pulsar icon indicating copy to clipboard operation
pulsar copied to clipboard

[Bug] version 2.8.0.7 Restarting broker consumes historical data

Open sqtce opened this issue 2 years ago • 7 comments

Search before asking

  • [X] I searched in the issues and found nothing similar.

Version

My brokers are using version 2.8.0.7 and there are three of them, used for handling delayed messages. Last night, due to a hardware issue, I restarted one of the brokers. As a result, all the messages that were supposed to be consumed by that broker on that day were re-delivered.

Minimal reproduce step

restarted

What did you expect to see?

How to resolve this issue?

What did you see instead?

re-delivered

Anything else?

No response

Are you willing to submit a PR?

  • [X] I'm willing to submit a PR!

sqtce avatar Jun 25 '23 03:06 sqtce

@sqtce Can you provide an appropriate title for easy understanding?

JooHyukKim avatar Jun 25 '23 06:06 JooHyukKim

Restarting broker consumes historical data

sqtce avatar Jun 26 '23 05:06 sqtce

There is no 2.8.0.7 in upstream. Do you use the version of any vendors'?

Also, 2.8.x is end of life in upstream, the least maintained version is 2.10.4 now.

tisonkun avatar Jun 30 '23 09:06 tisonkun

The issue had no activity for 30 days, mark with Stale label.

github-actions[bot] avatar Jul 31 '23 01:07 github-actions[bot]

Hello,

We are encountering the same behavior on our Pulsar cluster in 3.1. It's a topic with long retention time (90 days) with delayed message that are acknowledged once consumed. When restarting broker the pulsar backlog metric is getting back to the full backlog and we reprocess ~2M messages. Here is a grafana screenshot after the broker restart showing pulsar_subscription_back_log metrics on the topic image

If you have a workaround or a solution that will be great!

Ouriel avatar Apr 04 '24 15:04 Ouriel

Hello,

We are encountering the same behavior on our Pulsar cluster in 3.1. It's a topic with long retention time (90 days) with delayed message that are acknowledged once consumed. When restarting broker the pulsar backlog metric is getting back to the full backlog and we reprocess ~2M messages. Here is a grafana screenshot after the broker restart showing pulsar_subscription_back_log metrics on the topic image

If you have a workaround or a solution that will be great!

@Ouriel What's your code looks like? I mean how do you create the consumers?

dao-jun avatar Apr 04 '24 17:04 dao-jun

We are using pulsar through Infinitic.io framework Here is the consumer code : https://github.com/infiniticio/infinitic/blob/v0.13.0/infinitic-transport-pulsar/src/main/kotlin/io/infinitic/pulsar/consumers/Consumer.kt The setup of topic is also done by the framework here: https://github.com/infiniticio/infinitic/blob/v0.13.0/infinitic-transport-pulsar/src/main/kotlin/io/infinitic/pulsar/admin/PulsarInfiniticAdmin.kt

Ouriel avatar Apr 05 '24 07:04 Ouriel