Hi CTOs,
Let me introduce myself since not all of you know me, I'm Ronny from Backend-Infra team. Currently we use Datadog for our monitoring solution and we prepaid 580125 USD for the whole year (period: Nov 2019-October 2020, details are in https://29022131.atlassian.net/wiki/spaces/DAT/pages/1147275783/Traveloka-Datadog+Pricing). There is also extra charge if we monitor too many hosts/metrics in a particular month, and some months ago there were > 20000 USD monthly extra charge for 2-3 months. And there was a recent incident of a team sending too many new metrics without being aware of it, because there is no cost/usage visibility. So, I have a few questions regarding this.
The first question: how important is it for you to know the cost attribution per product domain based on usages?
Because the easiest way (less work for us) to attribute the cost is by splitting the usage equally between product domains or between CTOs, but this doesn't reflect the actual usage. Datadog recently has a feature to summarize the usages based on tag, and we can use ProductDomain tag in our AWS/GCP resources for this. But the feature is still far from perfect and we still need some manual work to group the monthly usages per product domain (and then estimate the costs based on usages). So we want to be sure that this is really important for you, and if it's not important then we will proceed with splitting the cost equally between CTOs.
The second question: if it's important, how should we attribute the cost for identifiable product domain?
Proposal:
Split the cost per Datadog features using the formula:
product_domain_usages / total_usages_across_Traveloka * total_feature_cost
For example, for custom-metrics feature, if in a certain month:
For other costs that are not included in our annual contract (example: Datadog Logs, APM, network hosts), we will try to use the similar formula if we can get enough information from Datadog. But if there is not enough information, then we will split it equally between applicable product domains/CTOs (depends on how much information we have).
The third question: how should we attribute the cost for unknown product domain?
Background: There are still some custom-metrics that don't have ProductDomain tag. So we can't identify which teams are sending those metrics (and most of the time the metrics are useless because we can't slice and dice the metrics easily).
Proposal: split the cost from unknown product domain equally between applicable CTOs.
Meanwhile, we are also trying to contact Datadog to be able to do this more accurately and more often. We will wait for your feedback until 13 March 2020, 23.45 GMT+7. If there is no feedback, we will proceed with our current proposal.
Thank you!