Metrics: Startup time

[stefa2k]

Description

There is currently no way to tell the startup time of lighthouse components, except using OS features. Accessing this information on OS level rises security concerns.

Present Behaviour

No metrics of startup time/up time present.

Expected Behaviour

Publishing information about startup time on process_start_time_seconds like teku and prysm do.

[paulhauner]

I'm not exactly sure what we'd publish in process_start_time_seconds. It's not immediately clear to me what we're trying to measure here; startup is fairly complex and involves several components running concurrently.

[stefa2k]

To my understanding Teku and Prysm return the unix timestamp of the start of the main process.

[Yoldark34]

@stefa2k
Workaround for beacon node with validator monitor.

validator_monitor_unaggregated_attestation_total*384
Gives you the qty of seconds since restarted

Reduce to the last value if you got multiple validator. It will provide you with a good enough estimation of when your beacon node restarted.

Still looking for a reliable not resource intensive way to know when the validator process restarded

[Yoldark34]

@stefa2k
For the validator process :

vc_signed_attestations_total*384 reduced / with validator_monitor_validators_total
I dont undestand why i can't do in a single query
vc_signed_attestations_total/validator_monitor_validators_total*384

Best regards

[paulhauner]

To my understanding Teku and Prysm return the unix timestamp of the start of the main process.

Oh I understand. Sorry, I was reading this as "the duration LH took to start up" instead of "the timestamp at which LH was started".

This seems quite easy to achieve.

[Yoldark34]

Thanks <3