docs: summarise how cortex compares to alternatives

[errordeveloper]

Most notably, it'd be good to have comparison of Cortex vs Thanos.

Additionally, it'd also disable to see how Cortex compares to:

1. simplest standalone Prometheus server and alert manager
2. multiple redundant Prometheus servers and alert manager
3. federated Prometheus (as this could be really anything, and gets complicated too quickly, comparison in broad terms would be of sufficient)

[tomwilkie]

This will also be useful for the upcoming TSDB panel at Promcon. We could compare on the following axis:

• architecture: Cortex has disaggregated and decoupled all the various functions of Prometheus (scraping, storage, quering, rule evaluation etc) into separate, horizontally scalable services. Thanos is similar in some respects here, having separate query federation layer and separate long term storage, but the scraping, storage & querying of recent data is heavily coupled as its depends on existing Prometheus.
• replication: Cortex follows a single-source of truth model; each target should be scraped once, and then the data is subsequently replicated; Thanos requires each target to be scraped multiple times, and relies on heuristics to dedupe the data at query time.
• TCO: Thanos should be cheaper to run than Cortex, as it doesn't have the extra chunk index.
• features: Thanos has rollups, Cortex doesn't.
• reliability: ...
• scalability: ...
• performance: ...
• operational complexity: ...

[BrianChristie]

Just curious if this got discussed at Promcon and if anyone would be willing to post their notes or point to a recording?
Thanks!

[tomwilkie]

Hi @BrianChristie - we did the TSDB panel at PromCon on the first day; the videos haven't been edited down yet, but when they are I'll make sure to include a link. That being said the panel had a broader scope then Thanos vs Cortex, so might not be what you're looking for. Do you have any specific questions?

[shaoxt]

I have question about the replication. I am running two prometheus pods in one cluster.
Each one has a remotewrite which is writing the metrics to cortex.
So I got two metrics
metric_name{cluster="cluster1"prometheus_replica="prometheus-k8s-0"}
metric_name{cluster="cluster1",prometheus_replica="prometheus-k8s-1"}
How can I dedupe in the query?

[JamesHealey94]

The TSDB panel at PromCon has now been uploaded: https://www.youtube.com/watch?v=3pTG_N8yGSU

[achilles42]

@shaoxt #990

[valyala]

There are also M3DB and VictoriaMetrics to put on the comparison table additionally to Thanos.

[stale]

This issue has been automatically marked as stale because it has not had any activity in the past 30 days. It will be closed in 7 days if no further activity occurs. Thank you for your contributions.

[gouthamve]

Maybe we can extract the end of https://docs.google.com/document/d/1mCS5C9wQxEyFQrknaY4nFvgWR34ZKUgFH7I2osMhbIY/edit and put it onto the website.