evaneos / burrow Goto Github PK

Environnement:
Message Broker: RabbitMQ
Driver: php-amqplib
Hosting: GKE
Orchestration: Kubernetes
Containerization: Docker

We have 2 workers, a publisher and a consumer.

After about 8 hours of inactivity by the publisher and when it start again to push messages into RabbitMQT, the container got an error: syslog: write tcp x.x.x.x:41132->x.x.x.x:5000: write: broken pipe.
No problem, the app crash and the container is restarted by Kubernetes.

But for 1 day by week the consumer did not have this error (that the publisher always have).
In this case, the consumer stay alive without consuming the new messages because the connection is lost but he is not aware of it.

We have updated our container by using the amqp c library instead and let know here if we still have this issue or not.

Daemon can not handle signal until it consume a message from AMQP, as we call pcntl_signal_dispatch after the comsuption. This issue happen only with PECL AMQP as consuming a queue is blocking whilst AMQPlib is non blocking.

A possible solution is to use this signal handler as suggested here

Hello!

Please look at https://github.com/queue-interop/queue-interop project. Using interfaces from it allows us reuse some implementations like enqueue, which supports a lot of transports. You can outsource some code