3 ms·

One of my consulting customers came to their account team a couple of years ago and said "We want help setting up a cluster, and specifically we want metrics an

One of my consulting customers came to their account team a couple of years ago and said "We want help setting up a cluster, and specifically we want metrics and logging sorted out" - the account team brought this to me, so I said "OK no problem, what are you hoping to monitor, and what decisions will you be making based on this data?" Dear reader, you may be unsurprised to learn that they had no fucking clue what they were supposed to be monitoring, or what they would do with the data. I declined the project.

hef19898 4y ago
Metrics are great if they measure the correct thing and are part ofba properly defined metrics system, and not just a dashboard. E.g., in material flow, every metric should measure onr process / dub-flow performance and output, as this outpit is used downstream. Special emphasis on interfaces between teams and departments. Ideally, those low level, high detail metrics are consolidated into a few high level ones. This makes it easier to measure overall performance, and do diagnose root causes. Most, well almost all, metric systems I encountered in my life so failed misserably at that.
theptip 4y ago
Isn’t this a case where the stock answer (RED / Golden 4 metrics) plus a link to the appropriate chapter of the SRE book is all you need? If you wire up a simple Datadog / Honeycomb dashboard with those, plus log exploring, you have added lots of value that will definitely be used, for a team that clearly has no idea why they need o11y.
- steveBK123 4y ago
  OK now I'm cracking up, I didn't realize o11y was a thing. I was joking with my GenZ teammate that we use s5s (servers) and macroservices. Can we make m6h (monolith) a thing too please?
  - akiselev 4y ago
    Can we go back to autocorrect and just spell the words out? At this point they seem like some sort of secret handshake of the Eleven Society of Extraordinary Consultants.
    - thaumasiotes 4y ago
      > Can we go back to autocorrect and just spell the words out? Those are two different requests. Autocorrect may be a step up in legibility from word length indicators, but it's a big step down from typos. I just swiped "accessibility" with a somewhat-below-par amount of care and attention, and the output was "aggressively".

hef19898 4y ago

Metrics are great if they measure the correct thing and are part ofba properly defined metrics system, and not just a dashboard. E.g., in material flow, every metric should measure onr process / dub-flow performance and output, as this outpit is used downstream. Special emphasis on interfaces between teams and departments. Ideally, those low level, high detail metrics are consolidated into a few high level ones. This makes it easier to measure overall performance, and do diagnose root causes. Most, well almost all, metric systems I encountered in my life so failed misserably at that.

theptip 4y ago

Isn’t this a case where the stock answer (RED / Golden 4 metrics) plus a link to the appropriate chapter of the SRE book is all you need? If you wire up a simple Datadog / Honeycomb dashboard with those, plus log exploring, you have added lots of value that will definitely be used, for a team that clearly has no idea why they need o11y.

steveBK123 4y ago
OK now I'm cracking up, I didn't realize o11y was a thing. I was joking with my GenZ teammate that we use s5s (servers) and macroservices. Can we make m6h (monolith) a thing too please?
- akiselev 4y ago
  Can we go back to autocorrect and just spell the words out? At this point they seem like some sort of secret handshake of the Eleven Society of Extraordinary Consultants.
  - thaumasiotes 4y ago
    > Can we go back to autocorrect and just spell the words out? Those are two different requests. Autocorrect may be a step up in legibility from word length indicators, but it's a big step down from typos. I just swiped "accessibility" with a somewhat-below-par amount of care and attention, and the output was "aggressively".