Optimize notmuch's performance

  1. Optimize notmuch's performance
    1. Xapian 1.2.x database format: chert vs. flint
    2. Tagging: use filters!

Xapian 1.2.x database format: chert vs. flint

Xapian 1.2 (in contrast to 1.0.x) uses a new database format called chert which is a lot faster than the old format flint.

Check whether you are on chert or flint:

$ ls $(notmuch config get database.path)/.notmuch/xapian/iam*
iamchert

If it does not say iamchert, but iamflint, than you should migrate (read on).

There might be a file named flintlock, it is no indicator for your database format, but also used in case of chert.

Migrate from flint to chert

Make sure you are using xapian 1.2.x (depends on your distribution)

XXX: add examples for different distributions

Backup your notmuch tags:

$ cd
$ notmuch dump notmuch-dump-b4chert

Move away your old database:

$ mv $(notmuch config get database.path)/.notmuch/ notmuch-db-b4chert

Find and import your messages, a new and fresh database will be created:

$ XAPIAN_PREFER_CHERT=1 notmuch new

I did not need the XAPIAN_PREFER_CHERT, but am not sure whether there are situations making it necessary (certain compile-time options to xapian?). This note is just here to be extended or removed by somebody more knowledgeable.

After the initial import it is definitely not necessary to define XAPIAN_PREFER_CHERT for subsequent calls of notmuch new.

Restore your tags:

$ notmuch restore notmuch-dump-b4chert

Check whether you are on chert now:

$ ls $(notmuch config get database.path)/.notmuch/xapian/iam*
iamchert

If it says iamflint, most likely you are still using xapian 1.0.x.

Clean-up: if you are sure everything is fine, you can delete the old things. The dump of your tags you might want to keep anyway as a backup - on the other hand you should have some mechanism in place to make regular backups of your notmuch database:

$ rm notmuch-dump-b4chert
$ rm -R notmuch-db-b4chert

Compact your database

In order to greatly reduce the size of your database, you should compact it. Carl says "This functionality is not yet available in the Xapian library interface or else I would probably make notmuch call it after building the database." and Pieter "[compacting] will cause writes to be a bit more expensive in the short term, but will significantly reduce the db's size, which your SSD will surely appreciate."

This needs xapian 1.2

Backup your database:

$ cd
$ notmuch dump notmuch-dump-b4compact

Compact it:

$ cd $(notmuch config get database.path)/.notmuch
$ xapian-compact --no-renumber xapian xapian-compacted

Remove old and renamed compacted, if compacting went well:

$ rm -R xapian
$ mv xapian-compacted xapian

More information about this in the email archives:

Tagging: use filters!

Instead of just:

$ notmuch tag +foo

use:

$ notmuch tag +foo -- not tag:foo

See the ongoing discussion: