After lunch, I'll be livetweeting an #SREcon talk on productionizing machine learning services by Google SREs @salim and @villaviejac, in their professional capacities... [ed: and I saw some 🔥🔥🔥 slides about ML privacy & ethics in their dress rehearsal so this gonna be good]
They are SREs but not necessarily ML scientists/researchers.
Data about what can go wrong has been gathered from 40+ interviews with varying teams. #SREcon
Myths: "machine learning is a black box." "you rarely have to rollback." "ML based monitoring is like other alerting." All nope. #SREcon
What if we could just use real code rather than domain specific languages for immutability? What are the advantages of this approach? [ed: p.s. I'm super excited to have met her in person finally!!!] #gluecon
Previously worked at Azure on Azure Functions. Letting people be really productive using serverless offerings.
Where did people struggle? Doing deployments in repeatable fashions. The problem is not the atoms of code, it's deploying the dependency graph. #gluecon
Database migrations and 17th century windmills by Wander Hillen. #bangbangcon
Windmills are Wander's favorite machines. Disclaimer, he's Dutch. We should let machines do the boring stuff so that human minds can do stuff like poetry and conferences and important stuff. #bangbangcon
Even windmills constructed in 1781 have so much automation; they have wind powered elevators as well as milling capability. and windmills to rotate the windmill! and ability to adjust how much wind it catches to slow down. #bangbangcon
Doxxing vs. unmasking:
* Doxxing = publicizing highly specific personal info of an individual (& freq. facilitates intimidation/threats).
* Unmasking = publicizing name/identifying photo/city-level location for bystander accountability for someone's pseudonymous behavior.
Unmasking serves public safety benefits by enabling decisions to not further facilitate (previously secret) bad behavior by an individual.
Doxxing of already identified person doesn't serve public safety benefits in same way (various ppl have different Nazi-harming stances).
Both *can* rely upon the same set of techniques - investigative work, poking around at public records, examining photographs, trying to find related individuals and connections. But they are *different* outcomes from a similar process.
Going to tipsy-tweet #SREcon lightning talks, probably not at my usual level of detail. Expect more editorializing and heckling!
p.s. see me up front if you want cider, the hotel doesn't seem to be serving it, and @ingridavendano and @Ana_M_Medina hooked me up with a 6-pack last night on a Safeway run. My win is your win too. 2 is my limit, so 4 are up for grabs. #SREcon
Also, obligatory note that it is 100% okay to not drink. Fuck alcohol culture etc. #SREcon
First plenary talk: @nicolefv and @jezhumble on measurement. "If you don't know where you're going, it doesn't matter how fast you get there." #SREcon
Outline of the talk: (1) where am I going, (2) why do we care, (3) improve performance/quality, (4) measure performance, (5) culture & how to measure. #SREcon
Maturity models are for chumps, says @nicolefv. Everyone has one, you're supposed to get to 5. Level caps in World of Warcraft as an example of level creep. [ed: this is a really interesting thing the CRE team at Google needs to consider in prod maturity assessments] #SREcon
But I cannot abide being bullied about the priorities we have set for what is most important to advocate for.
He has repeatedly bullied us about not working on his pet list of issues. And I am tired of having someone on the outside repeatedly tell us what to do and micromanage us -- while not having any skin in the game himself.
There are *far* more important things to worry about. We see all the internal *and* external issues. The public may not even wind up ever hearing about our successes. But we have risked a lot for them.
Currently @stitchfix, previously doing CTO consulting as a service, and before that director of GAE at Google & eng lead at Ebay. #DevOpsDays
Is there a dichotomy between speed and stability? High-performing orgs can do multiple deploys per day. Time to release is 1 hour vs. 1 week, failure recovery is 1 hour instead of a day, <15% of changes fail. #DevOpsDays