supermacro / a1pubsub Goto Github PK

Seeing logs like this in our logs:

[pubSubMessageHandler] - 0322fba9-4609-43b4-8e8e-38e9f4fe0352 - Failed to process pubsub request - {"error":"invalid_event_data","reason":"Data for subscription \"<REDACTED>\" did not pass validation. Data: [object Object]"}

The issue is here:

We're not calling JSON.stringify

Imagine the following architecture:

In this (very common) architecture, there is a scenario that is very common with PubSub Push Messages.

• Some system publishes a message
• GCP receives message and then pushes message to a subscriber
  • Let's suppose message has id 123
  • subscriber has the above architecture
• Load balancer receives message 123 and forwards to node A
• a1pubsub is currently processing message 123
• Google sends a duplicate message
  • recall that google does not guarantee exactly once delivery of messages ... if it did, then you wouldn't need a1pubsub :)
• Load balancer receives message 123 a 2nd time, but forwards it to node B
• a1pubsub is currently processing message 123

Oh nooo, node B was not aware that node A was already processing message 123.

Thankfully the fix is quite straight forward. And provides a reasonably high level of confidence. Assumption is that query_latency < latency_between_original_and_duplicate_pubsub_message. If this assumption is not true, then this fix doesn't work.

You can acquire a lock on messages by leveraging the EventStatus enum.

https://github.com/supermacro/a1pubsub/blob/master/src/index.ts#L42

And then inside of handlePubSubMessage, check if there's already a cached message that is in_progress. If true, then throw away the current message since it's a duplicate.

Would be great to have a CLI to make working with pubsub even easier:

• Create Topic
• Create Subscription
  • enforce the <eventName>__<actionToTake> naming pattern
  • generate ngrok link automatically for local development
  • set up authentication automatically