All Day DevOps is a 24 hours long, online conference about DevOps with 125 sessions in five tracks.
Note: these are my notes. They are not necessarily representative or complete summaries of the talks.
Container security monitoring using open source — Madhu Akula (AppSecCo)
Why is container security monitoring (CSM) important? An attacker could gain access to one of your containers. Or you may be using an image from the Docker Hub with malicious code. CSM can help you to take actions based on the monitoring alerts.
For CSM it is important to collect your logs on a central location. Then you can analyse them to find suspicious patterns. Based on the patterns we identified, we can set thresholds and create pattern to trigger alerts.
How to look at logs? E.g. using commands like “
docker events” or “
docker diff <container>” you can check which files have changed in your
container. Although not security specific, for system level monitoring you can
Sysdig Falco is not an enforcement tool, but an auditing and monitoring tool. With Falco you can load rules that trigger on events, e.g. when a shell was spawned or a sensitive file has been opened. Falco comes with rules, but you can also customize rules yourself.
You can even create an automated defence to e.g. deploy new versions when an alert triggers (as was shown in a demo).
7 Container design patterns you need to know — Christian Melendez (Equinix)
Most of your problems are not unique, both in software development and container usage. Having patterns makes it easier to communicate: they provide a shared language to known problems. Having patterns also allows you to reuse shared components (solutions).
Patterns for a single container
A base for other patterns.
- Establish resource boundaries
- Isolate resources from others
- Separate concerns — the container only has one responsibility
Extend an application container with extra tools.
- Co-schedule the sidecar in the same node (same pod in the case of Kubernetes)
- Shared resources (volumes)
The ambassador is a broker or proxy between an application and consumers.
- Reusable in other applications
An application container to modify the interface of the application to e.g. provide a consistent interface.
- Helps to set a communication standard
- Can translate between protocols
- Reusable modules, perhaps environment variables to configure behavior
- Examples: collect metrics, parse and store logs, custom health checks
Used to scale at the node level for loosely coupled applications. Communication is via network calls. They are the base for microservices.
Split a big task to into smaller tasks that can run simultaneous and combine the results afterwards.
- Partial processing
- Multiple threads for requests and process
- Response time is the sum of all requests
Leader election pattern
How to select a new leader when the old leader fails.
- Reuse, don’t invent your own wheel
- Container implementations like etcd, Zookeeper and consul
One manager container with multiple worker containers. The manger concurrently distributes the tasks to the workers.
How To Build An Effective Container-Based Local Development Environment — Martino Fornasa (Kiratech)
Having a local development workflow is important:
- Easier onboarding
- Consistent software dependency management
- Environment similar to production
- Local testing (faster than waiting for CI tool)
- You build knowledge
Creating a local environment is hard. Lots of variables to take into account like languages, preferences of developers, you need to keep allow a certain flexibility.
A first attempt led to issues:
- Sync issues
- Resources (CPU/memory)
And people hated it.
Goals for a proper solution:
- Tune for performance and reusability.
- Allow people to use their preferred tools
- Make it easy to do the right thing.
When your target is Kubernetes, your development workflow should also take this into account. Benefits of a Kubernetes workflow:
- Easier to setup complex environments
- Same/similar tools for development and deployment
- Each developer is able to run the full system (perhaps using external systems)
- Auto rebuild the image on each code change, push to registry and deploy in the cluster. Related tools: draft and skaffold. Skaffold is probably more convenient for interpreted languages.
- Run on cluster + sync the filesystem between the local system and the cluster. Higher risk of drift between local image and image built by your CI system. Tools: skaffold and ksync.
- Run locally + develop against remote resources. Tool: telepresence.
DevSecOps Toolchain: Blocking Dangerous Containers — Siamak Sadeghianfar (Red Hat) & Georgios Kryparos (Tink)
DevSecOps is a shift in both mindset (everyone is responsible for security) and security work (security as code).
We’ve gone from running applications on hardware to running them in a VM using a hypervisor. Then we started running applications in containers (which may or may not run in a VM). Now the world is even more complex with container orchestrators. In summary: many layers to take into account. And then we haven’t even touched on the registry for images…
Containers have brought good things (consistency from dev to production, faster deployments), but they have also brought security concerns:
- isolation breakouts
- outdated apps and dependencies
- secret management is an additional headache
- auditing and monitoring are still issues we need to deal with
You need vulnerability scanners, configuration auditors, secret scanners and traffic monitors on all levels: host, container engines, orchestrator and registry.
Tools you can use to help you:
- Automation! Shift security to the left.
- Shared visibility
- Train first, enforce later
- Collaboration between dev, sec and ops teams, but also e.g. quality assurance
Zero Gain vs. Zero Loss: Practical Tales of Enterprise Decision-Making — Amélie Koran (HHS)
Leadership would like to see as little loss as possible. Selling the value of what you do on a day-to-day basis to others in the organization is important.
A “play” is an action you can take to a affect a certain outcome. Examples: skills development and applicability of what you do to the mission of the business.
Play one: metaphors
Metaphors allow you to connect on a common touchstone. E.g. use physical security measures as a metaphor for “cyber” security measures; executives can conceptualize the former, but perhaps not the latter.
Don’t talk tech, talk about something they are familiar with. Examples: risk, saving money, spend a little to prevent a greater loss.
Play two: participation
Having a seat at the table is valuable. Think about design sprints, strategy sessions, budget planning sessions. This leads to deeper connections, reduces communication barriers, creates more empathy and enables awareness.
Play three: conceptualization
Defining stuff is important. Having a shared lexicon helps communication. The terms and definitions enable knowledge sharing.
Clear communication has all kinds of benefits like reducing questions, stress and anxiety.
Why do people want dashboards? They would love to have a single number to give them insight in how things are going. But what is the question that is to be answered? What decision should the dashboard support? Is what is being presented actionable?
- Aim to reduce user errors (via good UX/UI)
- Automate tests
- Don’t deploy to production, invest in development and test environments
- Perform root cause analysis for major issues
- Include the proper team members (for instance: include security people)
Monitoring the easy way — Daniel Barker (National Association of Insurance Commissioners)
- Based on the output the system state should be deterministic.
- Based on the input the system state should be deterministic.
Observability tools (in implementation order):
- Metrics aggregation
- Log aggregation
- Alerting and visualizations
- Distributed tracing (harder than the others to implement, not all systems need it)
Let’s have a look at each of those tools.
Metrics aggregation tools:
- Prometheus, used a lot with Kubernetes
The two last ones are great for long term storage for Prometheus.
OpenMetrics is a metrics exposition format. Implement your metrics once and different products can understand them. Datadog is currently the only commercial product that support this standard; it is still early days though.
Log aggregation tools:
- Splunk (commercial)
- ELK stack
- FluentD (not log aggregation, but you can use it to replace Logstash in your ELK stack because it is a bit more efficient)
- Assume your logs arrive in a random order in your log aggregation service so include a timestamp
- Use UTC for your timestamps
- Format logs in JSON
- Log all application errors, don’t eat them without logging them, otherwise you don’t find out about them until your users start complaining
- Turn logging on
- Make the log messages human readable because they are meant to be read by humans
- Do not log informational/debug data in production
- Do not log anything a human cannot read or react on
- Only log significant events
- Do not print personally identifiable information (PII)
Alerting and visualization
Only send alerts when action can be taken. Information is never an alert.
You would not go running around the town square yelling the weather.
Visualization: you should be able to look at it and understand the system better.
- Prometheus AlertManager
- StatsAgg (does a lot more than just alerting)
- Flame graphs
Track requests throughout the system
More information on monitoring.danbarker.codes.
Getting started with chaos engineering — Ana Medina (Gremlin)
Definition of chaos engineering: thoughtful, planned experiments designed to reveal the weakness in our systems.
Chaos engineering is not about unexpected or unmonitored experiments. It is also not about creating outages.
Like a vaccine, we inject harm to build immunity.
Use cases: outage reproduction, on-call training, strengthen products and battle test infrastructure.
Prerequisites for chaos engineering:
- Monitoring and observability in place
- On-call and incident management (a good SRE practice in general)
- Know the cost of downtime per hour
You can inject chaos on all layers. Nice places to do it are e.g. elasticsearch, cassandra, etc.
To get started with chaos engineering, pick one of the top 5 most critical systems. Whiteboard the system and determine what experiment you want to run (e.g. maximize CPU, kill processes, packet loss) and determine the blast radius (for example only impact 1% of your traffic).
Chaos day is a dedicated day where the whole company focuses on resilience There’s a tutorial on the Gremlin site if you want help to plan your own chaos day.
Examples of experiments:
- Reproduce actual outage
- Datacenter failure
- Large traffic spikes
- Maxing out CPU and disk
What is SRE and why every DevOps professional should care — Dave Rensin (Google)
How do we choose between DevOps and SRE?
- Reduce organizational silos
- Accept failure as normal
- Implement gradual changes
- Leverage tooling and automation
- Measure everything (otherwise you run your business by luck)
- Reliability is the most important feature of a system (every system, not just software).
- The only measure that actually matters is the reliability experienced by your users.
- 100% is always the wrong goal all the time. (Your users also don’t need 100% reliability, they won’t even notice it.)
The space between 100% and your availability target (e.g. 99.99%) is called the “error budget.” If you achieve higher reliability than you targeted for, you are wasting budget.
- A service level indicator tells us how our user is experiencing our system.
- A service level objective is an SLI plus a goal.
- A service level agreement, in contrast to SLI and SLO, points outward. SLA = SLO + margin + consequences. Purely a business discussion; there is no engineering involved.
- Metrics & monitoring (are we measuring and alerting on the right thing?)
- Capacity planning
- Change management (consistency is important because you want to automate this)
- Emergency response
- Culture (toil management, engineering alignment, blamelessness)
Why choose between SRE and Devops? It is not one or the other. SRE is a concrete, highly opinionated implementation of DevOps! (Also see the video class SRE implements DevOps.) It provides a consistent and optimized way of implementing DevOps principles.
Some nice quotes:
The closer to zero you can drive the blast radius of a mistake, the closer to infinity you can drive your risk.
(And SRE manages risk.)
SRE is what happens when you ask a software engineer to design operations.
Take the SLO and calculate your error budget. Then apply risk analysis and calculate how many bad minutes/year something cost (or could cost). Then you start working on the items that are > 25% of your error budget. Tool to play with this: CRE Risk Analysis Template
How to start with SRE? The most important thing is willingness to do SRE. You cannot adopt SRE unless you have executive buy-in. (Note that you ‘only’ need buy-in, it does not work if executives enforce implementation.)
The next important thing is to pick one application (defined as a “discrete failure domain”) first. This will reduce the number of people you need buy-in from. But it has to be a meaningful application, otherwise the lessons you learn won’t be internalized. Once the reliability and velocity are increasing and the frustration between dev and ops is decreasing, others will want to do it too!
When you have the willingness and know which application to work on, you start with defining your SLIs, SLOs and error budgets. Answer the question “what do my customers care about?” Pick the smallest number of things that answer that question. So not “CPU utilization” but e.g. “latency of the application.” Then determine the first set of objectives. Also determine a policy, for instance: feature freezes until you go below the error budget.
Note that the SLOs don’t have to be perfect. You can re-evaluate them periodically (monthly, quarterly for instance).
Now, before you do any engineering, get your alerting and monitoring in order. Make sure you monitor the things that are relevant for your SLOs. Usually you can turn off half of your alerts without impacting the experience of your user. More alerting is almost never better. More logging and measurements are (probably) better. Simple test: only alert on something your customer would want to be alerted on.
An important prerequisite for doing SRE is having a blameless culture! The highest return on investment of a mistake is if you learn as much as possible from it. You can only do that in a blameless culture.
Take one step at a time:
- Define SLO, SLOs and error budgets
- Audit and adjust monitoring and alerting
- Model and reward blameless postmortems
If you do this, you will improve things!
More information: google.com/sre
DevSecOps - securing a great developer experience — Stefan Streichsbier (GuardRails)
Some stats: 51% of the world’s population has internet access (June 2017) and there are 31 million developers on GitHub (October 2018).
In the future every company will become a software company — Marc Andreessen
Having a website used to be simple: you uploaded the needed files to your hosting provider via FTP. There was no to little collaboration (just one “web master”), no build steps (PHP, ASP, etc) and you tested by checking if there were parse errors.
Now there is a whole bunch of additional (and different) steps required. There are code repositories, build pipelines, several environments (test, production). Deployment on e.g. a Kubernetes cluster is much more complex than putting some files somewhere. And then there’s the complexity of all the development tools and all the products that the cloud providers offer.
So how does security fit in?
I’ve worked extensively with AWS over the last 4 years, and I can barely wrap my head around the scope of managing security in AWS. We have an entire department dedicated to security in our company, and none of them are remotely close to being experts in AWS security either.
I’m starting to get curious if there even is an expert who could set up and maintain a bulletproof AWS Account.
Steps in the evolution of security:
- Penetration testing
- Secure software development life cycle
Going to DevSecOps is not easy for typical security teams. They have to get into the mindset of software development. Security tools are typically also not made with developers in mind.
Tools are not everything. We need to have a proper workflow around it. We should not “automate the nagging” by building in security tools in the development pipeline.
Only a fraction of the 31 million developers on GitHub have access to (enterprise grade) security tools and penetration testing.
Quality had to face some of the challenges security is currently facing. Quality is now more and more a first class citizen.
- Usability: can the user accomplish his/her goal very effectively when they interface with a product?
- User experience: everything around using a certain product
- Developer experience: user experience where your audience is exclusively developers.
To increase the developer experience with regards to security we have to keep some things in mind:
- Signal vs noise
- Security tools should not add more noise.
- Instead it should focus on high impact issues.
- Make sure that issues are reported with a high level of accuracy.
- Lost in translation
- Security has to speak the same language as developers. Otherwise developers have to become security experts as well to understand what the security team is talking about.
- Use the right communication channel. So not a PDF report or Excel sheet per email, but notifications in the workflow (e.g. in a pull request, issue, chat application).
- Describe the issues clear so that developer know how to solve them.
- Make it easy
- Make it simple to get started and try out (not: “schedule an appointment with our sales rep)
- Tight integration with the workflow
- Provide all functionality in one tool.
How to make security a part of DevOps?
- Acknowledge that developers are key. Tools should bring them joy using it and provide value.
- Respect that security is a complex topic but provide clear focus and make it easy to consume for developers.
- Security has to become a commodity, not just for big enterprises.