Wondering about DMS and rspamd SPAM/HAM learning proces

[hapklaar]

I just enabled rspamd in my DMS setup and have some questions about spam handling:

1. The docs state that when RSPAMD_LEARN is set to 1 autolearning is enabled. On the same page it also states "Do you additionally want to enable the auto-spam-learning for the Bayes module? No problem: set-option-for-module classifier-bayes autolearn true." For now I have only set the DMS config parameter for this to 1.

Do I need to set either or both for autolearning, or do they both configure the same rspamd setting?

2. How can I also use a #SPAM and #HAM folder in my mailbox where I can move email that should be learned as such?

3. Since enabling aulolearn I have received many emails with a negative spam score according to the rspamd history, but on the rspamd statistics page it says only one has been learned as HAM:
   
   If I understand the rspamd docs correctly, it should learn all negative scored mails as HAM, but this obviously doesn't happen. Hope someone can explain why this is.

4. If I want to tweak the autolearn function, should I disable RSPAMD_LEARN and enable it again in custom-commands.conf with for example set-option-for-module classifier-bayes autolearn [-5, 5] ?

5. Does anyone know of a good guide to alternatively implement autolearning of SPAM when messages are moved into the JUNK folder and learned as HAM when they are moved out from the JUNK folder in this particular DMS+Dovecot+Rspamd setup? I found some guides for this, but they don't take into account the dockerized nature and want me to change config in the container which won't persist. I assume any config needs to be done using custom-commands.conf?

thanks in advance!

[hapklaar]

Let me add what I've learned so far:

1. ?
2. Can be done by configuring docker-data/dms/config/dovecot.cf, for example:

plugin {
  imapsieve_mailbox1_before = file:/usr/lib/dovecot/sieve-pipe/learn-spam.sieve
  imapsieve_mailbox1_causes = COPY APPEND
  imapsieve_mailbox1_name = #SPAM
  imapsieve_mailbox1_from = *
  imapsieve_mailbox2_before = file:/usr/lib/dovecot/sieve-pipe/learn-ham.sieve
  imapsieve_mailbox2_causes = COPY APPEND
  imapsieve_mailbox2_from = *
  imapsieve_mailbox2_name = #HAM
}

4. ?
5. Found this should already be working by default when rspamd is enabled. It didn't work for me because Outlook for Windows uses method 'APPEND' instead of 'COPY' to copy and move items between folders, which didn't trigger imapsieve as it only reacted to cause 'COPY'. Unfortunately the 'APPEND' method does not specify a FROM location, so even with the APPEND method added as a cause (see above example) , the ham learning script was never triggered (requires from Junk to work by default). This is why I chose 2 separate ham and spam folders, so APPENDing anything to one of those folders could trigger ham and spam learning respectively without knowing a FROM folder.