RT 4.4 slow sbin/rt-externalize-attachments with large and old (before RT 4.2) databases

Hi,

I'm currently upgrading one of our old RT instances from 3.8.1 to 4.4.0.
The instance started with RT 3.4.2 and has over 7,000,000 attachments.

sbin/rt-externalize-attachments is really slow, 1,500,000 attachments in
4 hours.

The MySQL processlist shows me always 'UPDATE Attachments SET Headers='
queries.
This is because rt-externalize-attachments calls
RT::Attachment->ContentLength [1] which checks if a Content-Length
header exist, and if not, update the header with a Content-Length header
value.

Unfortunately RT never set a Content-Length header for created
attachments before 4.2.0 [2]. While incoming mails and attachments
expected to have a Content-Length header, attachments created by RT
themselves (ticket create by WebUI/REST, OutgoingEmailRecord) doesn't
have one.

My question:
is it really necessary that rt-externalize-attachments update all
attachment which doesn't have a Content-Length header value?

I already moved attachments from one of our RT 4.2. instances out of the
database with RT::Extension::ExternalStorage, which fetches the content
length from the DB [3], which is much faster (1,500,000 attachments in 2
hours).

Chris

[1]
https://github.com/bestpractical/rt/blob/stable/lib/RT/Attachment.pm#L426-L439
[2]
https://github.com/bestpractical/rt/commit/337893c1f5b53a430667107b28b8212acdaf037e
[3]
https://github.com/bestpractical/rt-extension-externalstorage/blob/master/sbin/extract-attachments.in#L99-L105
---------
RT 4.4 and RTIR Training Sessions https://bestpractical.com/training
* Washington DC - May 23 & 24, 2016