USE vs RED - which is the better methodology for analyzing the performance of microservices 🤔? 🧵
Begin observing the performance characteristics of apps before the production rollout using Google's four golden signals. #ShiftLeft ibm.com/garage/method/…
The RED method takes an externally-visible (request-driven) view of the workload. For every workload, it checks request rate, error-rate, time each request takes. weave.works/blog/the-red-m…
The USE method takes an internal, resource-centric view of the workload. For every resource, it checks utilization, saturation, and errors. The goal is to understand how these resources behave in the presence of the load. brendangregg.com/usemethod.html
➡️ Start with RED and treat the services as a black box. Best for request-driven apps.
➡️ Combine with USE to check stateful/non-request-based systems and components as white box.
➡️ If none helps, check logs and traces too. copyconstruct.medium.com/logs-and-metri…
• • •
Missing some Tweet in this thread? You can try to
force a refresh
Slack’s Kafka journey to operational excellence at scale! Context and lessons 👇 🧵
➡️0.7 petabytes of data
➡️10 Kafka clusters x90 nodes
➡️6.5 Gbps throughput
💪small-but-mighty ops team slack.engineering/building-self-…
Challenge➡️ solution
Automation ➡️Chef, Terraform, CI/CD to build, release Kafka & ZK
Hot spotting➡️make partition counts to be a multiple of the broker count
Replication➡️limit the replication bandwidth
Imbalance ➡️ Cruise Control
Tune clusters➡️Chaos engineering with dark traffic
Slow cluster recovery➡️Enable jumbo frames
Visibility into cluster metadata➡️ Kafka manager
Visibility into the health of the Kafka consumers➡️Kafka offset exporter
Islands of knowledge➡️One-page runbook
📕Another week, another book review📕
"Learning Dapr"
tl;dr: A different perspective on distributed systems from the creators of Dapr. Introduction to @daprdev combined with creators early insights in one book 🧵👇
Starts with a detailed introduction to Dapr philosophy and architecture: Dapr is a language-neutral, unified programming model that abstracts infrastructure details from developers
What are the cloud promises & challenges; and how Dapr simplifies the creation of portable cloud native applications. With examples for service invocation and pub/sub through Dapr
Airbnb’s Journey To Post-Microservices
(HINT: microservies didn't solve all the problems) medium.com/qe-unit/airbnb…
Airbnb started with a Monolithic architecture inside a Monorepo. This carried them far, to $2.6B revenue before observing limitations with
➡️Velocity of software changes
➡️Component ownership
2020: teams and revenue grew ($5B), the architecture
changed (to microservices), but some issues remained: features require changes on multiple services and different teams. 🤔
2022 State of the Java Ecosystem (@newrelic customers segment) 🧵👇
• The most used production version: Java 11🌟
• The most popular vendor? oracle ↘️ amazon 🚀
• The rise of containers, i.e. Kubernetes 🤘
• The most common heap size & garbage collector?
In a year, Java 11 became the new standard:
↘️Java 8 from 85% to 46%
↗️Java 11 from 11% to 48%
Non-LTS such as Java 14 bellow 1%
Oracle proprietary JDK shrinks, whereas OpenJDKs based distros such as Amazon 🚀