Alibaba Cloud CDS-SLS Uncovers Secrets
This article, the second in a series of articles about the upcoming hybrid cloud product CDS-SLS (Cloud Defined Storage Simple Log Service), mainly provides an overview of CDS-SLS function points. As the carrier of digitalization, logs contain the operation information of background programs, business operations and other information.
The query and analysis of logs has evolved from the manual pssh+grep of the earliest machines to small-scale ELK or EFK (Elastic search, Logstash/Filebeat/Fluentd, Kibana) of each business. In addition to Kafka, if there is a need to collect Metrics, you need to add Collected, if there is a need for visualization, you need to add Grafana, and if there is a need for storage and backup, you will also introduce Ceph, For the consistency management of the basic configuration, Salt is introduced, and the hardware cost and operation and maintenance cost increase rapidly with each additional component in the high concurrency scenario.
As a cloud based log platform, CDS-SLS provides these components with high cohesion and low coupling. Offline users can automatically deploy all of the above functions on at least six machines. It effectively solves the pain points in traditional software in the fields of operation and maintenance, operations, financial management, data analysis reports and other big data scenarios in a low code mode.
CDS (Cloud Defined Storage) cloud defined storage. As an output form of software defined storage (SDS), CDS has a unified storage architecture and user experience with the public cloud, reduces the number of pedestals, provides flexible deployment scale and form, integrates multiple storage products, and provides operation, maintenance and management of enterprise storage.
CDS supports a mix of storage products, such as CDS-OSS+CDS-SLS, CDS-EBS+CDS-SLS, etc. In terms of products, there will be two output forms: the agile version (the minimum size of SLS is six, and it is planned to launch a more streamlined version reduced to four) and the enterprise version (six to hundreds of SLS). On the one hand, CDS improves the product competitiveness and product maturity of proprietary cloud enterprise and agile versions; On the other hand, access, backup and analysis of all kinds of data are realized in the end, edge, customer data center and other environments.
SLS (Simple Log Service) Alibaba Cloud log service. SLS originated from the Shennong monitoring service in Alibaba Cloud's early Apsaras Base, and has now developed into an integrated * Ops (DevOps, SecOps, FinOps) solution for cloud native observability that integrates collection, query analysis, and visualization.
Overview of the main functions of SLS
The log data in SLS is AppendOnly, with more writes and less reads. It is time sensitive but not strictly ordered. The query frequency and heat decrease rapidly with time. The CDS-SLS version is inherited from the SLS on Alibaba Cloud. At present, SLS has supported Alibaba's Double 11/Double 12 activities for many years in a row, as well as many major activities such as the Spring Festival Red Packet and the Anniversary Promotion, which have been fully verified in terms of stability, functionality and performance.
This paper focuses on SLS function points from the perspective of operation and maintenance. The main link of SLS includes data acquisition, data query and analysis, visualization and intelligent application. As a product that pays equal attention to computing and storage, in order to further reduce the hardware cost of offline users, some non universal functions will be cut to maximize the computing resources and storage resources of the hardware itself.
The above figure shows SLS functions from the perspective of public cloud users. For CDS-SLS offline users, you can see the corresponding sub modules of SLS services on the space-based platform, as well as the CPU and memory usage of each process. From the perspective of services, it can be divided into two categories: data and scheduling. The former is divided into 34 service roles, and the latter into 10 service roles. It will be easier to upgrade and expand the service after splitting.
SLS internal service splitting
SLs service master is mainly used to schedule related services. Each of its service roles has multiple instances to ensure high availability. The main service functions are concentrated in the sls backend server. The general hierarchical structure is as follows:
At present, the default underlying distributed storage of CDS-SLS is Pangu 2.0 system. As the storage base of Alibaba Cloud, Pangu 2.0 is characterized by high performance and stability. The internal business modules of SLS are also well split into microservices. The bottom layer is self-developed using C++to achieve extreme performance.
There are a large number of background parameters in each module of SLS that can be adjusted, but for the convenience of customers, the default values often meet the needs of most customers. Many designs follow the classic UNIX ideas of "Separation of mechanism and policy" and "Do one thing and do it well".
• For the flow control that users care about, the background provides accurate control of multiple dimensions, and the default parameters can cover most scenarios.
• The data collection agent (Logtail) has been verified on a large scale by millions of machines for many years, and its performance and stability are well guaranteed. Compared with open source software, it can greatly reduce the use of machine resources (up to 90%).
• The pipeline design of "query | analysis" well implements a single responsibility, and query and analysis correspond to different background services.
Special design of hybrid cloud scenarios
At present, there are two SLS related clusters in the space-based:
• The sls common cluster in the base shares the base's Pangu, which provides the most basic query and analysis for the self operation and maintenance of services in the Alibaba Cloud base. The storage time of the base resources is limited to 7 days. In the scenario where the hybrid cloud network isolation is not reachable, the operation and maintenance efficiency is significantly improved. The developers can quickly locate the problem by asking the field personnel to query through several keywords.
• The CDS-SLS cluster purchased separately by the user occupies an exclusive set of Pangu. Only SLS related processes are running in the cluster, which effectively alleviates the shortage of shared resources on the chassis. Therefore, the TTL of logs can be saved permanently, and there is a better console experience.
Most of the functions mentioned in this article are for CDS-SLS clusters purchased separately by users.
Localized credit innovation support
At present, CDS-SLS will support the CPU architectures of Haiguang, Kunpeng, Feiteng, etc., and will have the same strict acceptance tests as Intel X86. Later, more heterogeneous CPU and mixed scenarios will be tested for offline output scenarios.
HTTPS access will support national security TLS channel transmission, making data access in some financial or government enterprise industries more compliant.
Comparison and migration of open source ELK solutions
Background of ELK
Elastic is mainly implemented based on Lucene. In 2012, Elastic became a better software based on Lucene basic library package. In 2015, it launched ELK Stack (Elastic Logstash Kibana) to solve the problems of centralized log collection, storage and query. However, Lucene's design scenario is Information Retrieval, which is a document type. Therefore, there are certain restrictions on the observable analysis (Log/Trace/Metric) data, such as scale, query ability, and some customization functions (such as intelligent clustering LogReduce).
Function comparison of log collection terminal
At present, the log collection terminals of the mainstream open source community are generally Logstash and Fluentd. Early versions will compare some functions and performance. Refer to the test results for data
Knowledge Base Team
Knowledge Base Team
Knowledge Base Team
Knowledge Base Team
Explore More Special Offers
50,000 email package starts as low as USD 1.99, 120 short messages start at only USD 1.00