Bundle indexing and preview computation in a single listener

[chibenwa]

Why?

Mass delivering enail saturates network link.

The pitfall is that for each mailbox we need to re-download the all message in order to...

• Index it in OpenSearch
• Pre-compute the preview

So 13 MB x 2 x 10.000 emails * 8 bit ~= 2 Tb of traffic
With a link speed of ~1Gb/s that's 40 minutes of link saturration.

Which is exactly what we see: incoming link saturated, outgoing link easy peasy.

How?

We can bundle Index it in OpenSearch + Pre-compute the preview as a single listener in Twake mail side and thus download the message for post processing only once and not twice.

Easy to do, gain a factor 2 (and reduces OVH cloud bill)

[quantranhong1999]

WIP: #1508

[chibenwa]

Proposal:

inferface MessageIndexer {
    Mono<Void> index(MailboxMessage message, Added event);
}

Write an OpenSearchMessageIndexer.

Write a ComputePreviewMessageIndexer.

(Extract some logic in ComputePreviewListener so that it gets reused into ComputePreviewMessageIndexer )

Inject Set<MessageIndexer> into the OpenSearchMailboxSearchIndex.

[quantranhong1999]

apache/james-project#2631

I will create a dedicated JIRA ticket if this goes in the right direction.

[chibenwa]

Not really this is overly complex and I see many needless refactoring arising from the interface being wired in the wrong place.

Let me propose an alternative...