[ILUG] [Q] `sa-learn --no-sync': what does it (not?) do, and is `--sync' then needed?

[Justin Mason]

Brian Foster writes:
>  I have just revamped my spam-filtering techniques
>  to include the usage of SpamAssassin (v3.1.8).
>  the Bayesian filter was trained with c.40,000 of
>  the spams I've received, and with c.20,000 hams;
>  both the hams and spams cover the last c.5 years.
>  the training was done in `--local' mode (i.e., no
>  internet access).

Wow, that's quite a lot!  I'd suggest that it'd be fine (and faster) with
just the most recent 1000 or so.

>  to-date, and using essentially the shipped defaults,
>  I've had no false-spams, but only c.50% of the spams
>  are being caught.  that is notably lower than I was
>  hoping for, but we'll see what happens with time
>  and tweaks.  (possibly relevant here, spamd(1) is
>  currently being run `--local' (this could, perhaps,
>  be changed?).)

Yep, definitely change this -- unfortunately spam has evolved to really
require it.  Allowing network lookups will have a much greater
effect on accuracy than training will.

>  one of the key changes I made was, when refiling a
>  false-ham as spam, to run `sa-learn --spam' on the
>  misclassified-spam.  and this is where my current
>  issue is:  it's a bit slow, and on occasion, takes
>  a really really long time (multiple minutes whilst
>  consuming a great deal of CPU).  (this training is
>  also done `--local', so internet access is not the
>  problem here.)
> 
>  re-reading the sa-learn(1) man page, I note there
>  is a `--no-sync' option which _sounds_ like it may
>  deal with one or both slowness issues.  however,
>  the manual page is mostly opaque about about the
>  consequences of using this option, and seems to
>  suggest that after a series of `sa-learn --no-sync's,
>  an `sa-learn --sync' ought to be done.  IF that is
>  true, it's an issue:  it doesn't fit into my nominal
>  routine; and IF it's required, then I cannot ensure
>  it will "always" be done.  (besides, I've no clear
>  idea when or why it's required?)
> 
>   • so just what is `--no-sync' about?
>   • is a `--sync' subsequently required?  (if so, why?)
>   • what are the consequences of not (always) doing a
>      `--sync' afterwards (whether required or not)?

--no-sync will write changes to a journal; you do, then, need
to run --sync later to synchronise the journal to the DB.
If you don't run --sync, the changes will not be reflected
in your scan results.

>  and b.t.w., how safe is it to interrupt (^C) or
>  suspend (^Z) an überlong `sa-learn --spam'?

Both are safe -- just don't "kill -9", or it'll leave lock
files in your ~/.spamassassin dir.

--j.