On 04/05/2018 02:08 PM, Pavel Cernohorsky wrote:
Hello Dan, so you are saying that if WAL is somehow in memory only
(not properly persisted) and app or OS crashes in the middle of the
checkpoint operation, my main database file will get corrupted? And by
corrupted you mean as in "you will loose changes which were in the WAL
file", or "you will end up with unusable main database file, or file
where rows which were affected by the checkpoint will have wrong
contents (halfway through written, ...)". In other words, I may end up
with the main database file in some other state than just "like before
checkpointing" or "like after checkpointing"? I understood
checkpointing as a kind of atomic operation which "merges data in the
main database file and in the WAL". Is that understanding wrong?
That's correct. If you crash mid-checkpoint and lose the wal file, some
future queries may return inconsistent results or SQLITE_CORRUPT errors.
The suggestion made in another post to put the wal file on a tmpfs or
similar file-system is a good one. Then you will only risk corruption if
the OS crashes. There is no easy way to do that at the moment though,
you will have to do some hacking to get it to work.
Dan.
Thanks, Pavel
On 04/04/2018 06:33 PM, Dan Kennedy wrote:
On 04/04/2018 09:01 PM, Pavel Cernohorsky wrote:
Hello, does anybody know if there is some possibility to not have
WAL file as a normal file on the disk, but only in memory? I
understand that all the modifications to the database would get lost
in case of the application / OS crash, but for my application, I
only need the level of durability based on checkpointing. I just
need to guarantee that all the data are properly written to the main
database and synchronized to disk when manual (or even automatic)
WAL checkpoint is called, but I do not care if I loose data in
between the checkpoints. Of course database should never get corrupted.
If your app or the OS crashes halfway through a checkpoint and this
means that the WAL file is lost, the database is likely to be
corrupted. Is that a problem?
Dan.
My goal is to limit the number of IOps being performed to the disk.
Currently I use "PRAGMA synchronous = 1" and there is only one
process manipulating the database (multiple reader threads, only one
writer thread at one moment in time). Or if it is not possible to
have WAL in memory only, is there something like “PRAGMA
wal_synchronous = none_and_delete_wal_if_corrupted”?
Thanks for suggestions, kind regards,
Pavel
_______________________________________________
sqlite-users mailing list
sqlite-users@mailinglists.sqlite.org
http://mailinglists.sqlite.org/cgi-bin/mailman/listinfo/sqlite-users
_______________________________________________
sqlite-users mailing list
sqlite-users@mailinglists.sqlite.org
http://mailinglists.sqlite.org/cgi-bin/mailman/listinfo/sqlite-users
_______________________________________________
sqlite-users mailing list
sqlite-users@mailinglists.sqlite.org
http://mailinglists.sqlite.org/cgi-bin/mailman/listinfo/sqlite-users
_______________________________________________
sqlite-users mailing list
sqlite-users@mailinglists.sqlite.org
http://mailinglists.sqlite.org/cgi-bin/mailman/listinfo/sqlite-users