際際滷

際際滷Share a Scribd company logo

Ottimizzare le performance dell'API Server K8s come utilizzare cache e eventi per implementare il Change Data Capture

Luigi Fugaro
Luigi Fugaro

Il serve API di Kubernetes 竪 un componente fondamentale per la gestione dei cluster e l'interazione con le risorse del cluster. Tuttavia, le richieste ripetitive e le risposte voluminose possono causare problemi di prestazioni. In questo talk esploreremo come utilizzare i meccanismi di caching per migliorare le prestazioni del server API di Kubernetes e come utilizzare gli eventi per invalidare la cache in modo efficiente. Verranno presentati esempi concreti di implementazione e verranno discusse le best practice per l'utilizzo di questi meccanismi nell'ambiente di produzione. Imparare a utilizzare questi strumenti pu嘆 aiutare a ridurre i tempi di risposta e aumentare la scalabilit del cluster.

1 of 19
OTTIMIZZARE LE PERFORMANCE DELL'API SERVER K8S: COME UTILIZZARE CACHE ED EVENTI PER IMPLEMENTARE IL CHANGE DATA CAPTURE By Diego Braga, Product Manager @ Krateo PlatformOps & Luigi Fugaro, Sr. Solution Architect @ Redis
STOP QUERYING AND START CACHING - Kubernetes API server
Ottimizzare le performance dell'API Server K8s come utilizzare cache e eventi per implementare il Change Data Capture
WHY ARE WE PRESENTING THIS TALK? Kubernetes is becoming the de facto industrial standard for orchestrating cloud-native applications. What if we want to treat Kubernetes as a platform that is interrogated by different clients via API? This could generate a lot of requests to the API server. Kubernetes doesn't offer caching out of the box, so what could be the best pattern to implement it?
WHAT IS A CACHE Prompt: "Briefly describe what is a cache." Response: "A cache is a hardware or software component that stores data so future requests for that data can be served faster. The data stored in a cache might be the results of an earlier computation, or duplicates of data stored elsewhere. A cache hit occurs when the requested data can be found in a cache, while a cache miss occurs when it cannot." Caches are utilized because the data stored in a cache can often be delivered faster than re-fetching or recomputing the original data. This is due to the principles of the locality of reference, where programs access data and instructions in localized areas. Common types of caches include CPU caches, web caches, and database caches. By storing frequently accessed information, caches reduce data access latency, decrease bandwidth consumption, and improve overall system performance.
CACHE-ASIDE READ-THROUGH WRITE-THROUGH Cache-miss managed at application level. Cache-miss managed at cache provider level. Write data in the cache and automatically update the db. SYNC CACHING PATTERNS WRITE-BEHIND Write data in the cache and automatically update the db. ASYNC
CHANGE DATA CAPTURE Prompt: "Briefly describe what is change data capture." Response: "Change Data Capture (CDC) is a design pattern that enables tracking and capturing changes in data so that other software (like ETL tools, data replication tools, or data auditing tools) can respond to those changes. The changes might include insertions, updates, and deletions that are performed on a database table." AUTOMATION OFFLOADING CQRS DESIGN PATTERN
KUBERNETES API SERVER The API server is a component of the Kubernetes control plane that exposes the Kubernetes API. The API server is the front end for the Kubernetes control plane
KUBERNETES INFORMERS As the center of Kubernetes communication components, API Server receives requests, contacts etcd, and manages all the resource objects. So it would be under pressure easily. To ease API Servers pressure, Kubernetes employed a message system called Informer, which guarantees message delivery and provides high- performance real-time delivery, like any other famous message system. Informers query the resource data and store it in a local cache. Once stored, an event is only generated when it detects a change in the object (or resource) state.
KUBERNETES CLIENT The following client libraries are officially maintained by Kubernetes SIG API Machinery. C DOTNET GO HASKELL JAVA JAVASCRIPT PERL PYTHON RUBY
USE CASE Leveraging client-javascript, we wrote a backend that exposes an API to query Kubernetes resources: via query parameter is it possible to use Redis as cache or interact directly with the API server
USE CASE ARCHITECTURE CACHE WARM UP When the Informer is instantiated, all the existing resources are pushed into the cache AVOID CACHE MISS The client will always find an entry in the cache if the resource exists in Kubernetes CACHE UPDATE The Change Data Capture is implemented via the Informer that is notified for every resource change from the API server - and the microservice will set, update or delete the key 2 3 1 Client Webserver (express.js) Cache Kubernetes API server Informer (client-javascript) /cache=true /cache=false
CODE SNIPPET A new Informer is instantiated for any kind of Resource available in the cluster (namespaces or cluster-scoped) and it interacts directly with the cache
CODE SNIPPET If the 'cache' query parameter is set to 'true' (or not specified), the webservice will inquiry the cache, otherwise it will call directly the API server
LOG SNIPPET When a new Kubernetes Resource is created / updated / deleted (i.e. via 'kubectl'), the Change Data Capture pattern is implemented with the Informer that creates / updates / deletes the key in the cache
MAX No Cache MAX Cache 0 1000 2000 3000 4000 5000 10 requests 50 requests 100 requests 150 requests 200 requests USE CASE RESULTS Max 10 concurrent requests to inquiry a single resource Duration of 30 seconds
AVG No Cache AVG Cache 0.00 500.00 1000.00 1500.00 2000.00 10 requests 50 requests 100 requests 150 requests 200 requests USE CASE RESULTS Max 10 concurrent requests to inquiry a single resource Duration of 30 seconds
CONCLUSIONS Kubernetes is not only an orchestrator for containerized applications but a platform that can be elevated by leveraging all the existing features. One feature that is missing is caching for clients that interact with Kubernetes as a backend that exposes resources via API. Change Data Capture is a pattern that is already implemented in Kubernetes client libraries with the Informer object. There are different caching patterns that can be implemented at the application level or cache provider level. Placing a cache between the client and the API server can enhance latencies by one order of magnitude.
THANK YOU

More Related Content

Ottimizzare le performance dell'API Server K8s come utilizzare cache e eventi per implementare il Change Data Capture

  • 1. OTTIMIZZARE LE PERFORMANCE DELL'API SERVER K8S: COME UTILIZZARE CACHE ED EVENTI PER IMPLEMENTARE IL CHANGE DATA CAPTURE By Diego Braga, Product Manager @ Krateo PlatformOps & Luigi Fugaro, Sr. Solution Architect @ Redis
  • 2. STOP QUERYING AND START CACHING - Kubernetes API server
  • 4. WHY ARE WE PRESENTING THIS TALK? Kubernetes is becoming the de facto industrial standard for orchestrating cloud-native applications. What if we want to treat Kubernetes as a platform that is interrogated by different clients via API? This could generate a lot of requests to the API server. Kubernetes doesn't offer caching out of the box, so what could be the best pattern to implement it?
  • 5. WHAT IS A CACHE Prompt: "Briefly describe what is a cache." Response: "A cache is a hardware or software component that stores data so future requests for that data can be served faster. The data stored in a cache might be the results of an earlier computation, or duplicates of data stored elsewhere. A cache hit occurs when the requested data can be found in a cache, while a cache miss occurs when it cannot." Caches are utilized because the data stored in a cache can often be delivered faster than re-fetching or recomputing the original data. This is due to the principles of the locality of reference, where programs access data and instructions in localized areas. Common types of caches include CPU caches, web caches, and database caches. By storing frequently accessed information, caches reduce data access latency, decrease bandwidth consumption, and improve overall system performance.
  • 6. CACHE-ASIDE READ-THROUGH WRITE-THROUGH Cache-miss managed at application level. Cache-miss managed at cache provider level. Write data in the cache and automatically update the db. SYNC CACHING PATTERNS WRITE-BEHIND Write data in the cache and automatically update the db. ASYNC
  • 7. CHANGE DATA CAPTURE Prompt: "Briefly describe what is change data capture." Response: "Change Data Capture (CDC) is a design pattern that enables tracking and capturing changes in data so that other software (like ETL tools, data replication tools, or data auditing tools) can respond to those changes. The changes might include insertions, updates, and deletions that are performed on a database table." AUTOMATION OFFLOADING CQRS DESIGN PATTERN
  • 8. KUBERNETES API SERVER The API server is a component of the Kubernetes control plane that exposes the Kubernetes API. The API server is the front end for the Kubernetes control plane
  • 9. KUBERNETES INFORMERS As the center of Kubernetes communication components, API Server receives requests, contacts etcd, and manages all the resource objects. So it would be under pressure easily. To ease API Servers pressure, Kubernetes employed a message system called Informer, which guarantees message delivery and provides high- performance real-time delivery, like any other famous message system. Informers query the resource data and store it in a local cache. Once stored, an event is only generated when it detects a change in the object (or resource) state.
  • 10. KUBERNETES CLIENT The following client libraries are officially maintained by Kubernetes SIG API Machinery. C DOTNET GO HASKELL JAVA JAVASCRIPT PERL PYTHON RUBY
  • 11. USE CASE Leveraging client-javascript, we wrote a backend that exposes an API to query Kubernetes resources: via query parameter is it possible to use Redis as cache or interact directly with the API server
  • 12. USE CASE ARCHITECTURE CACHE WARM UP When the Informer is instantiated, all the existing resources are pushed into the cache AVOID CACHE MISS The client will always find an entry in the cache if the resource exists in Kubernetes CACHE UPDATE The Change Data Capture is implemented via the Informer that is notified for every resource change from the API server - and the microservice will set, update or delete the key 2 3 1 Client Webserver (express.js) Cache Kubernetes API server Informer (client-javascript) /cache=true /cache=false
  • 13. CODE SNIPPET A new Informer is instantiated for any kind of Resource available in the cluster (namespaces or cluster-scoped) and it interacts directly with the cache
  • 14. CODE SNIPPET If the 'cache' query parameter is set to 'true' (or not specified), the webservice will inquiry the cache, otherwise it will call directly the API server
  • 15. LOG SNIPPET When a new Kubernetes Resource is created / updated / deleted (i.e. via 'kubectl'), the Change Data Capture pattern is implemented with the Informer that creates / updates / deletes the key in the cache
  • 16. MAX No Cache MAX Cache 0 1000 2000 3000 4000 5000 10 requests 50 requests 100 requests 150 requests 200 requests USE CASE RESULTS Max 10 concurrent requests to inquiry a single resource Duration of 30 seconds
  • 17. AVG No Cache AVG Cache 0.00 500.00 1000.00 1500.00 2000.00 10 requests 50 requests 100 requests 150 requests 200 requests USE CASE RESULTS Max 10 concurrent requests to inquiry a single resource Duration of 30 seconds
  • 18. CONCLUSIONS Kubernetes is not only an orchestrator for containerized applications but a platform that can be elevated by leveraging all the existing features. One feature that is missing is caching for clients that interact with Kubernetes as a backend that exposes resources via API. Change Data Capture is a pattern that is already implemented in Kubernetes client libraries with the Informer object. There are different caching patterns that can be implemented at the application level or cache provider level. Placing a cache between the client and the API server can enhance latencies by one order of magnitude.