Failed messages get replayed some number of times (depends on the topology) and then the dead letter queue functionality of our queuing system takes over. We use SQS, but this is bog standard functionality in almost any queue system. The dead letter queue has alarms on it and we have some scripts that can be manually triggered to reprocess the data after investigating/correcting the issue that caused the failure.
On Mon, Oct 12, 2015 at 1:55 PM, Ankur Garg <[email protected]> wrote: > Hi , > > So I wanted to understand how ppl process messages which failed inside > Storm . > > Do you store it in memory in some data structure (say queue) or use any > other technique to store it . > > Please provide your inputs as to how you deal in production environment. > > Thanks > Ankur >
