Patrones de observabilidad para aplicaciones modernas

img

Metadev Whitepaper: Observability Patterns for Modern Applications.
© Metadev, Seville, Spain, EU. MMXXII.
All rights reserved.
Cover design & image: Paco Soria, 2022.
Produced with LATEX / MiKTEX and pdfTEX.

Descargar

About

Metadev S.L. is a company devoted to creating high quality software on the cloud. Metadev helps its customers to improve processes, train technical staff, and recommend practices for excellence in development and operation optimizing ROI on Software investments.
Metadev | info@metadev.pro

author

Pedro J. Molina, PhD
Metadev’s Founder

Executive Summary

This document presents key technical patterns to expose application’s health, metrics, logs and KPIs in a standardized way to allow aggregate reporting and quick decision making by stakeholders.

Contents

  1. 1.Goal

  2. 2.Patterns & practices

    • 2.1 Visible Version
    • 2.2 Ping/Heartbeat Endpoint
    • 2.3 Metrics Endpoint
    • 2.4 Status Page
    • 2.5 Health Endpoint
    • 2.6 Structured Log
    • 2.7 Centralized Log
    • 2.8 Application Instrumentation
    • 2.9 Configuration Management
    • 2.10 Central Configuration Management
    • 2.11 Audited Configuration Management

1. Goal

This whitepaper provides a set of condensed best practices distilled at Metadev when dealing with Application Monitoring and operation in production with the goal of saving time and effort and reduce overall downtime. We help customers to implement these practices to deploy and operate application on scale with safety and security. Expected audience: CTOs & Software Architects.

2. Patterns & practices

  • Visible Version
  • Ping / Heartbeat Endpoint
  • Metrics Endpoint
  • Status Page
  • Health Endpoint
  • Structured Log
  • Centralized Log
  • Application Instrumentation
  • Configuration Management
  • Central Configuration Management
  • Audited Configuration Management

2.1 Visible Version

Application’s users should be able to know the exact version of the product they are using. Showing in a prominent place of the UI (f.e. footer, welcome or about page) the version number. This information will be useful for:

  • Verify a deployment matches the expected version.
  • Discover/ensure versions on environments.
  • Report bugs with the corresponding exact version where they were found.

2.2 Ping/Heartbeat Endpoint

Applicable in Web applications, services and long running processes. It provides a specific endpoint f.e. GET /ping in a way a client can verify if a service instance is available or not responding. For a call as follows:

curl https ://server/ping

The service will respond with:

200 OK
Content-Type: application/json

{ " message ": " pong " }

In case of failure, the client will receive a timeout or a network error. Ping service does not need securitization (no credentials needed). However, can be constrained to respond to specific internal IPs if confronted with Denial-of-Service attacks. In services needing been monitored to verify availability and correct operation (almost all of them). Scalable services behind a load balancer, in containers or Kubernetes. This ping or heartbeat service allows a load balancer like nginx [9] or ha-proxy [2] to route workload to different nodes dynamically depending on latency observed in each node. Tools like Consul [6] or Nagios [14] allows to periodically monitor service instances.

2.3 Metrics Endpoint

A dedicated metric endpoint allows to know about the performance indicators of the service. A third-party system can query services to collect and centralize technical and business metrics. This information can be stored, aggregated and visualized in real time, also used as a alert triggers. Tooling like Prometheus [3], InfluxDB [12], or Graphana [13] allows to aggregate monitoring information, define business dashboards and trigger alerts as needed. In a call similar to this one (note the service provides API-Key authentication):

curl https://server:443/metrics -H "apikey = ab39z3d893"

The service will respond (example using Prometheus Metrics format):

200 OK
Content-Type: text/plain ; charset=utf-8

customers_enrolled 238
completed_courses 345
api_calls_counter{code="200"} 23456
api_calls_counter{code="404"} 132
load_average_1min_gauge 6.037109375
load_average_5min_gauge 8.2421875
load_average_15min_gauge 8.8232421875
memory_used_bytes_gauge 62867410944
memory_free_bytes_gauge 1557377024

In the sample you can see:

  • Operative metrics as memory, CPU or I/O usage.
  • Business metrics including customers enrollment and completed courses (to be defined as KPIs in each service).

Business services needing monitorization to verify availability and correct function. As default, all services would need to be monitored in production.

“What cannot be measured, cannot be optimized.” — Lord Kelvin.

2.4 Status Page

A Status page allows operators and developers to verify a service has everything needed to operate in a healthy state.

In particular, the focus is on checking the external dependencies of the service. Consider as external dependency every resource out of control of the application like:

  • Environment configuration
  • Databases availability and permissions
  • File system availability, permissions, and free space
  • Queues
  • Send Mail service availability
  • Third-party services

The Status page will perform a series of health checks to verify:

  • Environment configuration

    • Correct settings and complete
    • All compulsory parameters are passed
    • All parameter passed are in the correct format (are valid)
  • Databases

    • Connection string is valid
    • Can connect to database
    • Schema version match the expected one
    • Has permissions on DB to read/write as needed
  • File System

    • File System is accessible. Units are mounted (no timeouts).
    • Read/write permissions are correct
    • File system schema version is expected: folder and files structure conform the expected one
  • Queues

    • Queue exists
    • Permissions are correct
  • Send mail & third-party services

    • Is possible to connect to the external service (ping or probe)
    • Latency and connection times are under an acceptable threshold.
    • Service Works (a query to service3/status is possible if implemented).

Samples:https://status.bitbucket.org, health-check of a home router.

Business Services needing to be monitored for proper behavior (all services in production).

Critical in complex services with many dependencies. Many dependencies and distributed systems, increases the chance of failure.

The Status page reduces a lot the time for diagnostic a failure caused by an external dependency. Saves developer’s time chasing for ghosts when the root cause for the failure is external. Direct ROI in terms of time and money saved in operation.

2.5 Health Endpoint

Quite similar to Status Page pattern, but oriented to machines rather than humans. It will return JSON or XML and will be systematically monitored and collected on a fixed frequency.

Sample call:

curl https://server:443/health

Sample response:

200 OK
Content-Type : application/json

{
  status : " Degraded ",
  dependencies : [
    { name : "BD", status : "OK" },
    { name : " Queues ", status : "OK" },
    { name : " Email ", status : " FAIL ", message : "No response "},
    { name : " GoogleMaps ", status : " DEGRADED ", message : " Latency greater
    than 500 ms" }
  ]
}

When the number of services to track and monitor grows, it’s a good idea to centralize and automate health-check collection in a standardized way. Dashboards and alert systems can be setup feed by the data collected. It allows to respond quick and proactively to infrastructure failures or third-party services (in some cases before users are affected).

Allows to verify and track SLAs and effective availability of third-party services.

2.6 Structured Log

Standard log format shared across all applications in the company. It allows to implement common trace and correlation ids in the aggregated log. Select and apply a format for structured log. Samples:

  • Structured Login
  • Serilog
  • Stackify

When orchestrating several services collaborating to resolve a single user request or when the number of services grows.

2.7 Centralized Log

Centralizes traces and service logs to be aggregated in a central service. Allows to explore traces and filter logs using standard log exploitation tooling. Traces (with correlation-id) and structured logs allows to track individual requests crossing several services. Aggregate logs [1] using services like Elastic-Search [4] or Splunk [11].

Having a central log when the number of machines and nodes is greater than 5-8 is a key tool to escalate production troubleshooting issues.

2.8 Application Instrumentation

Application Performance Monitoring tools (APM) instrument service code to measure via sampling times and bottlenecks on CPU, memory usage, disk, IO, and time waiting for externals calls. Using product like New Relic APM [10], Application Insight [5] or DataDog [8].

Recommendable in production systems. Allows to discover real bottlenecks (not in theory ones) and to plan accordingly to this information the next optimization to tackle.

2.9 Configuration Management

Configuration Management is a disciple to control the configuration been passed to services in environments. When the number of services (S), versions deployed (V ) and environments (E) grows, the complexity (C) of handling the correct configuration increases as (C = S × V × E).

  • Keep configuration in code, versioned in source control. Separated by environment and version.
  • Track changes to configuration. Use a repository for all Configuration Management (different and independent from the service’s code).
  • Verify configuration is correct using automation and/or service health-checks.

When the number of services is greater than 5-7.

2.10 Central Configuration Management

As the number of services (or microservices) keeps growing, many of them have similar configuration. Therefore, makes sense to centralize a source of configuration. Products like Consul [6] provides a high-availability key/value store where service configuration can be persisted and served.

The inversion of control applied here configures in each service a well-known DNS name and a port to reach the central configuration service. This service will provide the required configuration based on the service requesting, version and current environment.

Use a product for Central Configuration Management like Consul [6]. Setup it in high-availability mode as configuration is a critical data for services to start-up.

When the number of services is greater than 12.

2.11 Audited Configuration Management

In scenarios where security is a key concern like Banks, Public Administration, aerospace, or military systems changes to Configuration Management must be tracked, reviewed, authorized and audited.

Changes to Configuration Managements can be done only by authorized principals and follow a review process workflow before application.

Implement such Audit and Workflow capabilities if security requirements are meet.

  • Secure the Configuration Management tool. Integrate it with a workflow process.
  • A git repository with signatures could be a great option. Can use PR as review process.
  • Consider Vault [7] or similar product for credential storage and secrets.

Consider it when strong security requirements apply.

References


If you liked this whitepaper:

icon1 Enroll into our newsletter for future whitepapers & techonology insight.

icon2 Feel free to forward this whitepaper to anyone interested in the topics covered.

icon3 Join the conversation & write us at info@metadev.pro.

Si todavía tienes preguntas

Envíanos un correo y conversamos.

¿Hablamos?