search

maestre3d / alexandria

The Alexandria Project is an open-source platform where people can share their knowledge through books, podcasts, docs and videos.
https://alexandria.damascus-engineering.com
GNU Affero General Public License v3.0
9 stars 1 forks source link

No information while transaction is being executed at client-side #34

Open maestre3d opened 4 years ago

maestre3d commented 4 years ago

Describe the bug Whenever we start a distributed transaction (SAGAs mostly), the client is not able to see what is the current state in real-time and even worst, if something fails, there is no way to know what happened.

To Reproduce Steps to reproduce the behavior:

  1. Make a POST HTTP request or a gRPC action call to any endpoint with dependencies (almost all).
  2. Wait for response, you shall see the created entity with state set to PENDING.
  3. You need to execute a short-polling retry policy if you want "real-time" state. Aside of that technique, there's no actual way to know the entity's state explicitly (if failed, then be able to see what happened)

Expected behavior We must have a way to see entity's state in real-time while a distributed transaction is being executed along with following error information if a failing scenario happened.

Additional context No.

maestre3d commented 4 years ago

We could take advantage of Redis and Apache Kafka capabilities to design and build a nice distributed environment for declarative transaction execution,

maestre3d commented 4 years ago

Now considering using synchronous inter-service communication when a transaction is needed using resiliency patterns such as rate-limit for received requests and circuit breaker (w/ Netflix Hystrix) and a retry strategy for external calls (outbound) to avoid cascading failures and many more problems.

We would be still using asynchronous communication for side-effect distributed propagation.

This is due the complex implementation (e.g. dist. tracing and metrics for each event bus) and maintenance of each distributed transaction using pure events.

maestre3d commented 4 years ago

In addition, we could take advantage of event sourcing and CQRS when required to avoid these issues. More details here.

maestre3d commented 4 years ago

A well-tested approach is a Self-Contained service, more information about it here.

Thus, we will have two options and this kind of approach could be used whenever an operation must be tracked by the client.

This could be a definitive standard, almost closing this issue.