"rebuilding indices from scratch"

[Carl Hörberg]

When a node in a cluster that wasn't gracefully shutdown (or sometimes even when gracefully shutdown too i feel like) starts up again the indicies are being rebuilt. This takes a very long time, often 10-30min, when there's several hundreds of GBs of persistent messages. Are the any speedups that can be made here? Does every byte have to be read from disk when rebuilding the indexes?

We run only mirrored queues so all messages are then thrown away when the node is up and are being resynced from the other nodes. I guess we in that case can just drop the msg_store_transient dir before starting?

[Michael Klishin]

It's a sequential scan over all messages.
I can't think of speed-ups other than "use an SSD."

MK

[Alvaro Videla]

Also AFAIK the transient message store is not scanned during startup/recovery

--
You received this message because you are subscribed to the Google Groups "rabbitmq-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to rabbitmq-user...@googlegroups.com.
To post to this group, send an email to rabbitm...@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

[Laing, Michael]

Yet another argument for transient queues and non-persistent messages. :) ml

To post to this group, send email to rabbitm...@googlegroups.com.

[Carl Hörberg]

Typo, I meant msg_store_persistent