--mtime slow (70s) for 400k messages

[yeled]

23:50 YELP-CHARLIE~[[email protected]] % keywsync -m /Users/charlie/Maildir/ -q 'not path:spodder/**' --mtime 1472423730 -k --no-replace-chars -d
** keyword <-> tag sync
replace chars: false
=> db: /Users/charlie/Maildir
=> direction: keyword-to-tag
=> query: not path:spodder/**
=> note: dryrun!
=> remove double x-keywords header: 1
mtime: only operating messages with mtime newer than: 2016-Aug-28 22:35:30
* db: current revision: 11562648
*  messages to check: 409777
*  query time: 654.258 ms.
=> done, checked: 409777 messages and changed: 0 messages (skipped: 409777) in 57847 ms [cpu], 74.9375 s [real time].

The large pause is after query time and then it completes.

The count of messages almost matches up:

0:09 YELP-CHARLIE~[[email protected]] % notmuch count 'not path:spodder/**'
409442

I run this regularly, and I have not modified 409,777 messages in the last 5 minutes.

[gauteh]

Charlie Allom writes on august 29, 2016 1:15:

23:50 YELP-CHARLIE~[[email protected]] % keywsync -m /Users/charlie/Maildir/ -q 'not path:spodder/**' --mtime 1472423730 -k --no-replace-chars -d
** keyword <-> tag sync
replace chars: false
=> db: /Users/charlie/Maildir
=> direction: keyword-to-tag
=> query: not path:spodder/**
=> note: dryrun!
=> remove double x-keywords header: 1
mtime: only operating messages with mtime newer than: 2016-Aug-28 22:35:30
* db: current revision: 11562648
*  messages to check: 409777
*  query time: 654.258 ms.
=> done, checked: 409777 messages and changed: 0 messages (skipped: 409777) in 57847 ms [cpu], 74.9375 s [real time].

The large pause is after query time and then it completes.

The count of messages almost matches up:

0:09 YELP-CHARLIE~[[email protected]] % notmuch count 'not path:spodder/**'
409442

I run this regularly, and I have not modified 409,777 messages in the last 5 minutes.

Yes, I'm believe so. The skipped count are the ones that are not checked
because of the mtime check. I have less email than you and spend:

 => done, checked: 67008 messages and changed: 4 messages (skipped: 67001) in 3586.48 ms [cpu], 9.49658 s [real time].

for keywords -> tags

and

 => done, checked: 7 messages and changed: 3 messages (skipped: 0) in 14.887 ms [cpu], 0.0724021 s [real time].

for tags -> keywords.

I am not sure this can be fixed unless we can do some file query where
messages are sorted by mtime. Otherwise, we are not guaranteed that all
message changes are caught.

[gauteh]

ls uses readdir and stores all files and sort them. This might be
faster than notmuch and get_filename. But then we'd have to use a
directory and not a query to find the files that have been modified.
I guess this wouldn't be a big deal for most people.

Perhaps allow query to be replaced with --directory for keyword -> tag sync.

[yeled]

ah, so the changed: X messages is what has matched the >= mtime?

[gauteh]

Charlie Allom writes on august 29, 2016 12:45:

ah, so the changed: X messages is what has matched the >= mtime?

Correct.

[gauteh]

Charlie Allom writes on august 29, 2016 12:45:

ah, so the changed: X messages is what has matched the >= mtime?

No, sorry. The 'checked:' is the ones that matched. 'changed:' is the
checked ones that needed modification.

[gauteh]

One should be able to use `find -mtime` to quickly get a list of the changed files and include this in the query by constructing a long "file1 OR file2" string. This could be implemented directly in abunchoftags, but I don't see that happening quite yet. `time find -mtime +1 > /dev/null` takes about 1.8s (real) on 71k e-mails for me.

[gauteh]

https://github.com/gauteh/gmailieer (brand-new experimental) might do this better