Prometheus metrics

[jmcruz1983]

A good improvement would be mev-boost producing prometheus metrics with /metrics endpoint

[metachris]

open to a PR for this. what metrics are you thinking of?

[jmcruz1983]

I was thinking about metrics around validators registered and blocks produced, wdyt?

[0xste]

Dropped a draft PR, largely the metrics collected thus far are for more operational/performance metrics, would be good to add validator and block-level metrics additionally, would be great to hear thoughts on this.

[jmcruz1983]

Yes,
I agree,
validator and block level metrics would be quite handy
Thanks!

[0xste]

Going for a sidecar approach for this, as opposed to the "deep" integration originally proposed in #294 given comments from @metachris

These changes are deep and extensive into the core system and logic. Considering that mev-boost is run at 50% of all validators, it is too risky and impactful for adding a large amount of new logic in the core flows at the current time.

Perhaps consider collecting the metrics from the logs, or thinking about a possible plugin interface. 🤔

---

Latest is on: #370
The idea is that you'd run a grok-exporter with an mev-boost flavoured configuration
Some care points:

• Changes to log format would affect the metrics exported
  • Mitigation: Config owned/managed/maintained in the mev-boost repo (would ensure alignment?)
• The fstab/grok_exporter project is not currently maintained afaics
  • Mitigation: it's a pretty simple (but powerful) go tool, i'm happy to contribute if any critical bugs come about, this monitoring is non-invasive, so you can always fall back to logs (or some other grok implementation if required)
• There is currently no maintained docker container for grok-exporter
  • Mitigation: Initial repo here and published to docker-hub here

[0xste]

Can we re-open the discussion on this?

I propose we implement this change in stages to de-risk the implementation

1. Implement the standalone server on the new port and expose only default metrics about the go runtime
   Impact:

• Changes made solely to cli/main.go
• Minimal Dependency Uplifts
• Still enable through a cli flag, only optionally bootstrapping the prom server
• Allow Custom Port Definition

2. Implement metrics for HTTP Server interaction

• http requests, statuses, latency, etc

3. Implement basic metrics for internal MEV details

• Bids, Relays, Validators

4+ Implement more advanced MEV statistics

• Blocks Produced, Relay Performance, Late Proposals in Slot,