Promotes "GCP Billing" dashboard to GA, with some modifications #955
Conversation
|
Let's talk through this today. |
|
(or when you're back) |
This dashboard, as it is, was created by @stephen-soltesz. It was located in Stephen's personal folder in Grafana in mlab-sandbox. This commit is using his work as the basis for an official dashboard to monitor GCP billing, where additional panels and metrics may be added in the future.
Additionally: * resizes numerous panels * removes legend from some panels * removes "stacked" graph display from most panels * renames panels slightly * removes any commens from BigQuery queries * various other tweaks I can't think of now
And a few other minor tweaks.
nkinkade
force-pushed
the
sandbox-kinkade
branch
from
ca2900c to
7c133e7
Compare
Queries the exported GCP billing data in BigQuery for any SKU like "Network Internet Egress%". This will be grossly higher than the Internet egress costs for just the virtual GCE platform nodes, since it includes everything originating from GCE, not just platform nodes. But should still provide a good idea of GCE Internet egress costs, of which virtual platform nodes will surely be a major contributor.
There was a problem hiding this comment.
@stephen-soltesz: Per our discussion the other day, I've restored the "stacked" graphs for costs per project and per service in the "All Projects" section. Additionally, I have added a new BigQuery query to the GCE egress panel, and have added rather lengthy description to the panel. PTAL?
Reviewable status: 0 of 1 approvals obtained (waiting on @stephen-soltesz)
There was a problem hiding this comment.
The "cost per service" panel does not use "$" units (you said you wanted to update that).
GCP egress costs are ~2x our estimate from prometheus metrics?
While there are many " to " variations, there may also be surprises in there... Maybe include this in the panel as a "hidden" query (so it can be enabled manually)?
SELECT
TIMESTAMP_TRUNC(usage_start_time, DAY) AS date,
sku.description as metric,
SUM(cost) as gce_egress
FROM
`mlab-oti.billing.unified`
WHERE
$__timeFilter(usage_start_time)
AND service.description = 'Compute Engine'
AND sku.description LIKE 'Network Internet Egress%'
AND cost > 0
AND project.id = 'mlab-oti'
GROUP BY
date, sku.description
ORDER BY
date, gce_egress
With this feedback,
Reviewable status:
complete! 1 of 1 approvals obtained
It lists only the egress SKUs where the cost per day exceeds $10, which is an arbitrary value, but weeds out quite a lot of SKUs where the cost is mere pennies per day, and make the chart more readable. Also, sets the cost per service per house data type to currency->USD.
There was a problem hiding this comment.
I changed the data type for the "cost per service" panel, thanks for catching that oversight.
gce_egress is much higher than the sum of octets_out, I surmise, because the query is summing all GCE Internet egress for the mlab-oti project, which would include GCE VMs and services that aren't part of the k8s platform, whereas octets_out is very specifically targeting egress traffic from only k8s platform VMs. There are fairly detailed descriptions in panel description field for each plot. Hover over the lowercase "i" in the top left of the panel to see if you think the descriptions I had added there shed any meaningful light on what you see and describe those discrepancies.
I also renamed gce_egress to gce_egress_sum, and octets_out to octets_out_sum to better characterize what they represent.
I added your query to the panel, but set AND cost > $10 to filter out all the SKUs where the hourly cost is mere pennies and not of much interest to us, I think. And the query is not hidden, but visible by default, since I think it's useful data.
Reviewable status:
|
Ah, yes of course. We have monitoring, we have the k8s management traffic, and that explains why there's "Americas to |
https://grafana.mlab-sandbox.measurementlab.net/d/a5mC51ZMk/gcp-billing
This PR takes the "GCP Billing" dashboard that @stephen-soltesz created with the following changes
This change is