Observability and Complex Systems (devopsdays AMS)

Distributed systems, microservices, containers and schedulers, polyglot persistence .. modern infrastructure patterns are fluid and dynamic, chaotic and transient. So why are we still using LAMP-stack era tools to debug and monitor them? We’ll cover some of the many shortcomings of traditional metrics and logs (and APM tools backed by metrics or logs), and show how complexity is their kryptonite.

So how do we handle the coming complexity Armageddon? It starts with a more modern approaches to observability, such as client-side structured data and event-driven debugging, distributed tracing, and more; no matter how many tags you add to a metric store, it still can’t tell a story like events can. It also means shifting perspective away from “monitoring” and to “instrumentation”.

Most problems are transient or self-healing, and you cannot possibly alert on (or even predict) the long tail of possible partial failures. So you need to turn finding arbitrarily complex causes into a support problem, not an engineering problem. How? Well … that’s the fun part.

Transcript

1. Charity Majors @mipsytipsy Observability & Complex Systems What got you here won't get you there, and other terrifying true tales from the computing frontier
2. @mipsytipsy engineer/cofounder/CEO https://charity.wtf “the only good diff is a red diff”
3. A short partial list of things I would like to touch on... "chaos engineering" you must be this tall to ride this ride. (are you? how do you evaluate this?) observability business intelligence, aka why nothing we are doing is remotely new why tools create silos the implications of democratizing access to data particularly for levels and career progressions how deploys must change the mis allocation of internal tooling energy away rom deploy software why you need to test in prod why you need a canary (probably) when to know you need a canary
4. "chaos engineering" you must be this tall to ride this ride. (are you? how do you evaluate this?) business intelligence, aka why nothing we are doing is remotely new why tools create silos the implications of democratizing access to data particularly for levels and career progressions how deploys must change the mis allocation of internal tooling energy away rom deploy software why you need to test in prod why you need a canary (probably) when to know you need a canary why you definitely need feature flags, no matter what test doesn't mean what you think it means continued ...
5. the future of development is observavbility-observability-driven development. "O-D-D yeah YOU KNOW ME" why we have to stop leaning on intuition and tribal knowledge before it is too late why AIOps is stupid and doomed why the team is your best source o wisdom why wisdom is not truth why ops needs to learn about design principles, stat why vendors are rushing to coopt the observability message before you notice they don't actually fulfill the demands, and why this makes me Very Stabby cont'd ... just a brief outline
6. "How did we get here?"
7. Monitoring (time series databases, dashboards, 'metric' tools) Logs (messy ass strings, really) More recently, APM and tracing. The trifecta:
8. "What do we need to get where we're going?"
9. Our idea of what the software development lifecycle even looks like is overdue for an upgrade in the era of distributed systems.
10. Deploying code is not a binary switch. Deploying code is a process of increasing your confidence in your code.
11. Development Production deploy
12. Observability Development Production
13. Observability Development Production
14. why now?
15. “Complexity is increasing” - Science
16. Architectural complexity Parse, 2015 LAMP stack, 2005
17. monitoring => observability known unknowns => unknown unknowns LAMP stack => distributed systems
18. We are all distributed systems engineers now the unknowns outstrip the knowns why does this matter more and more?
19. Distributed systems are particularly hostile to being cloned or imitated (or monitored). (clients, concurrency, chaotic traffic patterns, edge cases …)
20. Distributed systems have an infinitely long list of almost-impossible failure scenarios that make staging environments particularly worthless. this is a black hole for engineering time
21. Operational literacy Is not a nice-to-have
22. Without observability, you don't have "chaos engineering". You just have chaos. So what is observability?
23. Observability is NOT the same as monitoring.
24. @grepory, Monitorama 2016 “Monitoring is dead.” “Monitoring systems have not changed significantly in 20 years and has fallen behind the way we build software. Our software is now large distributed systems made up of many non-uniform interacting components while the core functionality of monitoring systems has stagnated.”
25. Observability “In control theory, observability is a measure of how well internal states of a system can be inferred from knowledge of its external outputs. The observability and controllability of a system are mathematical duals." — wikipedia … translate??!?
26. Can you understand what’s happening inside your systems, just by asking questions from the outside? Can you debug your code and its behavior using its output? Can you answer new questions without shipping new code? Observability... for software engineers:
27. Monitoring Represents the world from the perspective of a third party, and describes the health of the system and/or its components in aggregate. Observability Describes the world from the first-person perspective of the software, executing each request. Software explaining itself from the inside out.
28. We don’t *know* what the questions are, all we have are unreliable symptoms or reports. Complexity is exploding everywhere, but our tools are designed for a predictable world. As soon as we know the question, we usually know the answer too.
29. Welcome to distributed systems. it’s probably fine. (it might be fine?)
30. Many catastrophic states exist at any given time. Your system is never entirely ‘up’
31. Distributed systems have an infinitely long list of almost-impossible failure scenarios that make staging environments particularly worthless. this is a black hole for engineering time
32. You do it. You have to do it. Do it well.
33. Let’s try some examples! Can you quickly and reliably track down problems like these?
34. The app tier capacity is exceeded. Maybe we rolled out a build with a perf regression, or maybe some app instances are down. DB queries are slower than normal. Maybe we deployed a bad new query, or there is lock contention. Errors or latency are high. We will look at several dashboards that reflect common root causes, and one of them will show us why. “Photos are loading slowly for some people. Why?” Monitoring (old-school LAMP stack) monitor these things
35. “Photos are loading slowly for some people. Why?” (microservices) Any microservices running on c2.4xlarge instances and PIOPS storage in us-east-1b has a 1/20 chance of running on degraded hardware, and will take 20x longer to complete for requests that hit the disk with a blocking call. This disproportionately impacts people looking at older archives due to our fanout model. Canadian users who are using the French language pack on the iPad running iOS 9, are hitting a firmware condition which makes it fail saving to local cache … which is why it FEELS like photos are loading slowly Our newest SDK makes db queries sequentially if the developer has enabled an optional feature flag. Working as intended; the reporters all had debug mode enabled. But flag should be renamed for clarity sake. wtf do i ‘monitor’ for?! Monitoring?!?
36. Problems Symptoms "I have twenty microservices and a sharded db and three other data stores across three regions, and everything seems to be getting a little bit slower over the past two weeks but nothing has changed that we know of, and oddly, latency is usually back to the historical norm on Tuesdays. “All twenty app micro services have 10% of available nodes enter a simultaneous crash loop cycle, about five times a day, at unpredictable intervals. They have nothing in common afaik and it doesn’t seem to impact the stateful services. It clears up before we can debug it, every time.” “Our users can compose their own queries that we execute server-side, and we don’t surface it to them when they are accidentally doing full table scans or even multiple full table scans, so they blame us.” Observability (microservices)
37. Still More Symptoms “Several users in Romania and Eastern Europe are complaining that all push notifications have been down for them … for days.” “Disney is complaining that once in a while, but not always, they don’t see the photo they expected to see — they see someone else’s photo! When they refresh, it’s fixed. Actually, we’ve had a few other people report this too, we just didn’t believe them.” “Sometimes a bot takes off, or an app is featured on the iTunes store, and it takes us a long long time to track down which app or user is generating disproportionate pressure on shared components of our system (esp databases). It’s different every time.” Observability “We run a platform, and it’s hard to programmatically distinguish between problems that users are inflicting themselves and problems in our own code, since they all manifest as the same errors or timeouts." (microservices)
38. These are all unknown-unknowns that may have never happened before, or ever happen again (They are also the overwhelming majority of what you have to care about for the rest of your life.)
39. Three principles of software ownership: They who write the code Can and should deploy their code And watch it run it in production. (**and be on call for it)
40. When healthy teams with good cultural values and leadership alignment try to adopt software ownership and fail, the cause is usually an observability gap.
41. Software engineers spend too much time looking at code in elaborately falsified environments, and not enough time observing it in the real world. Tighten feedback loops. Give developers the observability tooling they need to become fluent in production and to debug their own systems. We aren’t “writing code”. We are “building systems”.
42. Observability for SWEs and the Future™ well-instrumented high cardinality high dimensionality event-driven structured well-owned sampled tested in prod.
43. Watch it run in production. Accept no substitute. Get used to observing your systems when they AREN’T on fire
44. Real data Real users Real traffic Real scale Real concurrency Real network Real deploys Real unpredictabilities.
45. You care about each and every tree, not the forest. "The health of the system no longer really matters" -- me
46. Zero users care what the “system” health is All users care about THEIR experience. Nines don’t matter if users aren’t happy. Nines don’t matter if users aren’t happy. Nines don’t matter if users aren’t happy. Nines don’t matter if users aren’t happy. Nines don’t matter if users aren’t happy.
47. well-instrumented high cardinality high dimensionality event-driven structured well-owned sampled tested in prod. Observability for SWEs and the Future™
48. You win … Drastically fewer paging alerts!
49. Charity Majors @mipsytipsy
50. • Srecon Charity Majors @mipsytipsy
51. You must be able to break down by 1/millions and THEN by anything/everything else High cardinality is not a nice-to-have ‘Platform problems’ are now everybody’s problems