Re: [sqlite] Is there any way to close a connection without checkpointing the WAL?
On 5 Dec 2016, at 9:46pm, Scott Hess wrote: > I don't think so, that seems like it could result in corruption. > Unless you mean something more like causing the OS to block all fsync > calls on the filesystem and release them as a single uber-sync? Yeah, that’s what I meant. It would just cause the device to suspend all writing until you gave it another command which released it all. Obviously it’d have to sync anyway if it ran out of caching space. Seems like the sort of call a device driver might have. And Chromium should be able to talk to its storage at that level. But I don’t recall if I’ve ever seen it. So I’m glad the development team has your back inside SQLite. Simon. ___ sqlite-users mailing list sqlite-users@mailinglists.sqlite.org http://mailinglists.sqlite.org/cgi-bin/mailman/listinfo/sqlite-users
Re: [sqlite] Is there any way to close a connection without checkpointing the WAL?
On Mon, Dec 5, 2016 at 1:34 PM, Simon Slavin wrote: > On 5 Dec 2016, at 9:26pm, Scott Hess wrote: >> An obvious solution would be to simply not call sqlite3_close(), >> though that has various other unfortunate side effects. > > Yeah. Don’t do that, eh ? :-). OK, the biggest unfortunate effect is that you don't release the various resources like memory and file descriptors, which in turn means that you spend your time populating exception policies for automated leak detectors and the like. Also, it means you have to have high confidence that you're skipping the close because of shutdown rather than some other reason (like you're closing the database before deleting it). All of this can probably be ground through eventually, but this kind of thing is likely to cause people to wonder if the code is doing the right thing. > What you’re actually trying to do is disable/delay fsync() for > a time for a particular storage device. Can that not be done > at device-driver level ? It would then affect all the programs > writing "I just quit" to their log files too, which would be an > additional benefit. I don't think so, that seems like it could result in corruption. Unless you mean something more like causing the OS to block all fsync calls on the filesystem and release them as a single uber-sync? -scott ___ sqlite-users mailing list sqlite-users@mailinglists.sqlite.org http://mailinglists.sqlite.org/cgi-bin/mailman/listinfo/sqlite-users
Re: [sqlite] Is there any way to close a connection without checkpointing the WAL?
On Mon, Dec 5, 2016 at 1:38 PM, Richard Hipp wrote: > On 12/5/16, Scott Hess wrote: >> Is there any clean way to request no WAL checkpoint on sqlite3_close()? > > sqlite3_db_config(SQLITE_DBCONFIG_NO_CKPT_ON_CLOSE, db). See > https://www.sqlite.org/draft/c3ref/c_dbconfig_enable_fkey.html at the > bottom. Oh! Now I wonder if I asked about this earlier :-). Thanks, scott ___ sqlite-users mailing list sqlite-users@mailinglists.sqlite.org http://mailinglists.sqlite.org/cgi-bin/mailman/listinfo/sqlite-users
Re: [sqlite] Is there any way to close a connection without checkpointing the WAL?
On 12/5/16, Scott Hess wrote: > > Is there any clean way to request no WAL checkpoint on sqlite3_close()? sqlite3_db_config(SQLITE_DBCONFIG_NO_CKPT_ON_CLOSE, db). See https://www.sqlite.org/draft/c3ref/c_dbconfig_enable_fkey.html at the bottom. -- D. Richard Hipp d...@sqlite.org ___ sqlite-users mailing list sqlite-users@mailinglists.sqlite.org http://mailinglists.sqlite.org/cgi-bin/mailman/listinfo/sqlite-users
Re: [sqlite] Is there any way to close a connection without checkpointing the WAL?
On 5 Dec 2016, at 9:26pm, Scott Hess wrote: > An obvious solution would be to simply not call sqlite3_close(), > though that has various other unfortunate side effects. Yeah. Don’t do that, eh ? What you’re actually trying to do is disable/delay fsync() for a time for a particular storage device. Can that not be done at device-driver level ? It would then affect all the programs writing "I just quit" to their log files too, which would be an additional benefit. Simon. ___ sqlite-users mailing list sqlite-users@mailinglists.sqlite.org http://mailinglists.sqlite.org/cgi-bin/mailman/listinfo/sqlite-users
[sqlite] Is there any way to close a connection without checkpointing the WAL?
At Chromium shutdown, various services desire to write data to their SQLite databases, which results in a (small) thundering herd of fsyncs, which makes shutdown slower than it could be. Normally, one could enable WAL mode to amortize the fsync cost across longer periods than a single transaction, but as best I can tell, sqlite3_close() requires the WAL checkpoint, so won't help. Is there any clean way to request no WAL checkpoint on sqlite3_close()? An obvious solution would be to simply not call sqlite3_close(), though that has various other unfortunate side effects. Thanks, scott ___ sqlite-users mailing list sqlite-users@mailinglists.sqlite.org http://mailinglists.sqlite.org/cgi-bin/mailman/listinfo/sqlite-users
