Re: Commit after some number of transactions

2021-06-27 Thread David Bremner
David Bremner  writes:

> The main rational is explained in the commit message to 
>
> [PATCH 4/5] lib: autocommit after some number of completed
>
> I'm not super-happy with the documentation in [5/5], as it explains
> things in terms of database concepts the user shouldn't really need to
> understand.
>

I have applied the series to master. Documentation improvements welcome.

d
___
notmuch mailing list -- notmuch@notmuchmail.org
To unsubscribe send an email to notmuch-le...@notmuchmail.org


Re: Commit after some number of transactions

2021-05-24 Thread David Bremner
David Bremner  writes:

> The main rational is explained in the commit message to 
>
> [PATCH 4/5] lib: autocommit after some number of completed
>
> I'm not super-happy with the documentation in [5/5], as it explains
> things in terms of database concepts the user shouldn't really need to
> understand.
>
> [PATCH 5/5] doc: document database.autocommit variable
>
> The default value of 8000 was chose not to cause any noticable
> slowdown when indexing the "large" corpus of about 200k messages.  The
> test machine is a recent Xeon with fast spinning rust drives; the
> whole index takes about 8.5 minutes on this machine. I'd be curious if
> other people notice a performance impact.  On the same machine the
> threshold of 8000 means that less than 30 seconds worth of work would
> be discarded.

An alternate approach would be to expose Xapian's commit() method as
notmuch_database_commit(). That would not really conflict with this
series, but arguably makes it unnecessary, if the clients (e.g. the
notmuch CLI) add appropriate calls to notmuch_database_commit.

I'm not 100% sure, but I think I may want to add notmuch_database_commit
in any case, for some other cases where automatic committing doesn't
work so well.
___
notmuch mailing list -- notmuch@notmuchmail.org
To unsubscribe send an email to notmuch-le...@notmuchmail.org