Re: dar became very slow (and process in D state?)
Andrea Vai <[email protected]>
| Newsgroups | gmane.comp.sysutils.backup.dar.support |
|---|---|
| Message-ID | <[email protected]> |
Il giorno ven, 17/01/2020 alle 21.32 +0100, Denis Corbin ha scritto: > > On 17/01/2020 12:47, Andrea Vai wrote: > [...] > > > > I have just tried and didn't make any difference, but I am > > wondering whether I am not understanding something. My problem > > appears while writing a new archive, so does --sequential-read > > option matter as well in that case? > > Yes, you are correct. The sequential-read does nothing (except > concerning the archive of reference if used) when creating an > archive. > > The archive is produced by appending data to a file, there still > might > be some seek backward if compression is worse than storing data > uncompressed (operation cp does not do). This one is not possible to > disable unless you disable compression at all. You can also write > the > archive to stdout to avoid having this occuring though dar will try > and get an error so the system call will still differ from what cp > would do. > > dar -c - .... > my.archive.1.dar > > Another condition that have influence on the way dar writes to an > archive is when a file has changed, dar will retry saving it up to 3 > times by default. See --retry-on-change option to disable this. > > > > > > the command I run is > > > > # SECONDS=0; mount UUID=ae7ac0ca-d0bc-46c8-81f1-d2b67e1f95f3 > > /mnt/pendrive && rm -f /mnt/pendrive/aero.1.dar && dar -c > > /mnt/pendrive/aero -R /home/andrea/Musica/MP3/Aerosmith && umount > > /mnt/pendrive; echo "Ci ho messo: $SECONDS secondi. > > > > which, as far as I know, doesn't involve reading a catalog, does > > it? > > You seems doing a full backup (I cannot see whether there is a > .darrc > or /etc/darrc but assuming there is none) without compression on > data > that do not seem subject to change. Yes, your assumptions are correct. And thank you for all the the explanations, > So I thus cannot see what dar does > so differently from the cp command in that context for the first > trigger the kernel bug while the second does not... Many thanks, Ciao :-) Andrea