unfortunately the only k8s cluster I have access to at the moment is minikube with the default storage provisioner; which doesn't provide the metrics required by this script.

The only guidance I can provide is that because the physical volume itself is not namespaced, it would probably require additional calls to find the pvc and from there the namespace could be extracted.

I'd love to help, but unfortunately the only k8s cluster I have access to at the moment is...

I wonder if it's a good idea to take this script and use it as the beginning of a new open source CLI tool (or, use it as inspiration for adding a feature to an existing CLI tool if one exists). This use case seems so popular. People like to be able to scan through all the data they've stored in their clusters' disks and learn more about the disks.

kubectl has support for plugins, which can be managed via krew (a kubectl plugin itself). It looks like a plugin that could be used in place of this script has already been created:
https://github.com/yashbhutwala/kubectl-df-pv

That's neat. That plugin's README says it's only compatible with GKE right now though. This script's approach is just pure Kubernetes, right?

Yeah, I'm using the kubernates rest api. But I took a quick look at the code for kubectl-df-pv, and it looks like it's using the same approach. I couldn't see anything that suggested this script would provide a greater level of support; but testing would confirm.

@redmcg
I have managed to add namespace column to your existing code

kubectl get --raw $NODESAPI/$node/proxy/stats/summary|jq -s '[flatten | .[].pods[].volume[]? | select(has("pvcRef")) | ''{namespace: .pvcRef.namespace, name: .pvcRef.name, capacityBytes, usedBytes, availableBytes, ''percentageUsed: (.usedBytes / .capacityBytes * 100)}] | sort_by(.namespace)'|jq -r '.[] | "(.namespace) (.name) (.capacityBytes) (.usedBytes) (.availableBytes) (.percentageUsed)"'|awk '{$3 = $3/(102410241024); $4 = $4/(10241024); $5= $5/(10241024*1024); $6 = sprintf("%.0f%%",$6); print $0}'

---

Namsespace PVC Capacity Used Available Used%

Thank you @redmcg

Good work @vsadanala ! And thanks for sharing.

Nice Work ! It's very useful !! TKS

nice ! thanks

I would had this > 50% usage

bash kubedf -h |tr -d '%' |awk '$NF > 50' | sed -e 's/$/%/'

Hi,
I just added an option to list the pvc information filtered with a namespace based on the script above:

#!/usr/bin/env bash

function getNodes() {
kubectl get --raw=/api/v1/nodes | jq -r '.items[].metadata.name'
}

function getPVCs() {
jq -s '[flatten | .[].pods[].volume[]? | select(has("pvcRef")) | '
'{namespace: .pvcRef.namespace, name: .pvcRef.name, capacityBytes, usedBytes, availableBytes, '
'percentageUsed: (.usedBytes / .capacityBytes * 100)}] | sort_by(.namespace)'
}

function column() {
awk '{ for (i = 1; i <= NF; i++) { d[NR, i] = $i; w[i] = length($i) > w[i] ? length($i) : w[i] } } '
'END { for (i = 1; i <= NR; i++) { printf("%-*s", w[1], d[i, 1]); for (j = 2; j <= NF; j++ ) { printf("%*s", w[j] + 1, d[i, j]) } print "" } }'
}

function defaultFormat() {
awk 'BEGIN { print "Namespace PVC 1K-blocks Used Available Use%" } '
'{$3 = $3/1024; $4 = $4/1024; $5 = $5/1024; $6 = sprintf("%.0f%%",$6); print $0}'
}

function humanFormat() {
awk 'BEGIN { print "Namespace PVC Size Used Avail Use%" } '
'{$6 = sprintf("%.0f%%",$6); printf("%s ", $1); printf("%s ", $2); system(sprintf("numfmt --to=iec %s %s %s | sed '''N;N;s/\n/ /g''' | tr -d \\n", $3, $4, $5)); print " " $6 }'
}

function format() {
jq '.[] | "(.namespace) (.name) (.capacityBytes) (.usedBytes) (.availableBytes) (.percentageUsed)"' |
sed 's/^"|"$//g' |
$format | column
}

function get_pvc_info(){
local _format=${1:-h} && shift
local namespace=${1:-""}
format=humanFormat

[[ "${_format,,}" != "h" ]] && format=defaultFormat

table=$(for node in $(getNodes); do
kubectl get --raw=/api/v1/nodes/${node}/proxy/stats/summary
done | getPVCs | format)
header=$(echo -e "${table}" | head -1)
echo -e "${header}"
[[ -n "${namespace}" ]] && echo -e "${table}" | grep -w "^${namespace} " || echo -e "${table}"
}

get_pvc_info "$@"