by Gene Kim on
@adrianco: #Monitorama - Please, no more Minutes, Milliseconds, Monoliths or Monitoring Tools! by @adrianco #cloud http://t.co/gLK7Q3LAtXg
.@adrianco: "My job these days: 'baffling-late-adopters as a service'" OMG. Amazing graph:https://pbs.twimg.com/media/Bm40zNwCIAEZpke.jpg
Slide showing @adrianco incredible contribs to monitoring over last 15+ years; "virtual adrian" Yes.https://pbs.twimg.com/media/Bm41Tt3CEAAyAFg.jpg
.@adrianco: "No more monitoring tools; we need analysis. Let's rename this conf to #analysisrama" (haha)
.@adrianco: "I want people to spend more time understanding systems, & dynamically controlling systems, feedback loops"
.@adrianco: "What's wrong with mins? Usually 8m delay after something bad, rollback, then 8 min more to see if fixed!"
.@adrianco: "CD/DevOps: lots of small chgs, but 1 chg much more likely to break; needs instantaneous detection to recover"
.@adrianco: "Netflix Hystrix/Turbine circuit breaker monitoring: 1 data pt per second;
.@adrianco: "Rule #2: total feedback loop (detect/fix) needs to be less than human perception (~10s)
.@adrianco: "Milliseconds too long; must JVM has ns timers
.@adrianco: "Rule #3: Validate your measurement system has enough accuracy and precision"
.@adrianco arguing that monolithic monitoring systems don't cut it: 'can't have gaps in your telemetry':https://pbs.twimg.com/media/Bm43fxdCcAAg1tq.jpg
.@adrianco: "@a32an: RT @hertling: #1: Spend more time working on code that analyzes the meaning of metrics than code that collects/stores/displays metrics. #mo…
.@adrianco: "Use in-band monitoring [uses same services/infrastructure as your service] & out-of-band [like SaaS]"
.@adrianco: "Your monitoring MUST be more available than the service you're actually monitoring." (!!)
.@adrianco: "High rate of chgs: ephemeral configs (can't hand-tweak); microservices w/complex calling patterns"
.@adrianco showing Gilt's amazing growth in services:https://pbs.twimg.com/media/Bm44pU8CYAAKxUR.jpg
.@adrianco: Haha. OMG. The Death Star architecture for Netflix, Gilt Group, Twitter:https://pbs.twimg.com/media/Bm443CKCMAEkNG8.jpg
.@adrianco desc using FFT to forward-predicting to auto-scale: see weekend bulge: biz metrics:https://pbs.twimg.com/media/Bm45VCHIEAAhfSr.jpg
@vingado12345678: RT @newrelic: "Monitoring systems need to be more available and scalable than the systems being monitored" @adrianco
.@adrianco: "In DevOps, devs are managing services, now driving APM based: biz transactions, JVM metrics, transaction errors (Netflix Servo, Yammer Metrics)
.@adrianco: "Embedding metrics (allowing use in other people's tool) is so useful, allows virality
.@adrianco: "Cloud assets bursty: Netflix code push (once every 40s) creates 100s of servers; often re-uses IP/MAC
.@adrianco: "NetflixOSS Edda: record a full history of your configuration
.@adrianco: "Many of our Cassandra clusters span 4 different regions;
.@adrianco's 5 New Rules of Monitoring:https://pbs.twimg.com/media/Bm47Rg-CMAAEE2N.jpg
.@adrianco: "There's no more architecture diagram anymore; everything always changing; ppl don't even try anymore at Netflix
.@adrianco: "Problem with many OSS monitoring tools: great backend, but often front-end not as good commercial tools
.@adrianco: "Netflix use JMeter to do canary testing, post functional testing; compares old vs new (CPU, biz metrics, latency
RT @bridgetkromhout: Great "blip" story - @beerops "We didn't deploy anything using math.random to take down the site, but..."
@metaforsoftware: RT @SeenFeed: Just trended for #monitorama: "data, @tboubez tip" (20 tweets): http://t.co/7lU4wx9cQE
RT @TerribleDev: "Consider load testing your monitoring" @beerops
@bridgetkromhout: Thx for all the help on #devops survey, btw! The results are astonishing... I'll give you sneak peek?RT @hertling: My notes from Katherine Daniels' talk at #monitorama: http://t.co/gDP2VRQtre @beerops
@interrante: RT @hertling: My notes from Adrian Cockcroft's keynote at #monitorama: http://t.co/9alwc8AWj4
RT @interrante: RT @hertling: My notes from @adrianco's keynote at #monitorama: http://t.co/9alwc8AWj4
Yes! .@adrianco slides! "Please, no more Minutes, Milliseconds, Monoliths or Monitoring Tools! by http://t.co/gLK7Q3LAtX
@aneel: there's a guy hacking on @OpenStack during #monitorama in the row in front of me #startuplife
(Nice) RT @HypertextRanch: As your service gets better your probes get worse if you don't tune specificity.
RT @selenamarie: PONY WANTED: something like nagios, but runs diagnostic routines in resp to an alert before paging #monitorama @danslimmon
RT @hertling: My notes from Dan Slimmon on Monitoring at #monitorama: http://t.co/iW186S2wdc @danslimmon