RubyConf 2018 Notes

rubyconf.md

RubyConf 2018 Notes

Keynote by Matz (Ruby Creator)

• Why do people create programming languages?
• C was designed to replace assembly to write OS
• Matz wrote Ruby for fun
• People write programming languages, why not me
• Japanese aesthetic/minimalism in Ruby

The games developers play (Andy Croll)

• Punitive, parental seniority can be used to protect codebase
• Take time to build mental health habits everyday
• Your adult can be a better arbiter based on keystone habits
• Sitting with discomfort necessary for learning and growing
• Examine overpowering feelings before acting on them

Graphics/Simulations (Ryan Davis)

• Author of Graphics gem in Ruby
• Primarily for graphics simulations but can also be used for game dev
• Generative art demos

It's Down! Simulating Incidents in Production (Kelsey Pedersen)

• TLDR; Stitch fix built a Faraday middleware to replace http calls from downstream services with 500s for certain systems to test how their service would react
• List use cases in runbook
  • Did app load
  • Can you view inventory/details
  • Did alerts go off
• Called game day exercies, entire remote team monitoring system behavior
• Gated by feature flags so can rollback anytime
• The service failed in unexpected way
• Also uncovered a lot lacking in their monitoring/alerting strategy

Uncoupling Systems (Jeremy Hanna)

• Separate but interconnected tower of responsibilities
• Anti pattern: Not discussing trade-offs of decisions
• Strategies
  • Async calls
  • Load balance the work
  • Replicate Data
  • Read through cache, explicitly tell when to expire and when to rebuild
    • Invalidation, stale data
  • Batch IO work in threadbare workloads
• events != streams
• last write wins, if can ensure that, process queue last in first out
• determining systems of record and derive meaning can get hard
• Out of band precompute as much as possible
• id where data is derived
• retryable
• Publish an artifact instead of redoing the work
• improvement over table scan

Optimizations in multiple dimensions (Jamie Gaskins)

• Optimization benefits/drawbacks may not be linear
• To monitor
  • Systems
    • RAM consumption
    • App bootime, delays (minimize restarts)
    • time for each transaction (latency)
    • transactions per second (throughput)
  • Team level
    • Time betn feature inception to release
    • Time betn deploys per developer
    • Time between bug discovery and bug fix deployment
• Optimizations
  • CPU bound
    • parallelization
    • caching
    • Frequency of GC runs
  • I/O
    • Remote APIs, persistent connections
    • Remote cache
    • Remote Filesystem
    • Databases
      • (Vacuum/Index)
      • Great if enough memory to fit whole dataset
• Latency vs Throughput
  • Most important metric at any given time is subjective
  • If latency is already good, only then we can handle throughput
• Algorithms, optimize for RAM/CPU space time complexity
  • Memory allocation within Ruby. 16kb page of memory. All ruby object live in those page
  • Ruby allocates a batch of RAM, and optimize that
  • MRI reserve RAM
• Memoization — caching at object level
  • lifetime of process level
• Small vs large datasets
  • If optimized for large datasets. MRI hash uses different underlying data structure for 3 keys vs more
  • Ruby sort mergesort for small, quick sort for large
• Remote cache
  • Costs roundtrip
  • Cache is about mitigating costs
  • App remote network latency to remote cache
  • 10 ms of cpu time not same as 10 ms of wall clock time for throughput reasons
  • Cache hit rate — want over 99%
  • When are we invalidating too little vs too much
  • If less than 90% util, cache isn't being useful
• Question if abstractions like AR help
• Perf under load vs perf at idle (prod workload)
  • Optimize for production workloads

Cache is King (Molly Struve)

• Batch processing
• Avoiding n+1 queries
• local cache
• Divide into two classes
  • Ruby to replace data source
  • Bulk serialization, hash cache, framework cache
• Use ruby to prevent useless datastore hits
  • Database guards
• return unless user_ids.any? (No db hit)
• RDS cpu on fire, wasn’t long running queries
  • Lots of fast queries
• Serialization of vulnerabilities
  • Active Model serializers
  • Use Bulk Serialization in batches of thousands
  • Bulk serialization deploys (no of db hits)
• RDS CPU utilization
• Ruby process data in bulk
• Elastic search vulnerabilities indexed by client
• get all serialized vulnerability cache
• Ruby hash to cache the index name
• Call made now once per client not per vulnerability
• 65% job speed up
• Redis is so fast easy to forget its a network call
• Hash cash always faster
• Shard mysql db
• Active Record Connection
• Proxy object
• Know your gems
• Active record none can avoid db hit
• Empty dataset hit
• Goal: Prevent useless datastore hits
• Took reporting time down from 10 hours to 3 hours
• How many MB per second of data is getting read from Redis
• Redis network traffic spikes resolved

Talk on GC in Ruby (Colin Fulton)

• OOM errors
• Ruby does not free garbage collected objects
• Ruby allocates page of memory to avoid calling expensive malloc/free
• Ruby can still leak memory
• Ruby can leak when accidentally make a reference to something can’t be deleted

Reducing memory usage in ruby (Aaron Patterson)

• Same talk as in GoRuCo 2018
• Direct ISeq memory
• ObjectSpace allocation_tracer

Ruby Code Generation(Alex Stephen)

• Use Ruby in Google for Google Cloud management
• magic-modules gem
• Auto-generation: Write once, write anywhere
• You want as much logic in templates as possible
• not clever
• be verbose
• Alterations make to template get to final state

Standard (Lightning talk)

• Justin Searls is building an opinionated Rubocop wrapper!

No Title Required (Jim Liu)

• How can you have outsize influence without official leadership role?
• Leadership is something really personal, we all have own timelines
• If you can positively influence people, you can have most influence
• Build consensus, celebrate wins, push for change
• Leadership is not a static milestone
• Flywheels, serious of steps that built off of one another and result in higher performance
  • Id, fix, prevent similar classes, teach others, repeat
• Not your idea: Once idea is out there it becomes everyone’s
• Observations on leadership
  • title != ability to lead
  • Give equal weight to engineers
  • Don’t be subconsciously influenced by those ideas of levels
• Leadership is continuous practice of positive influence
• Incentives and values of organization influence what kind of leadership bubbles to the top.
• Lots of different leadership styles
• More inclusive type of leadership styles
• You cannot be passive about leadership
  • We as individuals have to advocate for yourself
  • If measurable impact, be super direct and call it out
• Busy work: Best most/impactful impact
  • Busy work in one domain doesn’t mean busy work for everyone
• Say no a lot
• Don’t talk, listen
• Strive for continuous improvement
• Honest assessment of your skills
  • What am I good at? What do I want to improve?
  • Hour each month to talk about longer term goals
  • Areas of success/development (Photo)
• Feedback
  • Evaluation of trust (Feedback)
  • Ask specific feedback
  • How were the technical aspects of my presentation to the leadership team last Tuesday?
  • eedback good cadence, habit of giving feedback
• Work
  • Dive deep, top customer problem
  • How that get translated into code
  • Really understand Requirement -> Spec -> Code
  • Team appreciate, hires appreciate
  • Dive deep enough to uncover what really is happening
  • Better systems and tools
  • Best practices, designs
• Extra work, read source code, read interview feedback
• What components do I need to get better at
• Technical parts super important
• Project management, Launch plan (how will we release/rollback), Stakeholder communications, Metrics
• Measure as often as you can
  • Track OKRs
  • Writing on paper and come back later
• Always mentor, new hires. Best leaders take ideas to harness
• Find mentors
• Someone trusts you enough to answer
• Physical and mental health

The New Manager's Toolkit (Brandon Hays)

• Talk to everyone
• Team feels no career progress
• Design engineering ladder
• Help define and defend team values
• Performance improvement plan
• Manager primary job is to be available (Office hours)
  • Listening means following up
  • Backlog what is next up
• Interpersonal conflict: Trust is critical
  • Find common ground
  • Book: Crucial conversations
  • Culture change process
  • Team offsite, find team to do, managers
  • The power of vulnerability
• Programmers, why are we so focused on brains in jars rather than taking care of our body
• If you want to be able to take responsibility for others, work on yourself first
• As IC not exposed to what is happening at management level, how do we make it easy?
• Being your authentic self is good enough
• You get to take another swing at it
• Leave this industry a little better than you found it
• Podcast: Managing Up

Eiffel's Tower (Nicholas Means)

• Every organization is political
• To influence change, you must engage and participate in politics of company
• Networking and self promotion
  • Reframe as making friends and telling stories
• Grab coffee with your PM, sales
• Self promote: Doing good work is sometimes not enough
• Tell boss what you have been upto
• Telling stories
  • Book: Dale Carnegie
• Negotiation
  • There is always a middle ground
  • Go into negotiating understanding everyone’s needs, including your own
• Don’t become chokepoint of information for your team
  • Can’t save from all shit umbrella
  • Blocking all noise is comfortable, but keep from delivering all value they are capable of
  • If they never see org politics, never learn build their ideas or sell them
  • Obligation to help your teams how to navigate the org