Skip to content

Instantly share code, notes, and snippets.

@nrjones8
Created April 13, 2015 18:46
Show Gist options
  • Star 0 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save nrjones8/21deaa051ad0ffe6f718 to your computer and use it in GitHub Desktop.
Save nrjones8/21deaa051ad0ffe6f718 to your computer and use it in GitHub Desktop.
Notes from a few talks at Pycon

Building Secure Systems -- lvh

  • High level
  • Bugs --> security bugs --> particularly bad
  • "Tools" don't work for security
  • Unit tests don't catch it
  • Using some software...read the docs?
  • Google/SO don't always have best answers
  • Good practice ~= bad practice
  • Process is different in security:
    • Install it
    • Try to break it
    • Learn and repeat
  • Usually try to "make it work"; security is trying to "make if fail"
  • Not considered a priority -- features > security bugs
  • "Good practice looks just like bad practice"
  • Sony as an example -- people don't care about security, not economically motivating
  • Not something that's taught
  • Principle of "least authority" -- don't give power that's not needed
  • How do you make systems secure? Adversial! red/blue team
  • Some automated testing can catch security
  • "Attack surfaces" -- where can it be attacked?
  • "Threat model" -- what are you trying to prevent?
  • Example smartphone threat model
    • Contains everything
    • Protected by...a lock screen (passcode, fingerprint)
    • Fingerprints are on everything you touch!
    • Passcode can just be watched So...which is a better route? Use a threat model. Depends who you're worried about.
  • Random people? touch ID
  • Government? passcode might be better.

HTTPS: A Comedy of Errors Ashwini Orugati

  • Passive (just need the ability to listen) vs. active (misdirecting traffic/changing data -- MITM) attacks

  • SSL (old and busted) vs. TLS (probably what you mean)

  • Authentication via certificates

  • Certificate validation -- can we trust a site?

  • Server: certificate signed by CA (intermediary)

  • Not just authentication -- want to make sure site is who you think it is

  • TLS session:

    • Handshake (more detailed than this)
    • Have to agree on a bunch of details
    • Now traffic can be
  • Implementations: OpenSSL, BoringSSL, Secure Transport. They all have bugs/flaws.

  • API design sucks

  • Protocol flaws (downgrade attacks in establishing details)

  • Cookie stealing -- cookies always sent; secure flag on cookies?

  • Cookie injection

  • Users will click through certificate warnings

  • https://hynek.me/articles/hardening-your-web-servers-ssl-ciphers/

  • Tangled Web

Hyperactive: HTTP/2 and Python (Cory Benfield)

  • requests core-contributor

  • lukasa

  • Early implementer of HTTP/2

  • 1.1 is old -- first a standard in 1996

  • Our internet is different now -- 600 bytes of HTML no longer what we use

    • Inefficient use of TCP
    • Mostly doesn't reuse connections
    • Requires lots of concurrent connections
    • Too many roundtrips!
    • Not designed to be "real time"
    • Lots of hacking/misuse to make it fast
  • How'd we get to HTTP/2? Political discussion, not technical.

  • Differences?

    • HTTP/1.1 -- Text easy to debug, but complex to parse
    • Binary protocol: faster, easier to parse
    • Just a different wire format
    • Multiplexing
      • One TCP connection can carry many concurrent requests
      • "stream" -- single request/response (part of connection..?)
      • Priority/flow control of streams (does some resource depend on another one?)
    • Header compression (headers are big, let's fix taht. Why not use gzip? Could, but structured data would be better. HPACK designed specifically more HTTP headers
    • Early stream termination
    • Server push ("headline") -- can send responses to requests in advance. For priming caches.
    • Why not use server push? Server sends "request" that it anticipates to respond to
  • Poul-Henning Kamp -- HTTP/2.0 sucks. lol.

  • What's wrong with HTTP/2?

    • Difficult to reason about; hard to debug.
    • MAINTAINS STATE!
    • Interrupt testing for debugging
    • Awkward/terrible edge cases for backwards compatability with HTTP/1.1
    • Inherently concurrent -- makes it quite a bit trickier
    • asyncio <--> HTTP/2
  • https://http2.golang.org/

  • Lots of implementations of HTTP/2

  • Twitter, Chrome, Firefox all using it

  • Python -- hyper client library: httplib equivalent. Lower level.

  • Ideally, will put hyper under urllib3 under requests

  • http://daniel.haxx.se/http2/

  • https://http2.github.io/faq/

  • Current implementations

  • More widely used than IPv6. Poor IPv6.

Smart Services & Smart Clients (Frank Stratton)

  • Runscope -- (hosted) tests for APIs. Fancy pingdom.
  • Traffic inspector (middleware...?)
  • 60 services with only 8 engineers
  • But...why?
    • Scaling infrastructure
    • Scaling team (...how do microservices help this?)
  • Independent
    • Different codebases
    • Different deploys
    • Different owners
  • Microservices = SOA + DevOps
  • How should they communicate?
  • How do you "find a service?" "Smart client"

Smart Client

  • A wrapper around requests
  • Accepts "service URLs" (serivce://identity/...)
  • Retries certain requests
  • Can run requests asynchronously
  • "Atlas" -- register service metadata in Zookeeper & Dynamo

Smart Service

  • Built on top of Flask
  • Where to get API keys etc.?
  • SmartConfig -- gets keys from Atlas
  • Common logging config
  • Service skeletons with default values/configurations

Service automations

  • Puppet

  • "invest in it"

  • They also use Go, services can be language agnostic (well, anything that talks HTTP).

  • No shared databases between services; makes services dependent on one another.

  • One click deploys

  • 50 deploys/day (seems...excessive) -- to production or in general?

  • Some sort of "integration testing", but happens in production

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment